Title of Invention	A METHOD AND A DATA PROCESSING SYSTEM FOR MONITORING THE HEALTH OF A PLANT.
Abstract	This invention relates to a method for monitoring the health of a system comprises performing at each of a plurality of times the steps of: constructing a condition signature from a plurality of condition indicators including (a) a plurality of vibration measurements acquired from the system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from the system; predicting a normal signature from a model defining one or more inter-dependencies between said condition indicators, the normal signature corresponding to the condition signature for a healthy system; comparing the condition signature with the normal signature; and registering an event if the condition signature differs from the normal signature by more than a predetermined threshold.

Title of Invention

A METHOD AND A DATA PROCESSING SYSTEM FOR MONITORING THE HEALTH OF A PLANT.

Abstract

This invention relates to a method for monitoring the health of a system comprises performing at each of a plurality of times the steps of: constructing a condition signature from a plurality of condition indicators including (a) a plurality of vibration measurements acquired from the system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from the system; predicting a normal signature from a model defining one or more inter-dependencies between said condition indicators, the normal signature corresponding to the condition signature for a healthy system; comparing the condition signature with the normal signature; and registering an event if the condition signature differs from the normal signature by more than a predetermined threshold.

Full Text	FIELD OF INVENTION This invention relates to a methods and data processing systems for monitoring the health of a system. The methods and data processing systems of the invention are particularly, although not necessarily exclusively, suitable for monitoring the health of power plant, including for example gas turbine, spark ignition and compression ignition internal combustion engines. BACKGROUND OF THE INVENTION The health of the system can be considered a measure of the condition of a system against expected norms. A health system is one whose condition closely matches expectations, whereas an unhealthy system is one whose condition differs from what would be expected, indicating for example deterioration of, or a possible problem with the system. The ability to monitor the health of a system can therefore allow such deterioration and/or problems to be detected and, if necessary, addressed at an early stage. For example, US-5684718 describes a non-real time system for monitoring the operation of an electric generator in which vibration and load data are combined to produce a single which is then compared with stored data representative of maximum acceptance combinations of the two parameters. Essentially the system is an automated Nook up table' which issues warnings when vibrations have exceed acceptable limits. In order to determine the condition, and consequently health, of a system, it is normal to monitor and analyse a series of measurable indicators which themselves reflect performance parameters such as turbine and compressor operating temperature and pressures and spool speeds. To obtain a fuller overall picture of the engine's condition, these performance parameters can be supplemented with further condition indicators including, for example, vibration measurements and measurements of the particulate material entrained in the circulating oil. Particularly with complex mechanical systems such as gas turbines, the number of indicators that must be monitoring to obtain a useful overall picture of the system's condition can be high. This in turn means that the task of analysing the complete series of indicators to determine the health of the engine is a complex one, typically requiring a skilled expert to analyse the data off-line. Taking again the example of a gas turbine, it is known for example to collect performance and vibration data from the engine over time to be analysed off-line by one or more experts. Typically the performance data will be compared with simulated data for the same engine and, based on this comparison, an expert will from a view as to the health of the engine. Additionally, a small amount of vibration data will be reviewed. Giving a superficial view of gross changes in engine behaviour. If a problem is detected, the vibration data may then be analysed in more detail, often by another expert, to look for any abnormal indications which might be symptomatic of underlying mechanical problems which could lead to a loss of health and operability. SUMMARY OF THE INVENTION It is a general aim of the present invention to provide methods and data processing systems that facilitate the acquisition and analysis of condition indicators in a manner such that the overall health of a system can be more readily assessed. Accordingly, in general terms a first aspect of the invention provides a method for monitoring the health of a system, comprising: constructing a condition signature from a plurality of measured condition indicators acquired from the system; comparing the condition signature with a normal signature, corresponding to the signature for a healthy system; and registering an event if the condition signature differs from the normal signature by more than a predetermined threshold. The term 'signature', as used herein, pertain to the values of a plurality of condition indicators merged or fused into a unit or quality such as a set, vector or scalar. In the example of a vector signature, the indicators may correspond to respective elements of the vector. In the example of a scalar signature, the magnitude of the scalar may be determined by a mathematical function which acts upon the indicator values. By merging or fusing the condition indicators into a single signature in this manner, and providing a normal signature with which the fused data can be compared, the task of assessing the health of a system is greatly simplified. In particular, since the detection of an event amounts to an indication of a potential problem or an unhealthy system (i.e. a system condition that differs from what would normally be expected), the monitoring of the health can be largely automated, removing, or at least minimizing, the requirement for expert input during the monitoring process. This in turn means that it becomes feasible to continuously monitor the health of a system, and to provide useful information about the health of the system in real time during operation. Preferably the condition indicators that are combined to form the system condition signature include operational parameters, which in the case of a mechanical system may be speeds, pressure (e.g. gas pressures, oil pressures) and temperature for example. Other useful parameters may include what might be conventionally thought of as control or status parameters. For convenience, such parameters will be referred to using the single label of 'performance parameters' in the following text. Additionally, to obtain a fuller picture of a mechanical system's heath, it is particularly preferred that the signature includes one or more condition indicators related to the vibration of the system. Put, more generally, the condition indicators from which the system condition signature is constructed may be derived from two or more disparate sources of data. This illustrates a particular strength of this approach in that a great variety of different forms of condition indicator data can be encompassed in the system condition signature, providing a more comprehensive measure of the system's health than has previously been possible without multiple analyses. Preferably at least three condition indicators are used to contrast the condition signature. More preferably at least 10 and even more preferably at least 20 condition indicators are used to contrast the condition signature. In a preferred embodiment, the system comprises a gas turbine engine. The normal signature for the system can be derived from a predefined model of the system that is being monitored. This model can itself be developed off-line and then fixed for the duration of the operation of the health monitoring method. More preferably, however, the model is designed to be refined as the method proceeds in order that it might be better tuned to a specific system. Whichever approach is adopted, it is particularly preferred that the model is a 'learnt model' developed using a data-driven, or at least partially data-driven approach. That is to say the learnt model learns from training data comprising series of the condition indicators which have been labeled as normal (i.e. healthy) or abnormal (i.e. unhealthy) as the case may be. In fact, it is often the case that normal data is far more readily available than abnormal data and therefore the training data may only include examples of normal data. This still results in an effective model, because subsequent events can then be identified as departures from the learnt model of normality. The normal signature for the healthy system may be predicted from a model defining one or more inter-dependencies between the condition indicators. This enables the model to specify a continuous boundary in N-dimensional space (where each dimension relates to one of N condition indicators) corresponding to the limits of healthy system operation. This is in contrast to Nook up table' approaches for setting the limits of healthy system operation which we do not capture the (often complex) inter-relationships and correlations between condition indicators. So, for example, it is often the case that the onset of a problem or failure in a particular system manifests itself in small changes to a number of condition indicators which individually, however, remain in their respective allowable ranges. The 'look up table' approach, which is only able to sense gross shifts in individual parameters, would fail recognize that a problem or failure had occurred. In contrast, when the condition signature for the healthy system is predicted from a model defining one or more-inter-dependencies between the condition indicators the several small changes in the condition indicators may have the cumulative effect of driving the condition signature outside normal boundary in N-dimensional space. Preferably, the predetermined threshold corresponds to a statistically significant departure or variance from normally as defined by the normal signature. Thus, in the example of a normal signature provided by a learnt model, further development of the model (e.g. due to the input of more training data) will result in a corresponding variation in the predetermined threshold. In one embodiment, the invention providers a method for monitoring the health of a system, which comprises performing at each of a plurality of times the steps of: constructing a condition signature from a plurality of condition indicators including (a) a plurality of vibration measurement acquired from the system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from the system; predicting a normal signature from a model defining one or more inter- dependencies between said condition indicators, the normal signature corresponding to the condition signature for a healthy system; comparing the condition signature with the normal signature; and registering an event if the condition signature differs from the normal signature by more than a predetermined threshold. The model may comprise a matrix (e.g. a covariance matrix) with one or more non-zero off-diagonal terms to define the inter-dependencies. The step of comparing the condition signature with the normal signature may then involve calculating a value for the normalized innovations squared (NIS) which is defined below in the 'Description of the Embodiments'. Alternatively, the model may comprise a neural network. If there are N condition indicators, one embodiment is a neural network which is trained to predict the value of the Nth from the other N-1 indicators. The step of comparing the condition signature with the normal signature may then involve calculating a prediction error which is e.g. the square of the difference between the predicted value for N and the actual value. There may be N of these predictive networks operating in parallel for each of the condition indicators. In this case the total prediction error can be the sum of the prediction errors of each of the networks. In another embodiment, a neural network is trained to predict a subset of Nl condition indicators (such as vibration values, e.g. at a number of key frequencies) from another subset of N2 condition indicators (such as the performance parameters), where Nl + N2 = N Preferably the times define successive intervals of at most 1 sec duration (i.e. a 1 Hz repetition frequency). More preferably the times define successive intervals of at most 0.2 sec duration (a 5 Hz repetition frequency), even more preferably at most 0.1 sec (a 10 Hz repetition frequency). By acquiring and processing the condition indicator data at such rates, it is possible for the method to monitor the health of the system in real time. Therefore, if an abnormal event is registered at any time, immediate and appropriate action can be taken by the system operator. This can be particularly advantageous for the operation of safety critical plant such as aero gas turbine engines. The data acquisition rate can, however, be significantly faster than the processing rate. For example the data acquisition rate may be in the range 20 Hz to 80 kHz. Only a subset of the acquired data may then be processed. Thus in another embodiment, the invention provides a method for monitoring the health of a system, which comprises performing at each of a plurality of times defining successive intervals of at most 1 sec duration the steps