WO2005071582A2

WO2005071582A2 - Detection of abnormal behaviour in dynamic systems

Info

Publication number: WO2005071582A2
Application number: PCT/GB2005/000129
Authority: WO
Inventors: Basim Majeed; Behnam Azvine
Original assignee: British Telecommunications Public Limited Company
Priority date: 2004-01-27
Filing date: 2005-01-14
Publication date: 2005-08-04
Also published as: WO2005071582A3

Abstract

A system and method for predicting behavioural changes in dynamic systems are disclosed, the system comprising signal providing means (30) arranged to apply a configurable test signal to a dynamic system (31); monitoring means (32) arranged to monitor responses of the dynamic system to the application of said test signal, and to provide response data indicative of said responses; and response processing means (33) arranged to detect changes in said response data and provide predictions therefrom relating to whether the dynamic system is undergoing behavioural changes. Preferred embodiments further comprise means (36) for reconfiguring the test signal in response to an assessment of the response data obtained from the application of an initial test signal. This allows for the analysis procedure to be repeated with a test signal modified in order to provide improved response data.

Description

Detection of Abnormal Behaviour in Dynamic Systems

Technical Field

The present invention relates to methods and systems for monitoring the behaviour of dynamic systems, and in particular to methods and systems for detecting indications that a dynamic system is starting to behave differently, and for providing predictions relating to behavioural changes of dynamic systems.

Background to the Invention and Prior Art

Fault Detection and Isolation (FDI) Fault Detection and Isolation (FDI) systems aim to detect abnormal deviations in the behaviour of a system from the expected behaviour under the prevailing operating conditions. Such systems often aim to follow this by generating a list of fault candidates, which can be further refined by exploring further measurements and/or further analysis of the measured data.

Research over the past decade has concentrated mainly on methodologies and algorithms for FDI that rely on the availability of an accurate analytical model of the system, which can be in the form of a transfer function or state space representation. Major successes of FDI implementation have mainly been in applications where the system under observation exhibits a time-independent, linear dynamic behaviour over the operating region of interest. In such cases, linear state observers can track the expected response of the system to set-point and disturbance inputs, and by comparing this response with the actual one, residuals are generated that can lead to the identification of the source of the fault. See, for example, the article: "Frequency-Domain Optimization for Robust Fault Detection and Isolation in Dynamic Systems" (D. Sauter and F. Hamelin, IEEE Transactions on Automatic Control, vol. 44, No. 4, pp. 878-882, April 1999).

The main drawback of such FDI implementations is that an accurate model of the system is rarely available. Even in implementations where algorithms based on non-linear models are used, the accuracy of the fault detection results does not instil confidence in the system, and has thus prevented successful commercial deployment. See, for example, the article: "Model Based Sensor and Actuator Fault Detection and Isolation" (E. Larson, B. Parker Jr. and B. Clark, Proceedings of the American Control Conference, pp. 4215- 4219, May 2002). To compensate for these limitations, Artificial Intelligence (Al) based FDI techniques have been on the research scene for a few years.

Frequency Signature as a Tool for FDI

The frequency signature of a system is an important tool for determining whether the system is operating normally or under faulty conditions. In some cases, a continuous change of the frequency spectrum can provide a strong indication that a fault is impending, which should help in the scheduling of maintenance in order to avoid a catastrophic breakdown. Researchers have applied a number of techniques to analyse the frequency signature in order to detect faulty conditions. See for example, the article: "Detection of Faulty Components on Power Lines Using Radio Frequency Signatures and Signal Processing Techniques" (S. Shihab and K. Wong, Proceedings 2000 IEEE Power Engineering Society Winter Meeting, pp 2449-52 vol.4, 23-27 Jan. 2000) in which statistical classification of the frequency signature is used as fault indicator.

Despite the progress in the field of automatic diagnosis tools, the ultimate knowledge and interpretation of the frequency spectrum to determine the normality status of the system still lies with the human expert operator. This is because this type of knowledge is qualitative in nature and cannot be implemented using analytical or numerical techniques due to the fundamental lack of precision in the data.

For example, Figure 1 shows a system with two frequency modes being excited by a white noise source. In addition to the normal frequency signature of the system, the graph shows two frequency signatures for the system when it is affected by a non-linear sensor behaviour (saturation in this case).