of: constructing a condition signature from a plurality of condition indicators including (a) a plurality of vibration measurements acquired from the system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from the system; predicting a normal signature corresponding to the condition signature for a healthy system; comparing the condition signature with the normal signature; and registering an event if the condition signature differs from the normal signature by more than a predetermined threshold. where the condition signature is comprised of data from disparate sources, for instance performance and vibration data, a problem occurs in that the data may well not be synchronized in time. If this asynchronous data is combined to form the signature, a distorted picture of the system's health may well result. For similar reasons, training data used to develop a model of normal system behaviour should also be synchronized if distortions in the model are to be avoided. Thus preferably, the condition indicators are synchronously acquired from the system to a synchronization imprecision of at most 1 sec. More preferably the synchronization imprecision is at most 0.1, 0.075, 0.0625 or 0.02 sec. By 'synchronisation imprecision' we mean the maximum difference between the acquisition times of each pair of condition indicators forming a particular condition signature. Desirably, the measurements are acquired from the system at a synchronization imprecision which is less than the duration of the successive time intervals, e.g. if the time intervals are of 0.2 sec duration, the synchronization imprecision may be at most 0.075 sec. The invention also provide a data processing system for monitoring the health of a system suitable for performing the method outlined above. In general terms the data processing system comprises: data acquisition means for acquiring a plurality of measured condition indicators from the system; processor means for constructing a system condition signature from said plurality of measured condition indicators; comparator means for comparing the system condition signature with a predefined normal signature, corresponding to the signature for a healthy system; and means for registering an event if the comparator indicators signature differs from the normal signature by more than a predetermined threshold. The data processing system may further comprises a display means for displaying (a) one or more of the condition indicators, (b) the result of the comparison of the system condition signature with the normal signature and/or (c) an alert signal when the comparator indicates that the predetermined threshold has been transgressed (i.e. an event has been registered). In one embodiment, the invention provides a data processing system for monitoring the health of a system, comprising: data acquisition means for acquiring a plurality of condition indicators from the system at each of a plurality of times, the condition indicators including (a) a plurality of vibration measurements or (b) one or more vibration measurement and one or more performance parameters; processor means for constructing a condition signature from said condition indicators and for predicting a normal signature corresponding to the condition signature for a healthy system, the normal signature being predicted by a model which defines one or more inter-dependencies between said condition indicators; comparator means for comparing the condition signature with the normal signature; and registration means for registering an event if the comparator indicates that the condition signature differs from the normal signature by more than a predetermined threshold. In another embodiment, the invention provides a data processing system for monitoring the health of a system, comprising: data acquisition means for acquiring a plurality of condition indicators from the system at each of a plurality of times defining successive intervals of at most 1 sec duration, the condition indicators including (a) a plaurality of vibration measurements or (b) one or more vibration measurement and one or more performance parameter measurements; processor means for condition signature from said condition indicators and for predicting a normal signature corresponding to the condition signature for a healthy system; comparator means for comparing the condition signature with the normal signature; and registration means for registering an event if the comparator indicates that the condition signature differs from the normal signature by more than a predetermined threshold. A further aspect of the invention addressed the problem of the synchronous acquisition of the condition indicators. The invention proposes to associate time stamps (based on a common clock) with the acquired data and to synchronise the data on the basis of these time stamps. Accordingly, in this aspect, the invention provides a method of sychronising two or more data streams, each data stream comprising a series of sequentially acquired data elements (and relating e.g. to a respective condition indicator of the previous aspect), the method comprising: Associating a time stamp with each data element of each stream, the time stamp identifying the time of acquisition of the data element on the basis of a clock common to all data streams; selecting a first data element from a first stream and inspecting its time stamp; conducting a search of the data elements of the or each other stream to identify the data element in the or each other element having an associated time stamp closest to that of the selected element of the first stream; and marking said identified data element of the or each other stream and said selected element of eth first stream as being synchronized with one another. Because, the relative acquisition times of the data elements are generally more significant than their absolute acquisition times, the common clock may be operate within an absolute or relative framework. In an absolute framework one clock provides the time stamp for each data element of each data stream. In a relative framework each data stream has its own clock, and one of the clock is selected as the reference clock against which the acquisition times of the other data streams are measured. It may be convenient to use a mixture of absolute and relative frameworks. For example, if the data streams relates to performance parameter and vibration measurements, the performance parameters may be time stamped from one clock and the vibration measurements from another clock. The process can be repeated until the data elements in the first stream have been exhausted, in any subsequent processing of the data that is reliant on using synchronized data streams, only those data elements marked as being synchronized with one another are used. In the case where the acquisition rates of the data streams differ from one another, it is preferred that the first stream, with which the other streams are synchronized, is chosen to be the stream having the lowest acquisition rate. The invention further provides a data processing system for synchronizing two or more data streams, each data stream comprising a series of sequentially acquired data elements, comprising: means for associating a time stamp with each data element of each stream, the time stamp identifying the time of acquisition of the data element on the basis of a clock common to all data streams; means for selecting a first data element from a first stream and inspecting its time stamp; means for conduction search of the data elements of the or each other stream to identify the data element in the or each other element having an associated tome stamp closest to that of the selected element of the first stream; and means for making said identified data element of the or each other stream and said selected element of the first stream as being synchronized with one another. BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS The various aspect of the invention will be further described by way of example with reference to the accompanying drawings, in which: Fig. 1 schematically illustrates an exemplary data structure that can be adopted for operation of the second aspect of the invention; Fig. 2 shows a neural network architecture for a learnt model for operation of the first aspect of the invention; Fig. 3 shows a graph of the prediction error for the learnt model on a set of test data corresponding to a period of normal operating conditions for an engine; Fig. 4 shows a graph of the prediction error for the learnt model for a further engine operating period in which the engine experienced a bird strike; Fig. 5 shows the learning curve for a simple example of a system model for operation for the first aspect of the invention; Fig. 6 shows a comparison of observations and modeled estimates for a shaft speed measurement, illustrating evolution of the model of Fig. 5. Fig. 7 shows the measured low pressure shaft speed (N1V) for the period of the test data from a more elaborate example of the system model, Fig. 8 shows the value for the NIS over the same period as Fig. 7, Fig. 9 shows the values for the 13 condition indicators and the NIS over the same period from a further example of the system model, and Fig. 10 shows a schematic example of an on-the-engine health monitoring system. DETAIL DESCRIPTION OF A PREFERRED EMBODIMENT OF THE INVENTION The embodiment described below is an example of a data processing system employing both aspects of the invention discussed above. More specifically, it is a system for synchronous acquisition, analysis and display of performance parameters and vibration data from a power plant (e.g. a gas turbine), for monitoring the health of the plant. In accordance with a preferred from of the second aspect, the performance and vibration data streams are synchronized in real time and, in accordance with a preferred aspect of the first aspect of the invention, these data are combined or fused to construct a signature for the system that can be compared to a signature derived from a model representing a healthy power plant, in order to provide anomaly/event detection and hence fault diagnosis. The following discussion focuses on a application of the system to monitoring the health of a gas turbine aero-engine, but it will be appreciated that the methods can be adapted to other power plant, including for example ground-based and marine gas turbine, and spark ignition and compression ignition internal combustion engines, as well as other mechanical, thermodynamic, fluid, electrical or electronic system. The system acquires performance parameters from the gas turbine digitally via an Ethernet link at a rate between 20 and 40Hz. Typical performance parameters are measurements of pressure, temperature, thrust, altitude or March number. Vibration data is acquired from analogue vibration transducers which are sampled at user-selectable sampling rates (from 625Hz to 80KHz) via an analogue-to-digital converter. The amplitude spectrum of the vibration data is generated using the Fast Fourier Transform every 0.2 sec. The performance and vibration data streams are asynchronous and stored in separate files together with the corresponding timestamps. During review, as data is loaded into memory, synchronization is performed between the performance and spectrum data on a line by line basis. Markers 10,12 (see Fig. 1) are kept which record the last synchronized line in the vibration and performance data ring buffers 14,16. When new data is available in memory, the time stamp for the next vibration spectrum line is examined. The synchronization algorithm starts from the last previously synchronized location in the performance data and searches forwards or backwards based on the timestamps of the performance data (accurate to 0.05 sec) until the closest matching timestamp in the performance data ring buffer 16 is identified. This location in the performance data is recorded as being synchronized with the line in the vibration ring buffer 14. The algorithm then proceeds to the next line in the vibration ring buffer 14 (0.2 sec later) and so on until there is no more data available to synchronise. Clearly, therefore, if the performance parameters are acquired at 20 Hz (i.e. at 0.05 sec intervals) the synchronization precision is 0.075 sec (i.e. half the acquisition interval added to the accuracy of the timestamps) and if the performance parameters are acquired at 40 Hz (i.e. at 0.025 sec intervals) the synchronization precision is 0.0625 sec. Considering the synchronization algorithm in a little more detail, it can be seen from Fig. i that the algorithm maintains a synchronization table 18 that gives the index of the performance data entry that matches each vibration data line. The algorithm uses variables to mark the latest synchronized data in each buffer. The operation of the algorithm can be summarized by the following 'pseudo code': 1. Initialise the latest synchronized markers to the start of the vibration and performance data. 2. Loop while there is more data in both ring buffers. (a) Starting from the latest synchronized data item in each ring buffer, examine the time stamp, t, on the next entry in the vibration ring buffer. (b) search forward in the performance ring buffer until a time stamp greater than t is found. Select between this entry in the performance ring buffer and the previous entry for one which is closest to t and record the match in the synchronization table. Once synchronized, the analysis of this performance and vibration data relies on constructing models of normal jet engine behaviour and then detecting an event or an abnormality with respect to these models. Traditional aircraft engine monitoring systems are based on two distinct processes: the use of vibration signatures to indicate engine state, and a separate procedure, gas-path analysis, which is employed for determination of state from performance parameters. In the approach described now, however, performance-retaled parameters such as pressure and temperature can be fused with vibration data (such as tracked order vectors - the elements of which correspond to the signal amplitude in a narrow range centered on the main vibration frequencies for each shaft of the turbine). The aim is to take advantage of disparate sources of data to from a more comprehensive picture of engine state during normal operation. This in turn should allow a wider range of deviations to be identified. Furthermore, it is proposed to employ learnt data-driven models a normal engine. Thus, although models of the engine system are used, these are not fixed. Instead, they evolve with acquired training data. This offers the important advantage of robustness. The methods of data analysis described below may be termed 'novelty detection'. An advantage of the methods is that the role of the expert need only be retained in classifying training data as abnormal (i.e. novel) or normal. The use of Kalman filtering systems in novelty detection has been described in e.g. M. Gelb, Applied Optimal Estimation. MIT Press 1974. Two alternative data analysis methods are described below. They are distinguished by the amount of prior knowledge required to set up the system. In both cases, the role of the expert need only be retained in classifying training data as novel or normal. The first method relies on a prior leant model of normality. For example, normal vibration tracked order shaped are learnt using a simple clustering model for the normal data. The novelty of e.g. the vibration signature for an engine under test is assessed by comparing the closeness of its tracked order signature with the prototypical patterns in the clustering model of normality. This can be done, for example, by computing the shortest normalized Euclidean distance between the vector encoding the tracked order shaped to any of the (prototypical patterns) cluster centers in the model of normality (see Nairac at al, 'A System for the Analysis of Jet Engine Vibration Data', Integrated Computer-aided Engineering, 6(l):53-65, 1999). If this distance is beyond a previously set threshold, the vibration signature as represented by that tracked order is deemed to be outside the bounds of normality. In addition to the vibration tracked orders, the model of normality for the vibration spectra includes the following: sidebands, multiple harmonics, fractional harmonics and broadband power. The model is illustrated by an example in which a neural network having the architecture shown in Fig. 2 was developed as the learnt model. The neural network had an input layer 30 with four nodes consisting of four condition indicator measured relating to one shaft of a multi-shaft test engine. The condition indicators were the vibration amplitude, the phase and the shaft speed all at a specified time, and the shaft speed a time increment after the specified time. The output layer 32 of the network had two nodes for predicting respectively the change in vibration amplitude and change in phase after the time increment. The network had one hidden layer 34, each node of which contained a Gaussian radial basis function. The training phase for network used training data obtained from the test engine over a range of normal operating conditions. The centers and the spreads of the Gaussians were fixed using the cluster analysis described above and the weights of the connections between the nodes were then iterativeiy adjusted until the model converged. Figure 3 shows a graph of the prediction error (that is the sum of the prediction errors of the change in vibration amplitude and change in phase) for the model on a set of test data which also corresponded to the period of normal operating conditions for the engine. This graph provides a baseline of prediction error variation against which novel events can be judged. The neural network had an input layer 30 with four nodes for a condition signature consisting of four condition indicator measured relating to one shaft of a multi-shaft test engine. The condition indicators were the vibration amplitude, the phase and the shaft speed all at a specified time, and the shaft speed a time increment after the specified time. The output layer 32 of the network had two nodes for predicting respectively the change in vibration amplitude and change in phase after the time increment. The network had one hidden layer 34, each node of which contained a Gaussian radial basis function. The training phase for network used training data obtained from the test engine over a range of normal operating conditions. The centers and the spreads of the Gaussians were fixed using the cluster analysis described above and the weights of the connections between the nodes were then iteratively adjusted until the model coveraged. Fig. 3 shows a graph of the prediction error (i.e. the sum of the prediction errors of the change in vibration amplitude and change in phase) for the model on a set of test data which also corresponded to the period of normal operating conditions for the engine. This graph provides a baseline of prediction error variation against which novel events can be judged. Fig. 4 shows a graph of the prediction error for a further engine operating period. In this case, however, the engine experienced a bird strike. The largest peak in the graph corresponds to the moment of bird impact. Clearly the model was able to recognize this event. However, the changes prediction error signal (compared to the baseline of Fig. 3) after the event showed that the model was also able to detect post-impact abnormal engine behaviour. This provides confidence that the model can not only detect major events such as bird strikes, but also more subtle deviations from normality. The second method employs a process model which has a state vector associated with it (see below). The observation vector (i.e. the condition signature) has elements corresponding to measured values of performance parameters and vibration information so that two types of data are fused within the model. The fusion of the data is performed in real-time with a new output being generated by the system several times a second. An important aspect of the use of this model in the system is the use of learning. In a first, off-line, phase of learning, a generic model of the engine is learnt. The learning is data-driven using an algorithm such as Expectation-Maximisation in order to maximize the likelihood of the learnt model given the training data. Once such a generic model has been learnt off-line for a particular type of engine, learning can then be applied on-line in order to tune the model to an individual engine immediately after its pass-off test and after each maintenance procedure. Engine deterioration can also be learnt on-line. The learnt model can be turned to different flight conditions, such as cruising or landing, in order to detect novelty with even more sensitivity and specificity. The data-driven learnt model may be integrated with existing performance models which rely on the laws of thermodynamics and computational fluid dynamics (knowledge-based models). Such models can therefore be described as hybrid models because they are based on the integration of learnt and knowledge-based models. Looking in more detail at the learnt modeling approach, it is based on the application of Expectation Maximisation (EM) to parameter estimation in linear dynamical systems (see Ghahramani and Hinton, parameter Estimation for Linear Dynamical Systems, technical Report CRG-TR-96-2, University of Tornoto, 1996) and to non-linear systems (see Rowels and Ghahramani, 'A Unifying Review of Linear Gaussian Models', Neural Computation, 11, 305-345, 1999; and Ghahramani and Roweis, 'Learning in Nonlinear Dynamical System Using an EM Algorithm' in Kearns et al. (editors), advances in Neural Information Processing System, Volume 11, MIT Press, 1999). The EM learning algorithm is applied to a Kalman filter model. In the linear case, this is a system with a measurement process of the form Where y(i) is a set of observations of hidden state x(i), C is a covariance matrix, and measurement noise v(i) is zero-mean and normally distributed with covariance matrix R. y(i) and x(i) can be the same dimension. Non-zero off- diagonal terms in C allow the model to account for inter-dependencies between the performance parameter and vibration measurements of the condition and normal signatures. The state equation is with w(i) zero-mean and normally distributed with covariance matrix Q. At the beginning of the training phase A and C are initialized to small random values (e.g. with elements of the matrices ~ 10-5), and R and Q are initialized e.g. to I. Then during the training phase, for each condition signature y(i) in the training set, the method of rowels and Ghahramani applied to equation (2) to derive the most likely values for the elements of x(i), and the elements of C, R and v(i) are iteratively adjusted so that Cx(i) +v(i) converges to the respective condition signature (R and Q can be constrained throughout to be diagonal matrices). Convergence can be determined by the log-likelihood of the set of observations given the model. Instead of initializing the elements of A to small random values, it is also possible to adopt initial values that embody existing performance models of engine behaviour. After the training phase, the model would then be a hybrid of a knowledge-based and a data-driven model. By fusing these two methods of data-analysis, the accuracy of prior expert knowledge can be combined with the robustness of data-driven approaches. When the training phase has ended and the model is receiving real-time data consisting of a sequence of condition signature, the Kalman filter is again used to derive the most likely values for the elements of x(i) for each condition signature y(i). However, the elements of C and v(i) are now fixed, so Cx(i) +v(i) provides the normal signature for comparison with the condition signature., For example, comparison of the normal signature with the condition signature can be on the basis of the normalized innovations squared (NIS). The innovations sequence v is the difference between the condition signature and the normal signature, so The innovations should be zero-mean and white. The NIS combines the individual innovations sequences. The individual sequence are weighted by the term S(k)1, the inverse of the innovation covariance given by where P(kl k-1) is the prediction covariance. The model is first illustrated by a simple example (which does not use vibration measurements) where observations are made of the speeds of the three shafts of a test engine during cruise. The observed data y is simply the state x corrupted by noise, so The observations are used during the learning process, to generate a dynamical system model in which A, C, Q and R are learned from the data. At the beginning of the training phase A and C were initialized to small random values and R and Q were initialized to I. Fig. 5 illustrates the learning (log likehood) plot for the system. Fig. 6 shows the evolution of estimates of shaft 1 speeds during the learning process using the EM algorithm. In the example shown, the learning stage lasts for the first 25 iterations. From iteration 25 onwards, the system's dynamical properties are determined by the learned matrices (which are then kept fixed). Once trained, the systems can be used to detect events or abnormalities, that is to say divergences from the learnt model of normality. The events of particular interest are those that are unexpected, possibly indicating a problem with the engine for example. However, particularly where the models have been learned only for 'steady state' parts of the flight envelope (e.g. acceleration, cruise and deceleration), transients during operation of the engine will also be flagged up as events, although they are expected. For example, where a bleed valve is opened or closed, the operating condition of the engine will exhibit significant differences from a learnt model of steady state normality which does not include this event. Thus when using such a steady state model, measures can be employed to avoid these transient events. For instance, since the opening of a bleed valve is an event that occurs at a defined point in time, the data collected from the engine at the time and slightly either side of it (e.g. for 2 seconds before and after) can be eliminated from the data analysed by the health monitoring system. The approach is next illustrated with a more elaborate example which uses both performance parameters and vibration data. In this example the model was applied to data acquired from a test bed-mounted, multi-shaft, aero gas turbine engine which was undergoing a cyclic acceleration-deceleration test when it experienced an intermediate pressure turbine lock-plate event. Such tests are used to investigate engine behaviour under extreme conditions. The data was acquired synchronously as described above. A lock-plate event occurred in the engine on test day 152-00. The training data was 152-00 data for a period before the event, and the test data was 152-00 data for the period including the event. In this example a 14 dimensional model (i.e. y(i) and x(i) each had 14 elements) was used in which the condition indicator inputs were: - The tracked orders of the low pressure (tol), intermediate pressure (toi) and high pressure (toh) shafts, - The shaft speeds of the low pressure (NIV), intermediate pressure (N2V) and high pressure (N3V) shafts, - The ambient (POV), total inlet (P20V), high pressure compressor delivery (P30V) and exhaust (PEXV) pressures, - The total inlet (T20V) and high pressure compressor delivery (T30V) temperatures, - The turbine gas temperature trimmed (TGTTRM) - The demanded fuel flow (WFDEM) At the beginning of the training phase A and C were initialized to small random values