An expert human operator may be capable of recognising that the system is affected by a non-linear behaviour because he/she can detect that the fundamental components of the signal are reducing in power, while at the same time there is a relative increase in the power of the harmonic components. Other problems with the operation of the system can be detected from the frequency spectrum including attenuation (loss of signal), damping, amplification, loss of damping, noise interference, change in the system modes, complete breakdowns and so on. Figure 2 shows the case where a system with two natural frequencies is affected by a fault that results in a change of these frequencies.

Referring to prior patent documents, United States patent US 6,338,029 (Abbata et al) relates to a method for determining when electric motors in devices such as photocopying machines are acceptable. In order to do this, various tests are performed relating to specific characteristics of a motor, such as using a vibration sensor to measure the vibration energy of the motor in a steady state and thereby forming a vibration energy parameter. A decision as to whether or not the motor is acceptable can then be taken by considering the parameters, and it is stated that this can be done using fuzzy logic. Similarly, US 6,637,267 (Fiebelkom et al) relates to a diagnostic system for valves in , which structure-borne noise in valves is sensed and an evaluation unit evaluates a recorded measurement signal. Again, it is stated that fuzzy technology and self-learning diagnostic strategies based on neural structures may be used to evaluate the results, in order to achieve more automatic diagnosis, but this document only concerns itself with the treatment of data in the form of the measurement signal received from the item under test.

In the field of testing underground electric cables, United States patent US 5,347,212 discloses a method in which a non-harmonic selected frequency signal from a variable frequency generator is applied to the neutral of an underground cable connecting a pair of structures such as power transformers. The test is conducted by a surveyor wearing foot electrodes which contact the earth as the surveyor walks along the earth over the cable, and at each step, potentials are monitored, allowing a continuous corrosion and condition profile of the entire length of the cable to be generated.

I nternational patent application WO 99/40453 relates to occupancy sensors and motion detectors, and in particular to methods and apparatus for controlling an electrical load based upon occupancy and/or motion within a monitored zone. According to this publication, an interrogation signal is transmitted at a pre-determined frequency into a zone to be monitored. The interrogation signal may be transmitted either continuously or at controllable sample intervals. The return signal is then processed, either as an analog signal or digitized, to provide an indication of motion within a space. Thereafter, at a decision stage, an electrical load is either energized, de-energized, or otherwise controlled as a function of the indication of motion within the space. It is suggested that the "occupancy signal" may be generated either actively or passively, where examples of active generation are said to include use of ultrasonic, optical, acoustic, and microwave signals, and examples of passive generation are said to include use of passive infrared detectors that merely detect moving heat sources within a monitored zone.

From the above, it will be appreciated that there is great potential for an FDI system that can perform the role for which it has previously been necessary to have an expert human operator, and automatically interpret the shape of frequency signatures and relate them to fault conditions, particularly if such a system can perform this role with greater reliability than a human. In addition, the human expert does not have the ability or analysing power necessary to allow him/her to select the shape and/or frequency range of a test signal in order to optimise the prediction of abnormal conditions.

Summary of the Invention

In view of the above, there is provided, according to a first aspect of the present invention, a system for predicting behavioural changes in dynamic systems, comprising: signal providing means arranged to apply a reconfigurable test signal to a dynamic system; monitoring means arranged to monitor responses of the dynamic system to the application of said test signal, and to provide response data indicative of said responses; response processing means arranged to detect changes in said response data and provide predictions therefrom relating to whether the dynamic system is undergoing behavioural changes; and signal reconfiguring means arranged to interact with the signal providing means whereby to reconfigure the test signal.

According to the present invention, there is also provided a method of predicting behavioural changes in dynamic systems, comprising steps of: applying a reconfigurable test signal to a dynamic system; monitoring responses of the dynamic system to the application of said test signal, and providing response data indicative of said responses; detecting changes in said response data and providing predictions therefrom relating to whether the dynamic system is undergoing behavioural changes; reconfiguring the test signal; and applying the reconfigured test signal to the dynamic system. According to embodiments of the present invention there can thus be provided methods and systems for monitoring the behaviour of dynamic systems that can interpret the response of such systems under test to the application of specifically chosen or configured test signals and derive predictions therefrom that the system is, or will be, undergoing abnormal, perhaps faulty behaviour.