and R and Q were initialized to I. Fig. 7 shows the measured low pressure shaft speed (NIV) for part of the period of the test data, and Fig. 8 shows the value for the NIS calculated by the trained model over the same period. The first two sharp troughs in the NIV trace were caused by planned consecutive cyclic decelerations. Associated with each of these troughs are two NIS peaks. These peaks indicated that the engine was not behaving normally during the cyclic testing. In fact subsequent examination revealed that a lock plate had released earlier during the test and as a result abnormal blade rubbing was occurring during each of the deceleration cycles. During the third planned deceleration cycle in the period covered by Figs. 7 and 8 (i.e. at approximately data point 12850) the engine sustained blade damage which resulted in the sharp NIS peak and drop off in NIV. However, the earlier (all be it smaller) NIS peaks demonstrate that the monitoring system was able detect the effect of the lock plate release in real time and before substantial blade damage was sustained. If such a release had occurred in an in-service aero engine, it would therefore have been possible to generate an immediate warning so that timely action (such as engine inspection or maintenance) could have been performed. In contrast, sudden variations in NIV can occur normally, so Nlv alone is not a reliable indicator of abnormal behaviour. A further example also uses both performance parameter and vibration data. Again the model was applied to synchronously data acquired from a test bed- mounted, multi-shaft, aero gas turbine engine. However, in this case an oil seal leak developed in the engine. The fault occurred in the engine around data point 50410. The training data was from a period before the fault, and the test data was for a period including the fault. In this example a 13 dimensional model was used in which the condition indicator inputs were tol, toi, toh, N1V, N2V, N3V, P20V, P30V, PEXV, T20V, TGTTRM, AND WFDEM. Fig. 9 shows the values for the 13 condition indicators and the NIS (in the bottom graph) over the period including the event. The sharp NIS peak at data point 50410 again demonstrates that the monitoring system was able to detect the moment of the event. Although some of the other condition indicators also had peaks at this time, by themselves they cannot be reliably associated with abnormal (novel) behaviour. Subsequent examination showed that the event was an oil seal which allowed engine oil to enter the combustor. Because of this leak the engine control system reduced the amount of fuel entering the combustor (hence the drop in WEDEM) to maintain the engine thrust (given by PEXV / P20) at a constant level. After the event, with oil continuing to leak into the combustor, the engine was operating outside the scope of normal behaviour. This was successfully picked up by the NIS trace which after data point 50410 did not return to its pre-event level. Next we consider how a health monitoring system, incorporating the model described above, could be installed for in-flight analysis of aero gas turbines. An on-the engine system, shown schematically in Fig. 10, could generate of the order of 1 Gb of vibration and performance data (consisting mainly of pressures, temperatures and shaft speeds) per flight. The vibration data is usually analysed in the frequency domain. The vibration and performance data, as there are being generated by data acquisition means 20, are temporarily stored in ring buffer 22. The data is synchronized and subjected to novelty detection in processor and comparator means 24 which receives a synchronisation signal from data acquisition means 20 and the data from ring buffer 22. Those sections of the data corresponding to novel events are then tagged and recorded with no loss of information (i.e. highbandwidth data is recorded) in registration means 26 which has semi-permanent on-line and/or hard disk storage. When the flight is completed the stored data may be downloaded and subjected to more intensive ground-based analysis. The system may also include a display which is driven to allow information to be displayed either during acquisition or for review once an acquisition cycle has been completed, it preferably includes the following features: Ability to display the result of the comparison of the condition signature with the normal signature, e.g. in the form of the NIS or the prediction error. An unhealthy event nay be highlighted e.g. with an alert signal. Ability is display a combination of any two of vibration spectra, tracked orders, broadband power, performance parameters synchronized in time. Ability to extract and plot vibration spectra against engine speed. Ability to interrogate and print any of vibration spectra, tracked orders, broadband power and performance parameters. Automatic detection and display of features from vibration spectra (sidebands, harmonics, etc.) WE CLAIM 1. A method for monitoring the health of a plant, which comprises performing at each of a plurality of times the steps of: constructing a condition signature for a present time from a plurality of condition indicators having (a) a plurality of vibration measurements acquired from said system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from said plant; predicting a normal signature from a model defining one or more inter- dependencies between condition indicators used to construct the condition signature for a previous time, said normal signature corresponding to a condition signature for a healthy plant at the present time, and said model comprising a matrix with one or more non-zero off-diagonal terms to define said inter-dependencies; comparing said condition signature for the present time with said normal signature; and registering an event if said condition signature for the present time differs from said normal signature by more than a predetermined threshold. 2. A method as claimed in claim 1, wherein said model is a learned model. 3. A method as claimed in claim 1, wherein the step of comparing said condition signature with said normal signature involves calculating a value for the normalised innovations squared. 4. A method as claimed in claim 1, wherein said model comprises a neural network. 5. A method as claimed in claim 4, wherein the step of comparing said condition signature with said normal signature involves calculating a prediction error. 6. A method as claimed in claim 1, wherein said times define successive intervals of at most 1 sec duration. 7. A method for monitoring the health of a plant, which comprises performing at each of a plurality of times defining successive intervals of at most 1 sec duration the steps of: constructing a condition signature for a present time from a plurality of condition indicators including (a) a plurality of vibration measurements acquired from the system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from said system, said measurements being synchronously acquired from said plant to a synchronization imprecision of at most 1 sec; predicting, from condition indicators used to construct the condition signature for a previous time, a normal signature corresponding to a condition signature for a healthy system at the present time; comparing said condition signature for the present time with said normal signature; and registering an event if said condition signature for the present time differs from said normal signature by more than a predetermined threshold. 8. A method as claimed in claim 7, wherein said normal signature is predicted from a model defining one or more inter-dependencies between said condition indicators used to construct the condition signature for the previous time. 9. A method as claimed in claim 8, wherein said model is a learned model. 10. A method as claimed in claim 8, wherein said model comprises a matrix with one or more non-zero off-diagonal terms to define said interdependences. 11. A method as claimed in claim 10, wherein the step of comparing said condition signature with said normal signature involves calculating a value for the normalised innovations squared. 12. A method as claimed in claim 8, wherein said model comprises a neural network. 13. A method as claimed in claim 12, wherein the step of comparing said condition signature with said normal signature involves calculating a prediction error. 14. A method as claimed in any one of claim 1, wherein said measurements are synchronously acquired from said system to a synchronisation imprecision of at most 1 sec. 15. A method as claimed in any one of claims 1, wherein said system comprises a gas turbine engine. 16. A data processing system for monitoring the health of a system, comprising: data acquisition means for acquiring a plurality of condition indicators from said system at each of a plurality of times, said condition indicators including (a) a plurality of vibration measurements or (b) one or more vibration measurements and one or more performance parameter measurements; processor means for constructing a condition signature for a present time from said condition indicators and for predicting a normal signature corresponding to a condition signature for a healthy system at the present time, said normal signature being predicted by a model which defines one or more inter-dependencies between condition indicators used to construct the condition signature for a previous time, wherein said mode! comprises a matrix with one or more non-zero off-diagonal terms to define said inter-dependencies; comparator means for comparing said condition signature for the present time with said normal signature; and registration means for registering an event if said comparator for the present time indicates that said condition signature differs from said normal signature by more than a predetermined threshold. 17. A data processing system for monitoring the health of a system, comprising: data acquisition means for acquiring a plurality of condition indicators from said system at each of a plurality of times defining successive intervals of at most 1 sec duration, said measurements being synchronously acquired from said system to a synchronization imprecision of at most 1 sec, and said condition indicators including (a) a plurality of vibration measurements or (b) one or more vibration measurements and one or more performance parameter measurements; processor means for constructing a condition signature for a present time from said condition indicators and for predicting, from condition indicators used to construct the condition signature for a previous time, a normal signature corresponding to a condition signature for a healthy system at the present time; comparator means for comparing said condition signature for the present time with said normal signature; and registration means for registering an event if said comparator for the present time indicates that said condition signature differs from said normal signature by more than a predetermined threshold. 18.A method for monitoring the health of a system, which comprises performing at each of a plurality of times the steps of; constructing a condition signature for the present time from a plurality of condition indicators including (a) a plurality of vibration measurements acquired from said system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from said system, said measurements being synchronously acquired from said system to be a synchronization imprecision of at most i sec; predicting a normal signature from a model defining one or more inter-dependencies between condition indicators used to construct the condition signature for the previous time, said normal signature corresponding to a condition signature for a healthy system at the present time; comparing said condition signature for the present time with said normal signature; and registering an event if said condition signature for the present time differs from said normal signature by more than a predetermined threshold. 19. A data processing system for monitoring the health of a system, comprising: data acquisition means for synchronously acquiring a plurality of condition indicators from said system at each of a plurality of times to a synchronization imprecision of at most 1 sec, said condition indicators including (a) a plurality of vibration measurements or (b) one or more vibration measurements and one or more performance parameter measurements; processor means for constructing a condition signature for the present time from said condition indicators and for predicting a normal signature corresponding to a condition signature for a healthy system at the present time, said normal signature being predicted by a model which defines one or more inter-dependencies between condition indicators used to construct the condition signature for the present time; comparator means for comparing said condition signature for the present time with said normal signature; and registration means for registering an event if said comparator for the present time indicates that said condition signature differs from said normal signature by more than a predetermined threshold. This invention relates to a method for monitoring the health of a system comprises performing at each of a plurality of times the steps of: constructing a condition signature from a plurality of condition indicators including (a) a plurality of vibration measurements acquired from the system or (b) one or more vibration measurements and one or more performance parameter measurements acquired from the system; predicting a normal signature from a model defining one or more inter-dependencies between said condition indicators, the normal signature corresponding to the condition signature for a healthy system; comparing the condition signature with the normal signature; and registering an event if the condition signature differs from the normal signature by more than a predetermined threshold.

Full Text

FIELD OF INVENTION
This invention relates to a methods and data processing systems for monitoring
the health of a system. The methods and data processing systems of the
invention are particularly, although not necessarily exclusively, suitable for
monitoring the health of power plant, including for example gas turbine, spark
ignition and compression ignition internal combustion engines.
BACKGROUND OF THE INVENTION
The health of the system can be considered a measure of the condition of a
system against expected norms. A health system is one whose condition closely
matches expectations, whereas an unhealthy system is one whose condition
differs from what would be expected, indicating for example deterioration of, or
a possible problem with the system. The ability to monitor the health of a system
can therefore allow such deterioration and/or problems to be detected and, if
necessary, addressed at an early stage.
For example, US-5684718 describes a non-real time system for monitoring the
operation of an electric generator in which vibration and load data are combined
to produce a single which is then compared with stored data representative of
maximum acceptance combinations of the two parameters. Essentially the
system is an automated Nook up table' which issues warnings when vibrations
have exceed acceptable limits.
In order to determine the condition, and consequently health, of a system, it is
normal to monitor and analyse a series of measurable indicators which
themselves reflect performance parameters such as turbine and compressor
operating temperature and pressures and spool speeds. To obtain a fuller overall
picture of the engine's condition, these performance parameters can be
supplemented with further condition indicators including, for example, vibration
measurements and measurements of the particulate material entrained in the
circulating oil.
Particularly with complex mechanical systems such as gas turbines, the number
of indicators that must be monitoring to obtain a useful overall picture of the
system's condition can be high. This in turn means that the task of analysing the
complete series of indicators to determine the health of the engine is a complex
one, typically requiring a skilled expert to analyse the data off-line.
Taking again the example of a gas turbine, it is known for example to collect
performance and vibration data from the engine over time to be analysed off-line
by one or more experts. Typically the performance data will be compared with
simulated data for the same engine and, based on this comparison, an expert
will from a view as to the health of the engine. Additionally, a small amount of
vibration data will be reviewed. Giving a superficial view of gross changes in
engine behaviour. If a problem is detected, the vibration data may then be
analysed in more detail, often by another expert, to look for any abnormal
indications which might be symptomatic of underlying mechanical problems
which could lead to a loss of health and operability.
SUMMARY OF THE INVENTION
It is a general aim of the present invention to provide methods and data
processing systems that facilitate the acquisition and analysis of condition
indicators in a manner such that the overall health of a system can be more
readily assessed.
Accordingly, in general terms a first aspect of the invention provides a method
for monitoring the health of a system, comprising:
constructing a condition signature from a plurality of measured condition
indicators acquired from the system;
comparing the condition signature with a normal signature, corresponding to the
signature for a healthy system; and
registering an event if the condition signature differs from the normal signature
by more than a predetermined threshold.
The term 'signature', as used herein, pertain to the values of a plurality of
condition indicators merged or fused into a unit or quality such as a set, vector
or scalar. In the example of a vector signature, the indicators may correspond to
respective elements of the vector. In the example of a scalar signature, the
magnitude of the scalar may be determined by a mathematical function which
acts upon the indicator values.
By merging or fusing the condition indicators into a single signature in this
manner, and providing a normal signature with which the fused data can be
compared, the task of assessing the health of a system is greatly simplified. In
particular, since the detection of an event amounts to an indication of a potential
problem or an unhealthy system (i.e. a system condition that differs from what
would normally be expected), the monitoring of the health can be largely
automated, removing, or at least minimizing, the requirement for expert input
during the monitoring process. This in turn means that it becomes feasible to
continuously monitor the health of a system, and to provide useful information
about the health of the system in real time during operation.
Preferably the condition indicators that are combined to form the system
condition signature include operational parameters, which in the case of a
mechanical system may be speeds, pressure (e.g. gas pressures, oil pressures)
and temperature for example. Other useful parameters may include what might
be conventionally thought of as control or status parameters. For convenience,
such parameters will be referred to using the single label of 'performance
parameters' in the following text.
Additionally, to obtain a fuller picture of a mechanical system's heath, it is
particularly preferred that the signature includes one or more condition indicators
related to the vibration of the system.
Put, more generally, the condition indicators from which the system condition
signature is constructed may be derived from two or more disparate sources of
data. This illustrates a particular strength of this approach in that a great variety
of different forms of condition indicator data can be encompassed in the system
condition signature, providing a more comprehensive measure of the system's
health than has previously been possible without multiple analyses.
Preferably at least three condition indicators are used to contrast the condition
signature. More preferably at least 10 and even more preferably at least 20
condition indicators are used to contrast the condition signature.
In a preferred embodiment, the system comprises a gas turbine engine.
The normal signature for the system can be derived from a predefined model of
the system that is being monitored. This model can itself be developed off-line
and then fixed for the duration of the operation of the health monitoring method.
More preferably, however, the model is designed to be refined as the method
proceeds in order that it might be better tuned to a specific system.
Whichever approach is adopted, it is particularly preferred that the model is a
'learnt model' developed using a data-driven, or at least partially data-driven
approach. That is to say the learnt model learns from training data comprising
series of the condition indicators which have been labeled as normal (i.e.
healthy) or abnormal (i.e. unhealthy) as the case may be. In fact, it is often the
case that normal data is far more readily available than abnormal data and
therefore the training data may only include examples of normal data. This still
results in an effective model, because subsequent events can then be identified
as departures from the learnt model of normality.
The normal signature for the healthy system may be predicted from a model
defining one or more inter-dependencies between the condition indicators. This
enables the model to specify a continuous boundary in N-dimensional space
(where each dimension relates to one of N condition indicators) corresponding to
the limits of healthy system operation. This is in contrast to Nook up table'
approaches for setting the limits of healthy system operation which we do not
capture the (often complex) inter-relationships and correlations between
condition indicators.
So, for example, it is often the case that the onset of a problem or failure in a
particular system manifests itself in small changes to a number of condition
indicators which individually, however, remain in their respective allowable
ranges. The 'look up table' approach, which is only able to sense gross shifts in
individual parameters, would fail recognize that a problem or failure had
occurred. In contrast, when the condition signature for the healthy system is
predicted from a model defining one or more-inter-dependencies between the
condition indicators the several small changes in the condition indicators may
have the cumulative effect of driving the condition signature outside normal
boundary in N-dimensional space.
Preferably, the predetermined threshold corresponds to a statistically significant
departure or variance from normally as defined by the normal signature. Thus, in
the example of a normal signature provided by a learnt model, further
development of the model (e.g. due to the input of more training data) will result
in a corresponding variation in the predetermined threshold.
In one embodiment, the invention providers a method for monitoring the health
of a system, which comprises performing at each of a plurality of times the steps
of:
constructing a condition signature from a plurality of condition indicators
including (a) a plurality of vibration measurement acquired from the system or
(b) one or more vibration measurements and one or more performance
parameter measurements acquired from the system;
predicting a normal signature from a model defining one or more inter-
dependencies between said condition indicators, the normal signature
corresponding to the condition signature for a healthy system;
comparing the condition signature with the normal signature; and
registering an event if the condition signature differs from the normal signature
by more than a predetermined threshold.
The model may comprise a matrix (e.g. a covariance matrix) with one or more
non-zero off-diagonal terms to define the inter-dependencies. The step of
comparing the condition signature with the normal signature may then involve
calculating a value for the normalized innovations squared (NIS) which is defined
below in the 'Description of the Embodiments'.
Alternatively, the model may comprise a neural network. If there are N condition
indicators, one embodiment is a neural network which is trained to predict the
value of the Nth from the other N-1 indicators. The step of comparing the
condition signature with the normal signature may then involve calculating a
prediction error which is e.g. the square of the difference between the predicted
value for N and the actual value. There may be N of these predictive networks
operating in parallel for each of the condition indicators. In this case the total
prediction error can be the sum of the prediction errors of each of the networks.
In another embodiment, a neural network is trained to predict a subset of Nl
condition indicators (such as vibration values, e.g. at a number of key
frequencies) from another subset of N2 condition indicators (such as the
performance parameters), where Nl + N2 = N
Preferably the times define successive intervals of at most 1 sec duration (i.e. a 1
Hz repetition frequency). More preferably the times define successive intervals of
at most 0.2 sec duration (a 5 Hz repetition frequency), even more preferably at
most 0.1 sec (a 10 Hz repetition frequency). By acquiring and processing the
condition indicator data at such rates, it is possible for the method to monitor the
health of the system in real time. Therefore, if an abnormal event is registered at
any time, immediate and appropriate action can be taken by the system
operator. This can be particularly advantageous for the operation of safety
critical plant such as aero gas turbine engines.
The data acquisition rate can, however, be significantly faster than the
processing rate. For example the data acquisition rate may be in the range 20 Hz
to 80 kHz. Only a subset of the acquired data may then be processed.
Thus in another embodiment, the invention provides a method for monitoring
the health of a system, which comprises performing at each of a plurality of
times defining successive intervals of at most 1 sec duration the steps of:
constructing a condition signature from a plurality of condition indicators
including (a) a plurality of vibration measurements acquired from the system or
(b) one or more vibration measurements and one or more performance
parameter measurements acquired from the system;
predicting a normal signature corresponding to the condition signature for a
healthy system;
comparing the condition signature with the normal signature; and
registering an event if the condition signature differs from the normal signature
by more than a predetermined threshold.
where the condition signature is comprised of data from disparate sources, for
instance performance and vibration data, a problem occurs in that the data may
well not be synchronized in time. If this asynchronous data is combined to form
the signature, a distorted picture of the system's health may well result. For
similar reasons, training data used to develop a model of normal system
behaviour should also be synchronized if distortions in the model are to be
avoided.
Thus preferably, the condition indicators are synchronously acquired from the
system to a synchronization imprecision of at most 1 sec. More preferably the
synchronization imprecision is at most 0.1, 0.075, 0.0625 or 0.02 sec. By
'synchronisation imprecision' we mean the maximum difference between the
acquisition times of each pair of condition indicators forming a particular
condition signature. Desirably, the measurements are acquired from the system
at a synchronization imprecision which is less than the duration of the successive
time intervals, e.g. if the time intervals are of 0.2 sec duration, the
synchronization imprecision may be at most 0.075 sec.
The invention also provide a data processing system for monitoring the health of
a system suitable for performing the method outlined above. In general terms
the data processing system comprises:
data acquisition means for acquiring a plurality of measured condition indicators
from the system;
processor means for constructing a system condition signature from said plurality
of measured condition indicators;
comparator means for comparing the system condition signature with a
predefined normal signature, corresponding to the signature for a healthy
system; and
means for registering an event if the comparator indicators signature differs from
the normal signature by more than a predetermined threshold.
The data processing system may further comprises a display means for
displaying (a) one or more of the condition indicators, (b) the result of the
comparison of the system condition signature with the normal signature and/or
(c) an alert signal when the comparator indicates that the predetermined
threshold has been transgressed (i.e. an event has been registered).
In one embodiment, the invention provides a data processing system for
monitoring the health of a system, comprising:
data acquisition means for acquiring a plurality of condition indicators from the
system at each of a plurality of times, the condition indicators including (a) a
plurality of vibration measurements or (b) one or more vibration measurement
and one or more performance parameters;
processor means for constructing a condition signature from said condition
indicators and for predicting a normal signature corresponding to the condition
signature for a healthy system, the normal signature being predicted by a model
which defines one or more inter-dependencies between said condition indicators;
comparator means for comparing the condition signature with the normal
signature; and
registration means for registering an event if the comparator indicates that the
condition signature differs from the normal signature by more than a
predetermined threshold.
In another embodiment, the invention provides a data processing system for
monitoring the health of a system, comprising:
data acquisition means for acquiring a plurality of condition indicators from the
system at each of a plurality of times defining successive intervals of at most 1
sec duration, the condition indicators including (a) a plaurality of vibration
measurements or (b) one or more vibration measurement and one or more
performance parameter measurements;
processor means for condition signature from said condition indicators and for
predicting a normal signature corresponding to the condition signature for a
healthy system;
comparator means for comparing the condition signature with the normal
signature; and
registration means for registering an event if the comparator indicates that the
condition signature differs from the normal signature by more than a
predetermined threshold.
A further aspect of the invention addressed the problem of the synchronous
acquisition of the condition indicators. The invention proposes to associate time
stamps (based on a common clock) with the acquired data and to synchronise
the data on the basis of these time stamps.