Depending on the likely natural and resonant frequencies of the dynamic system under test, the test signal may be a mechanical or an acoustic signal, in which case mechanical sensors such as accelerometers or vibration sensors may be appropriate. Alternatively the test signal may be an electrical signal or another type of signal, in which case capacitative, resistance, or other types of sensors may be appropriate.

A specific application of a system according to the invention is in the testing of fibre optic cables by vibration. Detection of abnormal patterns may be taken as evidence of such problems as hot cables or partially damaged cables underground. Reconfiguration of the test signal can lead to a better choice of vibration amplitude and frequency in order to determine more accurately, more sensitively or in more detail whether cable operates or runs correctly between two physical points in a network. Such a system could improve the early detection and thus prevention of otherwise unforeseen circumstances such as underground fire incidents.

In general, the signal providing means may be arranged to apply an initial test signal comprising components such as sine wave, white noise, pulse train or step input components, depending on the situation. As will be explained below, systems according to preferred embodiments of the invention are capable of using the response of the system under test to the initial signal to determine how the test signal could be reconfigured in order to allow for an improved testing procedure.

In general, irrespective of the type of test signal, the monitoring means may be arranged to provide the response data in the form of a power spectrum, by use of algorithms such as Fast Fourier Transform (FFT) algorithms or other signal processing methods. The response processing means may then comprise one or more of the following spectrum characteristic detection means: a fundamental component magnitude detection means for detecting changes in the magnitudes of fundamental components of the power spectrum; a fundamental component frequency detection means for detecting changes in the frequencies of fundamental components of the power spectrum; a harmonic components detection means for detecting the existence of harmonics of fundamental components in the power spectrum; and a spurious components detection means for detecting the existence of spurious frequency components which do not correspond to harmonics of the fundamental components in the power spectrum. In relation to each characteristic there may be fuzzy classification means for classifying the detected characteristic using fuzzy sets.

According to preferred embodiments of the invention, the response processing means may comprise predicting means for predicting whether the dynamic system is undergoing behavioural changes using fuzzy rules, and output means such as a display device for providing output to a user in the event that a prediction is made that the dynamic system is undergoing behavioural changes. The response processing means may further comprise means for determining which of said fuzzy rules led to such a prediction, and provide an indication to a user relating to this.

With reference to the above, it will be understood that a human expert monitoring a plot of a frequency response of a system may attempt to use visual clues from an observed frequency spectrum, but will be incapable of identifying individual characteristics of the spectrum and making a qualitative assessment based thereon. A human expert is incapable of carrying out analysis of the type facilitated by FFT algorithms, for example. Embodiments of the present invention allow for analysis based on individual characteristics such as the magnitudes or frequencies of fundamental components, the harmonic noise components and the spurious noise components, in order to reach a conclusion regarding the "health" or behavioural status of the monitored system. The system may thus embody the fault detection knowledge of an expert in a fuzzy knowledge rule base that consists of rules relating to one or more characteristics such as these. From analysis of characteristics such as the magnitudes and frequencies of fundamental components, the levels of harmonic distortion and the levels of spurious noise, an indication may be given regarding the type of fault that is believed to be affecting the system under test.

According to embodiments of the invention in which fuzzy reasoning is applied, logic and prediction techniques more closely approximating those of an expert operator may be used in order to interpret the response of the system under test, and this may be used in determining likely types of system faults.

By determining the fuzzy rule or rules responsible for a prediction that the system is behaving abnormally, it may be possible to ascertain the characteristic or characteristics to which those fuzzy rules relate, and this information may allow a diagnosis of the type of fault to be output to the user.