Accordingly, in this aspect, the invention provides a method of sychronising two
or more data streams, each data stream comprising a series of sequentially
acquired data elements (and relating e.g. to a respective condition indicator of
the previous aspect), the method comprising:
Associating a time stamp with each data element of each stream, the time stamp
identifying the time of acquisition of the data element on the basis of a clock
common to all data streams;
selecting a first data element from a first stream and inspecting its time stamp;
conducting a search of the data elements of the or each other stream to identify
the data element in the or each other element having an associated time stamp
closest to that of the selected element of the first stream; and
marking said identified data element of the or each other stream and said
selected element of eth first stream as being synchronized with one another.
Because, the relative acquisition times of the data elements are generally more
significant than their absolute acquisition times, the common clock may be
operate within an absolute or relative framework. In an absolute framework one
clock provides the time stamp for each data element of each data stream. In a
relative framework each data stream has its own clock, and one of the clock is
selected as the reference clock against which the acquisition times of the other
data streams are measured.
It may be convenient to use a mixture of absolute and relative frameworks. For
example, if the data streams relates to performance parameter and vibration
measurements, the performance parameters may be time stamped from one
clock and the vibration measurements from another clock.
The process can be repeated until the data elements in the first stream have
been exhausted, in any subsequent processing of the data that is reliant on using
synchronized data streams, only those data elements marked as being
synchronized with one another are used.
In the case where the acquisition rates of the data streams differ from one
another, it is preferred that the first stream, with which the other streams are
synchronized, is chosen to be the stream having the lowest acquisition rate.
The invention further provides a data processing system for synchronizing two or
more data streams, each data stream comprising a series of sequentially
acquired data elements, comprising:
means for associating a time stamp with each data element of each stream, the
time stamp identifying the time of acquisition of the data element on the basis of
a clock common to all data streams;
means for selecting a first data element from a first stream and inspecting its
time stamp;
means for conduction search of the data elements of the or each other stream to
identify the data element in the or each other element having an associated
tome stamp closest to that of the selected element of the first stream; and
means for making said identified data element of the or each other stream and
said selected element of the first stream as being synchronized with one another.
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS
The various aspect of the invention will be further described by way of example
with reference to the accompanying drawings, in which:
Fig. 1 schematically illustrates an exemplary data structure that can be adopted
for operation of the second aspect of the invention;
Fig. 2 shows a neural network architecture for a learnt model for operation of the
first aspect of the invention;
Fig. 3 shows a graph of the prediction error for the learnt model on a set of test
data corresponding to a period of normal operating conditions for an engine;
Fig. 4 shows a graph of the prediction error for the learnt model for a further
engine operating period in which the engine experienced a bird strike;
Fig. 5 shows the learning curve for a simple example of a system model for
operation for the first aspect of the invention;
Fig. 6 shows a comparison of observations and modeled estimates for a shaft
speed measurement, illustrating evolution of the model of Fig. 5.
Fig. 7 shows the measured low pressure shaft speed (N1V) for the period of the
test data from a more elaborate example of the system model,
Fig. 8 shows the value for the NIS over the same period as Fig. 7,
Fig. 9 shows the values for the 13 condition indicators and the NIS over the
same period from a further example of the system model, and
Fig. 10 shows a schematic example of an on-the-engine health monitoring
system.
DETAIL DESCRIPTION OF A PREFERRED EMBODIMENT OF
THE INVENTION
The embodiment described below is an example of a data processing system
employing both aspects of the invention discussed above. More specifically, it is
a system for synchronous acquisition, analysis and display of performance
parameters and vibration data from a power plant (e.g. a gas turbine), for
monitoring the health of the plant.
In accordance with a preferred from of the second aspect, the performance and
vibration data streams are synchronized in real time and, in accordance with a
preferred aspect of the first aspect of the invention, these data are combined or
fused to construct a signature for the system that can be compared to a
signature derived from a model representing a healthy power plant, in order to
provide anomaly/event detection and hence fault diagnosis.
The following discussion focuses on a application of the system to monitoring the
health of a gas turbine aero-engine, but it will be appreciated that the methods
can be adapted to other power plant, including for example ground-based and
marine gas turbine, and spark ignition and compression ignition internal
combustion engines, as well as other mechanical, thermodynamic, fluid, electrical
or electronic system. The system acquires performance parameters from the gas
turbine digitally via an Ethernet link at a rate between 20 and 40Hz. Typical
performance parameters are measurements of pressure, temperature, thrust,
altitude or March number. Vibration data is acquired from analogue vibration
transducers which are sampled at user-selectable sampling rates (from 625Hz to
80KHz) via an analogue-to-digital converter. The amplitude spectrum of the
vibration data is generated using the Fast Fourier Transform every 0.2 sec.
The performance and vibration data streams are asynchronous and stored in
separate files together with the corresponding timestamps. During review, as
data is loaded into memory, synchronization is performed between the
performance and spectrum data on a line by line basis.
Markers 10,12 (see Fig. 1) are kept which record the last synchronized line in the
vibration and performance data ring buffers 14,16. When new data is available in
memory, the time stamp for the next vibration spectrum line is examined. The
synchronization algorithm starts from the last previously synchronized location in
the performance data and searches forwards or backwards based on the
timestamps of the performance data (accurate to 0.05 sec) until the closest
matching timestamp in the performance data ring buffer 16 is identified. This
location in the performance data is recorded as being synchronized with the line
in the vibration ring buffer 14. The algorithm then proceeds to the next line in
the vibration ring buffer 14 (0.2 sec later) and so on until there is no more data
available to synchronise.
Clearly, therefore, if the performance parameters are acquired at 20 Hz (i.e. at
0.05 sec intervals) the synchronization precision is 0.075 sec (i.e. half the
acquisition interval added to the accuracy of the timestamps) and if the
performance parameters are acquired at 40 Hz (i.e. at 0.025 sec intervals) the
synchronization precision is 0.0625 sec.
Considering the synchronization algorithm in a little more detail, it can be seen
from Fig. i that the algorithm maintains a synchronization table 18 that gives the
index of the performance data entry that matches each vibration data line. The
algorithm uses variables to mark the latest synchronized data in each buffer. The
operation of the algorithm can be summarized by the following 'pseudo code':
1. Initialise the latest synchronized markers to the start of the vibration and
performance data.
2. Loop while there is more data in both ring buffers.
(a) Starting from the latest synchronized data item in each ring buffer,
examine the time stamp, t, on the next entry in the vibration ring buffer.
(b) search forward in the performance ring buffer until a time stamp
greater than t is found. Select between this entry in the performance ring
buffer and the previous entry for one which is closest to t and record the
match in the synchronization table.
Once synchronized, the analysis of this performance and vibration data relies on
constructing models of normal jet engine behaviour and then detecting an event
or an abnormality with respect to these models.
Traditional aircraft engine monitoring systems are based on two distinct
processes: the use of vibration signatures to indicate engine state, and a
separate procedure, gas-path analysis, which is employed for determination of
state from performance parameters. In the approach described now, however,
performance-retaled parameters such as pressure and temperature can be fused
with vibration data (such as tracked order vectors - the elements of which
correspond to the signal amplitude in a narrow range centered on the main
vibration frequencies for each shaft of the turbine). The aim is to take advantage
of disparate sources of data to from a more comprehensive picture of engine
state during normal operation. This in turn should allow a wider range of
deviations to be identified.
Furthermore, it is proposed to employ learnt data-driven models a normal
engine. Thus, although models of the engine system are used, these are not
fixed. Instead, they evolve with acquired training data. This offers the important
advantage of robustness.
The methods of data analysis described below may be termed 'novelty
detection'. An advantage of the methods is that the role of the expert need only
be retained in classifying training data as abnormal (i.e. novel) or normal. The
use of Kalman filtering systems in novelty detection has been described in e.g.
M. Gelb, Applied Optimal Estimation. MIT Press 1974.
Two alternative data analysis methods are described below. They are
distinguished by the amount of prior knowledge required to set up the system. In
both cases, the role of the expert need only be retained in classifying training
data as novel or normal.
The first method relies on a prior leant model of normality. For example, normal
vibration tracked order shaped are learnt using a simple clustering model for the
normal data. The novelty of e.g. the vibration signature for an engine under test
is assessed by comparing the closeness of its tracked order signature with the
prototypical patterns in the clustering model of normality. This can be done, for
example, by computing the shortest normalized Euclidean distance between the
vector encoding the tracked order shaped to any of the (prototypical patterns)
cluster centers in the model of normality (see Nairac at al, 'A System for the
Analysis of Jet Engine Vibration Data', Integrated Computer-aided Engineering,
6(l):53-65, 1999). If this distance is beyond a previously set threshold, the
vibration signature as represented by that tracked order is deemed to be outside
the bounds of normality. In addition to the vibration tracked orders, the model of
normality for the vibration spectra includes the following: sidebands, multiple
harmonics, fractional harmonics and broadband power.
The model is illustrated by an example in which a neural network having the
architecture shown in Fig. 2 was developed as the learnt model.
The neural network had an input layer 30 with four nodes consisting of four
condition indicator measured relating to one shaft of a multi-shaft test engine.
The condition indicators were the vibration amplitude, the phase and the shaft
speed all at a specified time, and the shaft speed a time increment after the
specified time.
The output layer 32 of the network had two nodes for predicting respectively the
change in vibration amplitude and change in phase after the time increment.
The network had one hidden layer 34, each node of which contained a Gaussian
radial basis function.
The training phase for network used training data obtained from the test engine
over a range of normal operating conditions. The centers and the spreads of the
Gaussians were fixed using the cluster analysis described above and the weights
of the connections between the nodes were then iterativeiy adjusted until the
model converged.
Figure 3 shows a graph of the prediction error (that is the sum of the prediction
errors of the change in vibration amplitude and change in phase) for the model
on a set of test data which also corresponded to the period of normal operating
conditions for the engine. This graph provides a baseline of prediction error
variation against which novel events can be judged.
The neural network had an input layer 30 with four nodes for a condition
signature consisting of four condition indicator measured relating to one shaft of
a multi-shaft test engine. The condition indicators were the vibration amplitude,
the phase and the shaft speed all at a specified time, and the shaft speed a time
increment after the specified time.
The output layer 32 of the network had two nodes for predicting respectively the
change in vibration amplitude and change in phase after the time increment.
The network had one hidden layer 34, each node of which contained a Gaussian
radial basis function.
The training phase for network used training data obtained from the test engine
over a range of normal operating conditions. The centers and the spreads of the
Gaussians were fixed using the cluster analysis described above and the weights
of the connections between the nodes were then iteratively adjusted until the
model coveraged.
Fig. 3 shows a graph of the prediction error (i.e. the sum of the prediction errors
of the change in vibration amplitude and change in phase) for the model on a set
of test data which also corresponded to the period of normal operating
conditions for the engine. This graph provides a baseline of prediction error
variation against which novel events can be judged.
Fig. 4 shows a graph of the prediction error for a further engine operating
period. In this case, however, the engine experienced a bird strike. The largest
peak in the graph corresponds to the moment of bird impact. Clearly the model
was able to recognize this event. However, the changes prediction error signal
(compared to the baseline of Fig. 3) after the event showed that the model was
also able to detect post-impact abnormal engine behaviour. This provides
confidence that the model can not only detect major events such as bird strikes,
but also more subtle deviations from normality.
The second method employs a process model which has a state vector
associated with it (see below). The observation vector (i.e. the condition
signature) has elements corresponding to measured values of performance
parameters and vibration information so that two types of data are fused within
the model. The fusion of the data is performed in real-time with a new output
being generated by the system several times a second.
An important aspect of the use of this model in the system is the use of learning.
In a first, off-line, phase of learning, a generic model of the engine is learnt. The
learning is data-driven using an algorithm such as Expectation-Maximisation in
order to maximize the likelihood of the learnt model given the training data.
Once such a generic model has been learnt off-line for a particular type of
engine, learning can then be applied on-line in order to tune the model to an
individual engine immediately after its pass-off test and after each maintenance
procedure. Engine deterioration can also be learnt on-line. The learnt model can
be turned to different flight conditions, such as cruising or landing, in order to
detect novelty with even more sensitivity and specificity.
The data-driven learnt model may be integrated with existing performance
models which rely on the laws of thermodynamics and computational fluid
dynamics (knowledge-based models). Such models can therefore be described as
hybrid models because they are based on the integration of learnt and
knowledge-based models.
Looking in more detail at the learnt modeling approach, it is based on the
application of Expectation Maximisation (EM) to parameter estimation in linear
dynamical systems (see Ghahramani and Hinton, parameter Estimation for Linear
Dynamical Systems, technical Report CRG-TR-96-2, University of Tornoto, 1996)
and to non-linear systems (see Rowels and Ghahramani, 'A Unifying Review of
Linear Gaussian Models', Neural Computation, 11, 305-345, 1999; and
Ghahramani and Roweis, 'Learning in Nonlinear Dynamical System Using an EM
Algorithm' in Kearns et al. (editors), advances in Neural Information Processing
System, Volume 11, MIT Press, 1999).
The EM learning algorithm is applied to a Kalman filter model. In the linear case,
this is a system with a measurement process of the form