As mentioned above, by virtue of the signal reconfiguring means arranged to interact with the signal providing means whereby to reconfigure the test signal, preferred embodiments of the invention are capable of using the response of the system under test to an initial test signal, or to one or more previously applied test signals, in order to determine how the test signal could be reconfigured in order to allow for an improved testing procedure. The test signal reconfiguration may be done manually (i.e. under the control of a user). Alternatively signal reconfiguring may be done automatically (i.e. without the need for instructions from a user) or semi-automatically (i.e. partly under the control of a user). Such embodiments may include prediction assessment means for determining a measure of confidence relating to any prediction of abnormal behaviour which arises. Such measures of confidence may be determined using fuzzy logic, for example. With such embodiments, it may be arranged that predictions of abnormal behaviour are only provided to the user in the event that the measure of confidence in the prediction is found to be sufficiently high. If predictions cannot be made with a sufficiently high measure of confidence, such embodiments may be arranged to reconfigure the test signal automatically.

If the test signal is reconfigured, whether automatically according to predetermined rules, manually by a user to whom output has been provided, or semi-automatically (i.e. according to a combination of instructions from the user and from the system), an attempt may then be made to adjust the characteristics of the test signal (e.g. level and/or bandwidth). This may be in order to "home in" on a region of a frequency spectrum that appears to be providing an indication that a behavioural change is occurring, but where greater resolution is required to confirm this. Alternatively, this may be in order to completely change the form of test signal being used if the initial signal is found not to cause a potentially useful response. The FDI task may then be repeated and a new degree of confidence calculated. However the test signal is changed, the test process can then be repeated until a behavioural prediction or fault decision can be obtained with a sufficiently high degree of confidence.

By virtue of the above, systems according to preferred embodiments of the invention are capable of intelligently modifying characteristics such as the shape and/or frequency range of test signals in order that they best suit the situation or the specific diagnosis goals, by automatically selecting/or homing in on, suitable test signals that would achieve the required FDI goal.

Brief Description of the Drawings Further features and advantages of the present invention will become apparent from the following description of embodiments thereof, presented by way of example only, and by reference to the accompanying drawings, wherein like reference numerals refer to like parts, and wherein: Figure 1 is a graph illustrating the response spectrum of a dynamic system having two frequency modes, being excited by a white noise source; Figure 2 is graph illustrating changes in the response spectrum of a dynamic system when the system is behaviour of the system is affected by a fault; Figure 3 is a block diagram of a system according to an embodiment of the present invention; Figure 4 is a block diagram of the fault detector module of Figure 3; Figure 5 is a flowchart illustrating a fault detection procedure according to a preferred embodiment of the present invention; and Figure 6 is a graph illustrating a further response spectrum that may be achieved using a fault detection procedure according to that illustrated by Figure 5.

Detailed Description of the Implementation

An intelligent fault detection system according to a preferred embodiment of the present invention will now be described. Such a system may automatically detect abnormal system behaviour, and may also:

- Intelligently modify the characteristics such as the shape and frequency range of a test signal such that it best suits the diagnosis goals; - Apply fuzzy reasoning using the knowledge of the expert operator in order to interpret the shape of the frequency spectrum of the system under test, and to relate this interpretation to system faults.

Figure 3 shows a block diagram of a system according to a preferred embodiment of the present invention.

With reference to Figure 3, a Configurable Signal Source 30 initially generates general purpose test signals with characteristics that can be controlled by input signals to the block. The test signals that can be generated may include signals such as sine wave signals, white noise, pulse train signals, and step input (although it is envisaged that most deployments will use white noise and/or sine wave signals as their initial test signals). Other types of signals may however be found to be applicable in relation to particular systems. The test signals produced by the Configurable Signal Source 30 are applied to the system under test 31 , which is the system being investigated for potential faults.

Responses produced by the system under test 31 are picked up by appropriate sensors of a Frequency Spectrum Evaluator 32, which may calculate the power spectrum of the response using an algorithm such as a Fast Fourier Transform (FFT) based algorithm. The power spectrum produced is then analysed by an Intelligent Fault Detector module 33 which encompasses the expertise of the operator about the shape of the output in a fuzzy rule base. A fuzzy rule base is preferably used because the nature of the knowledge is imprecise. With reference to Figure 4, the intelligent fault detector module 33 consists of a number of computational sub-modules which are used to fuzzify important characteristics of the power spectrum. As shown in Figure 4, the sub-modules shown in this example are as follows:

- A Fundamental Component Magnitude Evaluator 41: This evaluates changes in the magnitude of the fundamental components of the power spectrum. The results of this sub- module are fuzzy classifications using fuzzy membership functions (e.g. "small", "medium", "large")

- A Fundamental Component Frequency Evaluator 42: This evaluates changes in the frequency of the fundamental components of the power spectrum. The results of this module are again fuzzy classifications using fuzzy membership functions such as small, medium and large.

- A Harmonics Components Evaluator 43: This checks the spectrum qualitatively to determine whether it contains harmonics of the fundamental components and classifies their power in a fuzzy way (small, medium, large) in comparison with the fundamental power.

- A Spurious Components Evaluator 44: This checks the spectrum qualitatively to determine whether it contains spurious frequency components that do not correspond to harmonics of the fundamental components, and if so, classifies their power in a fuzzy manner (small, medium, large) in comparison with the fundamental power.

The processed response data from the computational sub-modules indicative of characteristics of the power spectrum is then provided to a module 45 which comprises a fuzzy rule base and reasoning engine.

An example of the type of fuzzy rules that may be used is as follows: [if first.mag.change=medium OR first.mag.change=large OR second.mag.change=medium OR second.mag.change=large] AND first.freq.change=small AND second.freq.change=small AND harmonics.exist=true AND harmonics.mag=(significant) %for at least one Then Error.Category,nonlinear=high

Using known fuzzy learning algorithms, the system may learn several such rules from observation or from supervised learning schemes.

Going back to Figure 3, the fault detection system according to a preferred embodiment of the present invention also has an interface 34 for receiving User Input. This may allow a user to submit a User Enquiry to the system. This may be to enter an initial FDI query, or run a general test, or provide instructions to the system in order to improve the fault detection process. More on this will be explained later. The system comprises a Results Display device 35, such as a standard graphic user interface. This allows for display of the FDI conclusions, and may also allow for the display of an associated degree of confidence in the conclusion (see later).

The fault detection system further includes an Intelligent Test Signal Configurator 36. This block is concerned mainly with providing instructions for changing characteristics of the test signal, such as the amplitude and/or the frequency content of the test signal, in order to obtain a better insight into the fault condition. The selection of the test signal characteristics may be dependent on the initial fault diagnosis (if there is one), or on the user query, or on both. For example, when the test signal has a low frequency component and the diagnosis is that the amplitude of the fundamental component has decreased, then the user might require a check to be made as to whether there is a non-linear effect. In this case the Signal Configurator 36 will provide instructions to extend the frequency content of the test signal to cover the first few harmonics of the fundamental signal, and thus examine the total harmonic distortion of the output.

With reference to Figure 5, a fault detection system such as that outlined above could be deployed in the following manner, according to a preferred embodiment of the invention. In the flow chart of figure 5, the fault detection process is shown as consisting of two stages. In Stage 1, firstly a default test signal such as a wide band noise signal is used to evaluate the frequency spectrum. The test signal is applied to the system under test (Step 1), and the response of the system thereto is collected (Step 2). The response is then converted into the form of a power spectrum (Step 3) and provided to the Intelligent Fault Detector (module 33 in Figures 3 and 4) which performs the detection of spectrum characteristics (Step 4) and the analysis to establish an initial fault assessment (Step 5). With reference to rules such as those stored in the fuzzy rule base, the reasoning engine applies its logic to the data derived from the response spectrum, and provides an initial fault report, together with a degree of confidence in the fault report.

The degree of confidence of the fault report can be determined using the fuzzy output of the diagnosis rules. If for example one rule is triggered with a "high" output and many others are triggered with a "low" output then the degree of confidence is "high" by choosing the first rule and ignoring the rest. On the other hand, if more than one rule is triggered with a "high" or "medium" output, then a weighted combination of these should be chosen to collect the degree of confidence in a compound diagnosis. This is conditional on there not being a conflict between the respective diagnoses from the two such that they are incompatible in forming such a compound diagnosis. Conflicting outcomes have to be considered at the time of design of the knowledge base.

If the degree of confidence in the fault report is higher than a certain threshold (in a fuzzy sense) the fault detection process may be terminated at the end of Stage 1 , and the fault report sent to the display unit (Step 12) for display to the user.