Where y(i) is a set of observations of hidden state x(i), C is a covariance matrix,
and measurement noise v(i) is zero-mean and normally distributed with
covariance matrix R. y(i) and x(i) can be the same dimension. Non-zero off-
diagonal terms in C allow the model to account for inter-dependencies between
the performance parameter and vibration measurements of the condition and
normal signatures. The state equation is

with w(i) zero-mean and normally distributed with covariance matrix Q.
At the beginning of the training phase A and C are initialized to small random
values (e.g. with elements of the matrices ~ 10-5), and R and Q are initialized
e.g. to I. Then during the training phase, for each condition signature y(i) in the
training set, the method of rowels and Ghahramani applied to equation (2) to
derive the most likely values for the elements of x(i), and the elements of C, R
and v(i) are iteratively adjusted so that Cx(i) +v(i) converges to the respective
condition signature (R and Q can be constrained throughout to be diagonal
matrices). Convergence can be determined by the log-likelihood of the set of
observations given the model.
Instead of initializing the elements of A to small random values, it is also possible
to adopt initial values that embody existing performance models of engine
behaviour. After the training phase, the model would then be a hybrid of a
knowledge-based and a data-driven model. By fusing these two methods of
data-analysis, the accuracy of prior expert knowledge can be combined with the
robustness of data-driven approaches.
When the training phase has ended and the model is receiving real-time data
consisting of a sequence of condition signature, the Kalman filter is again used to
derive the most likely values for the elements of x(i) for each condition signature
y(i). However, the elements of C and v(i) are now fixed, so Cx(i) +v(i) provides
the normal signature for comparison with the condition signature.,
For example, comparison of the normal signature with the condition signature
can be on the basis of the normalized innovations squared (NIS). The
innovations sequence v is the difference between the condition signature and the
normal signature, so

The innovations should be zero-mean and white.
The NIS combines the individual innovations sequences.

The individual sequence are weighted by the term S(k)1, the inverse of the
innovation covariance given by

where P(kl k-1) is the prediction covariance.
The model is first illustrated by a simple example (which does not use vibration
measurements) where observations are made of the speeds of the three shafts
of a test engine during cruise. The observed data y is simply the state x
corrupted by noise, so

The observations are used during the learning process, to generate a dynamical
system model in which A, C, Q and R are learned from the data. At the beginning
of the training phase A and C were initialized to small random values and R and
Q were initialized to I.
Fig. 5 illustrates the learning (log likehood) plot for the system. Fig. 6 shows the
evolution of estimates of shaft 1 speeds during the learning process using the EM
algorithm. In the example shown, the learning stage lasts for the first 25
iterations. From iteration 25 onwards, the system's dynamical properties are
determined by the learned matrices (which are then kept fixed).
Once trained, the systems can be used to detect events or abnormalities, that is
to say divergences from the learnt model of normality. The events of particular
interest are those that are unexpected, possibly indicating a problem with the
engine for example. However, particularly where the models have been learned
only for 'steady state' parts of the flight envelope (e.g. acceleration, cruise and
deceleration), transients during operation of the engine will also be flagged up as
events, although they are expected. For example, where a bleed valve is opened
or closed, the operating condition of the engine will exhibit significant differences
from a learnt model of steady state normality which does not include this event.
Thus when using such a steady state model, measures can be employed to avoid
these transient events. For instance, since the opening of a bleed valve is an
event that occurs at a defined point in time, the data collected from the engine
at the time and slightly either side of it (e.g. for 2 seconds before and after) can
be eliminated from the data analysed by the health monitoring system.
The approach is next illustrated with a more elaborate example which uses both
performance parameters and vibration data. In this example the model was
applied to data acquired from a test bed-mounted, multi-shaft, aero gas turbine
engine which was undergoing a cyclic acceleration-deceleration test when it
experienced an intermediate pressure turbine lock-plate event. Such tests are
used to investigate engine behaviour under extreme conditions. The data was
acquired synchronously as described above.
A lock-plate event occurred in the engine on test day 152-00. The training data
was 152-00 data for a period before the event, and the test data was 152-00
data for the period including the event. In this example a 14 dimensional model
(i.e. y(i) and x(i) each had 14 elements) was used in which the condition
indicator inputs were:
- The tracked orders of the low pressure (tol), intermediate pressure (toi)
and high pressure (toh) shafts,
- The shaft speeds of the low pressure (NIV), intermediate pressure (N2V)
and high pressure (N3V) shafts,
- The ambient (POV), total inlet (P20V), high pressure compressor delivery
(P30V) and exhaust (PEXV) pressures,
- The total inlet (T20V) and high pressure compressor delivery (T30V)
temperatures,
- The turbine gas temperature trimmed (TGTTRM)
- The demanded fuel flow (WFDEM)
At the beginning of the training phase A and C were initialized to small random
values and R and Q were initialized to I.
Fig. 7 shows the measured low pressure shaft speed (NIV) for part of the period
of the test data, and Fig. 8 shows the value for the NIS calculated by the trained
model over the same period.
The first two sharp troughs in the NIV trace were caused by planned consecutive
cyclic decelerations. Associated with each of these troughs are two NIS peaks.
These peaks indicated that the engine was not behaving normally during the
cyclic testing. In fact subsequent examination revealed that a lock plate had
released earlier during the test and as a result abnormal blade rubbing was
occurring during each of the deceleration cycles.
During the third planned deceleration cycle in the period covered by Figs. 7 and
8 (i.e. at approximately data point 12850) the engine sustained blade damage
which resulted in the sharp NIS peak and drop off in NIV.
However, the earlier (all be it smaller) NIS peaks demonstrate that the
monitoring system was able detect the effect of the lock plate release in real
time and before substantial blade damage was sustained. If such a release had
occurred in an in-service aero engine, it would therefore have been possible to
generate an immediate warning so that timely action (such as engine inspection
or maintenance) could have been performed. In contrast, sudden variations in
NIV can occur normally, so Nlv alone is not a reliable indicator of abnormal
behaviour.
A further example also uses both performance parameter and vibration data.
Again the model was applied to synchronously data acquired from a test bed-
mounted, multi-shaft, aero gas turbine engine. However, in this case an oil seal
leak developed in the engine.
The fault occurred in the engine around data point 50410. The training data was
from a period before the fault, and the test data was for a period including the
fault. In this example a 13 dimensional model was used in which the condition
indicator inputs were tol, toi, toh, N1V, N2V, N3V, P20V, P30V, PEXV, T20V,
TGTTRM, AND WFDEM.
Fig. 9 shows the values for the 13 condition indicators and the NIS (in the
bottom graph) over the period including the event. The sharp NIS peak at data
point 50410 again demonstrates that the monitoring system was able to detect
the moment of the event. Although some of the other condition indicators also
had peaks at this time, by themselves they cannot be reliably associated with
abnormal (novel) behaviour.
Subsequent examination showed that the event was an oil seal which allowed
engine oil to enter the combustor. Because of this leak the engine control system
reduced the amount of fuel entering the combustor (hence the drop in WEDEM)
to maintain the engine thrust (given by PEXV / P20) at a constant level.
After the event, with oil continuing to leak into the combustor, the engine was
operating outside the scope of normal behaviour. This was successfully picked up
by the NIS trace which after data point 50410 did not return to its pre-event
level.
Next we consider how a health monitoring system, incorporating the model
described above, could be installed for in-flight analysis of aero gas turbines.
An on-the engine system, shown schematically in Fig. 10, could generate of the
order of 1 Gb of vibration and performance data (consisting mainly of pressures,
temperatures and shaft speeds) per flight. The vibration data is usually analysed
in the frequency domain. The vibration and performance data, as there are being
generated by data acquisition means 20, are temporarily stored in ring buffer 22.
The data is synchronized and subjected to novelty detection in processor and
comparator means 24 which receives a synchronisation signal from data
acquisition means 20 and the data from ring buffer 22. Those sections of the
data corresponding to novel events are then tagged and recorded with no loss of
information (i.e. highbandwidth data is recorded) in registration means 26 which
has semi-permanent on-line and/or hard disk storage. When the flight is
completed the stored data may be downloaded and subjected to more intensive
ground-based analysis.
The system may also include a display which is driven to allow information to be
displayed either during acquisition or for review once an acquisition cycle has
been completed, it preferably includes the following features:
Ability to display the result of the comparison of the condition signature
with the normal signature, e.g. in the form of the NIS or the prediction
error. An unhealthy event nay be highlighted e.g. with an alert signal.
Ability is display a combination of any two of vibration spectra, tracked
orders, broadband power, performance parameters synchronized in time.
Ability to extract and plot vibration spectra against engine speed.
Ability to interrogate and print any of vibration spectra, tracked orders,
broadband power and performance parameters.
Automatic detection and display of features from vibration spectra
(sidebands, harmonics, etc.)
WE CLAIM
1. A method for monitoring the health of a plant, which comprises
performing at each of a plurality of times the steps of:
constructing a condition signature for a present time from a plurality of
condition indicators having (a) a plurality of vibration measurements
acquired from said system or (b) one or more vibration measurements
and one or more performance parameter measurements acquired from
said plant;
predicting a normal signature from a model defining one or more inter-
dependencies between condition indicators used to construct the condition
signature for a previous time, said normal signature corresponding to a
condition signature for a healthy plant at the present time, and said model
comprising a matrix with one or more non-zero off-diagonal terms to
define said inter-dependencies;
comparing said condition signature for the present time with said normal
signature; and
registering an event if said condition signature for the present time differs
from said normal signature by more than a predetermined threshold.
2. A method as claimed in claim 1, wherein said model is a learned model.
3. A method as claimed in claim 1, wherein the step of comparing said
condition signature with said normal signature involves calculating a value
for the normalised innovations squared.
4. A method as claimed in claim 1, wherein said model comprises a neural
network.
5. A method as claimed in claim 4, wherein the step of comparing said
condition signature with said normal signature involves calculating a
prediction error.
6. A method as claimed in claim 1, wherein said times define successive
intervals of at most 1 sec duration.
7. A method for monitoring the health of a plant, which comprises
performing at each of a plurality of times defining successive intervals of
at most 1 sec duration the steps of: constructing a condition signature for
a present time from a plurality of condition indicators including (a) a
plurality of vibration measurements acquired from the system or (b) one
or more vibration measurements and one or more performance parameter
measurements acquired from said system, said measurements being
synchronously acquired from said plant to a synchronization imprecision of
at most 1 sec;
predicting, from condition indicators used to construct the condition
signature for a previous time, a normal signature corresponding to a
condition signature for a healthy system at the present time;
comparing said condition signature for the present time with said normal
signature; and
registering an event if said condition signature for the present time differs
from said normal signature by more than a predetermined threshold.
8. A method as claimed in claim 7, wherein said normal signature is
predicted from a model defining one or more inter-dependencies between
said condition indicators used to construct the condition signature for the
previous time.
9. A method as claimed in claim 8, wherein said model is a learned model.
10. A method as claimed in claim 8, wherein said model comprises a matrix
with one or more non-zero off-diagonal terms to define said
interdependences.
11. A method as claimed in claim 10, wherein the step of comparing said
condition signature with said normal signature involves calculating a value
for the normalised innovations squared.
12. A method as claimed in claim 8, wherein said model comprises a neural
network.
13. A method as claimed in claim 12, wherein the step of comparing said
condition signature with said normal signature involves calculating a
prediction error.
14. A method as claimed in any one of claim 1, wherein said measurements
are synchronously acquired from said system to a synchronisation
imprecision of at most 1 sec.
15. A method as claimed in any one of claims 1, wherein said system
comprises a gas turbine engine.
16. A data processing system for monitoring the health of a system,
comprising:
data acquisition means for acquiring a plurality of condition
indicators from said system at each of a plurality of times, said condition
indicators including (a) a plurality of vibration measurements or (b) one or
more vibration measurements and one or more performance parameter
measurements;
processor means for constructing a condition signature for a
present time from said condition indicators and for predicting a normal
signature corresponding to a condition signature for a healthy system at
the present time, said normal signature being predicted by a model which
defines one or more inter-dependencies between condition indicators used
to construct the condition signature for a previous time, wherein said
mode! comprises a matrix with one or more non-zero off-diagonal terms
to define said inter-dependencies;
comparator means for comparing said condition signature for the
present time with said normal signature; and
registration means for registering an event if said comparator for
the present time indicates that said condition signature differs from said
normal signature by more than a predetermined threshold.
17. A data processing system for monitoring the health of a system,
comprising:
data acquisition means for acquiring a plurality of condition
indicators from said system at each of a plurality of times defining
successive intervals of at most 1 sec duration, said measurements being
synchronously acquired from said system to a synchronization imprecision
of at most 1 sec, and said condition indicators including (a) a plurality of
vibration measurements or (b) one or more vibration measurements and
one or more performance parameter measurements;
processor means for constructing a condition signature for a
present time from said condition indicators and for predicting, from
condition indicators used to construct the condition signature for a
previous time, a normal signature corresponding to a condition signature
for a healthy system at the present time;
comparator means for comparing said condition signature for the
present time with said normal signature; and
registration means for registering an event if said comparator for
the present time indicates that said condition signature differs from said
normal signature by more than a predetermined threshold.
18.A method for monitoring the health of a system, which comprises
performing at each of a plurality of times the steps of;
constructing a condition signature for the present time from a
plurality of condition indicators including (a) a plurality of vibration
measurements acquired from said system or (b) one or more vibration
measurements and one or more performance parameter measurements
acquired from said system, said measurements being synchronously
acquired from said system to be a synchronization imprecision of at most
i sec;
predicting a normal signature from a model defining one or more
inter-dependencies between condition indicators used to construct the
condition signature for the previous time, said normal signature
corresponding to a condition signature for a healthy system at the present
time;
comparing said condition signature for the present time with said
normal signature; and
registering an event if said condition signature for the present time
differs from said normal signature by more than a predetermined
threshold.
19. A data processing system for monitoring the health of a system,
comprising:
data acquisition means for synchronously acquiring a plurality of
condition indicators from said system at each of a plurality of times to a
synchronization imprecision of at most 1 sec, said condition indicators
including (a) a plurality of vibration measurements or (b) one or more
vibration measurements and one or more performance parameter
measurements;
processor means for constructing a condition signature for the
present time from said condition indicators and for predicting a normal
signature corresponding to a condition signature for a healthy system at
the present time, said normal signature being predicted by a model which
defines one or more inter-dependencies between condition indicators used
to construct the condition signature for the present time;
comparator means for comparing said condition signature for the
present time with said normal signature; and
registration means for registering an event if said comparator for
the present time indicates that said condition signature differs from said
normal signature by more than a predetermined threshold.
This invention relates to a method for monitoring the health of a system
comprises performing at each of a plurality of times the steps of: constructing a
condition signature from a plurality of condition indicators including (a) a plurality
of vibration measurements acquired from the system or (b) one or more
vibration measurements and one or more performance parameter measurements
acquired from the system; predicting a normal signature from a model defining
one or more inter-dependencies between said condition indicators, the normal
signature corresponding to the condition signature for a healthy system;
comparing the condition signature with the normal signature; and registering an
event if the condition signature differs from the normal signature by more than a
predetermined threshold.