The second stage of the process is triggered if the degree of confidence is lower than the threshold. In this case the Intelligent Signal Configurator adjusts the input signal bandwidth and/or amplitude and/or other characteristics of the input signal to test the system under more specific conditions (Step 6). As explained earlier, the reconfiguring of the test signal may be carried out fully automatically (i.e. without user input), or manually by the user, or semi-automatically (i.e. by a combination of the above methods).

If Stage 2 of the process is triggered, it starts with the application of the reconfigured test signal to the system under test (Step 7). As before, the response of the system is collected (Step 8) and converted into a form such as a power spectrum (Step 9), before being provided to the Intelligent Fault Detector (33 in Figures 3 and 4) which performs the detection of spectrum characteristics (Step 10). Second stage analysis, similar to the first stage analysis, is then carried out to establish a further fault assessment (Step 11). A degree of confidence can again be determined in relation to the fault report. The fault detection system may be configured such that the second-stage fault report is sent to the display unit (Step 12), together with an indication of the degree of confidence, or the system may be configured such that a low level of confidence in the second-stage fault report triggers one or more further stages of reconfiguring the test signal and repeating the steps of Stage 2. This may occur automatically, or under the control of a user.

Figure 6 shows the type of spectrum that may result from performing Stage 2 of the process illustrated by Figure 5, in order to obtain better information about a system having a response spectrum such as those of Figures 1 and 2. Here the test signal configurator (36 in Figures 3) has applied a narrow bandwidth signal around one of the fundamental harmonics in order to assess the effect of saturation ^'on that part of the spectrum.

Claims

1) A system for predicting behavioural changes in dynamic systems, comprising: signal providing means arranged to apply a reconfigurable test signal to a dynamic system; monitoring means arranged to monitor responses of the dynamic system to the application of said test signal, and to provide response data indicative of said responses; response processing means arranged to detect changes in said response data and provide predictions therefrom relating to whether the dynamic system is undergoing behavioural changes; and signal reconfiguring means arranged to interact with the signal providing means whereby to reconfigure the test signal.

2) A system according to claim 1 , wherein the signal reconfiguring means is arranged to interact with the signal providing means in a manner dependent on the response of the dynamic system to the application of one or more previously applied test signals.

3) A system according to claim 1 or 2, further comprising user input means for receiving instructions from a user, said signal reconfiguring means being responsive to said user input means whereby to reconfigure the test signal in accordance with instructions received from the user.

4) A system according to claim 1 , 2 or 3, further comprising prediction assessment means arranged to determine a measure of confidence relating to predictions provided by the response processing means.

5) A system according to claim 4, wherein said prediction assessment means is arranged to determine said measures of confidence using fuzzy logic.

6) A system according to claim 4 or 5, further comprising output means for providing information to a user relating to said predictions in the event that the prediction assessment means determines the measure of confidence to be above a predetermined threshold. 7) A system according to any of claims 4, 5 or 6, further comprising instructing means for providing instructions to the signal reconfiguring means, said signal reconfiguring means being arranged to reconfigure the test signal in accordance with instructions received from the instructing means.

8) A system according to claim 7, wherein said instructing means is responsive to measures of confidence determined by the prediction assessment means, and is arranged to provide instructions to the signal reconfiguring means in the event that the prediction assessment means determines the measure of confidence to be below a predetermined threshold.

9) A system according to any of the preceding claims, wherein the signal providing means is arranged to apply an initial test signal comprising components of one or more or the following types: sine wave; white noise; pulse train; and step input.

10) A system according to any of the preceding claims, wherein the test signal is a mechanical or an acoustic signal.

11) A system according to any of the preceding claims, wherein the monitoring means comprises mechanical or acoustic sensing means.

12) A system according to any of claims 1 to 9, wherein the test signal is an electrical signal.

13) A system according to claim 12, wherein the monitoring means comprises electrical sensing means.

14) A system according to any of the preceding claims, wherein the monitoring means is arranged to provide the response data in the form of a power spectrum.