Documents:

IN-PCT-2002-(11-01-2012)-1540-KOL-CORRESPONDENCE.pdf

IN-PCT-2002-(11-01-2012)-1540-KOL-OTHERS.pdf

IN-PCT-2002-1540-KOL-(11-01-2012-)-ASSIGNMENT.pdf

IN-PCT-2002-1540-KOL-(11-01-2012-)-CORRESPONDENCE.pdf

IN-PCT-2002-1540-KOL-(11-01-2012-)-PA-CERTIFIED COPIES.pdf

IN-PCT-2002-1540-KOL-(15-10-2012)-CORRESPONDENCE.pdf

IN-PCT-2002-1540-KOL-(29-11-2012)-ASSIGNMENT.pdf

IN-PCT-2002-1540-KOL-(29-11-2012)-CORRESPONDENCE-1.pdf

IN-PCT-2002-1540-KOL-(29-11-2012)-CORRESPONDENCE.pdf

IN-PCT-2002-1540-KOL-(29-11-2012)-OTHERS.pdf

IN-PCT-2002-1540-KOL-(29-11-2012)-PA.pdf

IN-PCT-2002-1540-KOL-ASSIGNMENT 1.1.pdf

IN-PCT-2002-1540-KOL-ASSIGNMENT.pdf

IN-PCT-2002-1540-KOL-CORRESPONDENCE 1.2.pdf

IN-PCT-2002-1540-KOL-CORRESPONDENCE-1.1.pdf

IN-PCT-2002-1540-KOL-CORRESPONDENCE.pdf

IN-PCT-2002-1540-KOL-FORM 16.pdf

IN-PCT-2002-1540-KOL-FORM-27.pdf

in-pct-2002-1540-kol-granted-abstract.pdf

in-pct-2002-1540-kol-granted-assignment.pdf

in-pct-2002-1540-kol-granted-claims.pdf

in-pct-2002-1540-kol-granted-correspondence.pdf

in-pct-2002-1540-kol-granted-description (complete).pdf

in-pct-2002-1540-kol-granted-drawings.pdf

in-pct-2002-1540-kol-granted-examination report.pdf

in-pct-2002-1540-kol-granted-form 1.pdf

in-pct-2002-1540-kol-granted-form 18.pdf

in-pct-2002-1540-kol-granted-form 2.pdf

in-pct-2002-1540-kol-granted-form 26.pdf

in-pct-2002-1540-kol-granted-form 3.pdf

in-pct-2002-1540-kol-granted-form 5.pdf

in-pct-2002-1540-kol-granted-form 6.pdf

in-pct-2002-1540-kol-granted-reply to examination report.pdf

in-pct-2002-1540-kol-granted-specification.pdf

in-pct-2002-1540-kol-granted-translated copy of priority document.pdf

IN-PCT-2002-1540-KOL-PA 1.1.pdf

IN-PCT-2002-1540-KOL-PA.pdf

« Previous Patent

Next Patent »

Patent Number

225479

Indian Patent Application Number

IN/PCT/2002/1540/KOL

PG Journal Number

46/2008

Publication Date

14-Nov-2008

Grant Date

12-Nov-2008

Date of Filing

17-Dec-2002

Name of Patentee

OXFORD BIOSIGNALS LIMITED ,

Applicant Address

MAGDALEN CENTRE, OXFORD SCIENCE PARK, OXFORD OX4 4GA

Inventors:

#	Inventor's Name	Inventor's Address
1	UTETE SIMUKAL	C/O. ST. HUGH'S COLLEGE, OXFORD, OXFORDSHIRE OX2 6LE
2	ANUZIS PAUL	86 LOCKO ROAD, SPONDON, DERBY, DERBYSHIRE DE21 7AR
3	KING STEVE P	30 WILLOW DRIVE, NEWHALL, SWADLINCOTE, DERBYSHIRE DELL ONW
4	KING DENNIS M	94 WESTERN ROAD, MICKLEOVER, DERBY, DERBYSHIRE DE3 5GQ
5	TARASSENKO LIONEL	68 OLD ROAD, HEADINGTON, OXFORD, OXFORDSHIER OX3 7LP
6	HAYTON PAUL	1 LANGFORD COTTAGE, LONDON ROAD, BISESTER, OXFORDSHIRE OX6 OJN

PCT International Classification Number

G01H 1/00

PCT International Application Number

PCT/GB01/03020

PCT International Filing date

2001-07-05

PCT Conventions:

#	PCT Application Number	Date of Convention	Priority Country
1	0016561.3	2000-07-05	U.K.