15) A system according to claim 14, wherein the response processing means comprises one or more of the following spectrum characteristic detection means: a fundamental component magnitude detection means for detecting changes in the magnitudes of fundamental components of the power spectrum; a fundamental component frequency detection means for detecting changes in the frequencies of fundamental components of the power spectrum; a harmonic components detection means for detecting the existence of harmonics of fundamental components in the power spectrum; a spurious components detection means for detecting the existence of spurious frequency components which do not correspond to harmonics of the fundamental components in the power spectrum.

16) A system according to claim 15, wherein the or each spectrum characteristic detection means comprises fuzzy classification means for classifying its detected characteristics using fuzzy sets.

17) A system according to any of the preceding claims, wherein the response processing means comprises predicting means for predicting whether the dynamic system is undergoing behavioural changes using fuzzy rules.

18) A system according to claim 17, further comprising output means for providing output to a user in the event that the predicting means predicts that the dynamic system is undergoing behavioural changes, said output including an indication relating to said prediction.

19) A system according to claim 17 or 18, wherein the response processing means further comprises means for determining, in the event that the predicting means predicts that the dynamic system is undergoing behavioural changes, which of said fuzzy rules led to said prediction.

20) A system according to claim 19, further comprising output means for providing output to a user in the event that the predicting means predicts that the dynamic system is undergoing behavioural changes, said output including an indication as to which of said fuzzy rules led to said prediction.

21) A method of predicting behavioural changes in dynamic systems, comprising steps of: applying a reconfigurable test signal to a dynamic system; monitoring responses of the dynamic system to the application of said test signal, and providing response data indicative of said responses; detecting changes in said response data and providing predictions therefrom relating to whether the dynamic system is undergoing behavioural changes; reconfiguring the test signal; and applying the reconfigured test signal to the dynamic system.

22) A method according to claim 21, wherein the test signal is reconfigured in a manner dependent on the response of the dynamic system to the application of one or more previously applied test signals.

23) A method according to claim 21 or 22, further comprising a step of receiving instructions from a user, said signal reconfiguring step comprising reconfiguring the test signal in accordance with instructions received from the user.

24) A method according to claim 21 , 22 or 23, further comprising a prediction assessment step whereby there is determined a measure of confidence relating to predictions as to whether the dynamic system is undergoing behavioural changes.

25) A method according to claim 24, wherein said prediction assessment step comprises determining said measures of confidence using fuzzy logic.

26) A method according to claim 24 or 25, further comprising a step of providing information to a user relating to said predictions in the event that said measure of confidence is determined to be above a predetermined threshold.

27) A method according to any of claims 24, 25 or 26, further comprising a step of reconfiguring the test signal in the event that a measure of confidence is determined to be below a predetermined threshold.

28) A method according to any of claims 21 to 27, wherein an initial test signal is applied comprising components of one or more or the following types: sine wave; white noise; pulse train; and step input. 29) A method according to any of claims 21 to 28, wherein the test signal is a mechanical or an acoustic signal.

30) A method according to any of claims 21 to 28, wherein the test signal is an electrical signal.

31) A method according to any of claims 21 to 30, wherein the response data is provided in the form of a power spectrum.

32) A method according to claim 31 , wherein the step of detecting changes comprises one or more of the following spectrum characteristic detection steps: detecting changes in the magnitudes of fundamental components of the power spectrum; detecting changes in the frequencies of fundamental components of the power spectrum; detecting the existence of harmonics of fundamental components in the power spectrum; detecting the existence of spurious frequency components which do not correspond to harmonics of the fundamental components in the power spectrum.

33) A method according to claim 32, wherein the or each spectrum characteristic detection step comprises classifying detected characteristics using fuzzy sets.

34) A method according to any of claims 21 to 33, wherein the step of detecting changes comprises predicting whether the dynamic system is undergoing behavioural changes using fuzzy rules.

35) A method according to claim 34, further comprising a step of providing output to a user in the event that a prediction is made that the dynamic system is undergoing behavioural changes, said output including an indication relating to said prediction.

36) A method according to claim 34 or 35, further comprising a step of determining, in the event that a prediction is made that the dynamic system is undergoing behavioural changes, which of said fuzzy rules led to said prediction. 37) A method according to claim 36, further comprising a step of providing output to a user in the event that a prediction is made that the dynamic system is undergoing behavioural changes, said output including an indication as to which of said fuzzy rules led to said prediction.