Sensor fault detection system

• US 6,016,465 A
• Filed: 01/12/1998
• Issued: 01/18/2000
• Est. Priority Date: 07/10/1995
• Status: Expired due to Fees

• Current Assignee: Qinetiq Limited (QinetiQ Group PLC)
• Original Assignee: Secretary of State For Defence
• Inventors: Kelly, Ronald W
• Primary Examiner(s): SHAH, KAMINI S
• Application Number:
  US09/005,547
• Time in Patent Office: 736 Days
• Field of Search: 702/116, 702/60-69, 702/179, 702/189, 702/191, 702/193, 702/195, 376/215-217, 345/914, 345/915
• US Class Current: 702/116
• CPC Class Codes: G05B 23/0235 based on a comparison with predetermined threshold or range, e.g. "classical methods", carried out during normal operation; threshold adaptation or choice; when or how to compare with the threshold
• CPC Class Codes: G05B 9/02 electric

Abstract

The sensor fault detector is one in which each sensor signal is compared with a reference signal provided by e.g. a model of the system being monitored. Prior art sensor fault detectors subtract the output of the sensor and the corresponding reference signal from each other, and a fault is declared when the residual difference exceeds a prescribed threshold. However, modelling errors (such as scaling discrepancies), d.c bias and noise mean that faults are often wrongly declared, or alternatively that only large faults can be detected. These problems are overcome according to the invention by comparing the "shape" of the sensor signal with that of the corresponding reference signal (i.e. comparing the signal outlines) over a finite time window, using e.g. correlation techniques. A fault is declared when the shapes differ by a prescribed amount.

Claims

1. A method of detecting sensor faults in a system having at least one sensor providing a sensor signal for monitoring or controlling system parameters, the method comprising the following steps:

(a) generating reference signals of the system;

(b) recording the value of each sensor signal and the corresponding reference signals at time intervals;

(c) comparing the shape of an outline formed from a series of values of each sensor signal with the shape of an outline formed from a series of corresponding modified reference signals by means of a correlation coefficient;

(d) continuously updating the respective series of signals with the most recent values and

(e) declaring a fault when the correlation coefficient falls outside a prescribed range, wherein the modified reference signal is generated by;

(f) multiplying the reference signal by a scaling factor to convert each sensor signal and its corresponding reference signal;

(g) calculating the mean values of each sensor signal and its corresponding reference signal;

(h) subtracting the respective mean values from each sensor signal and its corresponding reference signal to obtain new values for each sensor signal and its corresponding reference signal;

(i) applying deviation thresholds to either side of the new values for each sensor signal;

(j) superimposing the value of each reference signal over the thresholds around the new value of its corresponding sensor signal, and

(k) formulating a modified reference signal from a hybrid of values of the new sensor signal and the new reference signal.

4. A method of detecting sensor faults as claimed in claim 1 wherein in step (k) the modified reference signal comprises values of the new sensor signal where the new reference signal fails within said deviation thresholds, and those values of the new reference signal itself where the new reference signal fails outside said deviation thresholds.

5. A method of detecting sensor faults as claimed in claim 4 wherein in step (k) the modified reference comprises the new sensor signal values when the new reference signal lies within said deviation thresholds and comprising the new reference signal with the deviation threshold value added or subtracted therefrom when the new reference signal lies outside said deviation thresholds, such that the modified reference signal only differs from the sensor signal by the amount the new reference signal lies outside the deviation thresholds.

6. A method of detecting sensor faults as claimed in claim 1 comprising the additional step between steps (h) and (i) of calculating the areas under the new reference signal and new sensor signal and scaling that signal of them with smallest area by the ratio of the largest area to the smallest area.

7. A method for detecting sensor faults as claimed in claim 6 comprising the additional step of applying a dither signal to the either the new reference signal or the new sensor signal.

8. A method of detecting sensor faults as claimed in claim 1 wherein the deviation thresholds are allowed to vary in magnitude during operation of sensor fault detection.

2. A method of detecting sensor faults in a system having at least one sensor providing a sensor signal for monitoring or controlling system parameters, the method comprising the following steps:

(a) generating reference sensor signals of the system;

(b) recording the value of each sensor signal and the corresponding reference sensor signals at regular intervals;

(c) comparing the shape characteristics of a series of values of each sensor signal with the shape characteristics of a series of corresponding modified reference signals by means of a correlation coefficient;

(d) continuously updating the respective series of signals with the most recent values and

(e) declaring a fault when the correlation coefficient falls outside a prescribed range, wherein the modified reference signal is generated by;

(f) multiplying the reference signal by a scaling factor to convert each sensor signal and its corresponding reference signal;

(g) calculating the mean values of each sensor signal and its corresponding reference signal;

(h) subtracting the respective mean values from each sensor signal and its corresponding reference signal to obtain new values for each sensor signal and its corresponding reference signal;

(i) applying deviation thresholds to either side of the new values for each sensor signal;

(j) superimposing the value of each reference signal over the thresholds around the new value of its corresponding sensor signal, and

(k) formulating a modified reference signal from a hybrid of values of the new sensor signal and the new reference signal, wherein said correlation coefficient is of the form ##EQU3## Where x_n is the nth sample of the sensor signal;

y_n the nth sample of the reference signal;

x is the mean value of the sensor signal taken over N samples and

y is the mean value of the reference signal taken over N samples.

3. A method of detecting sensor faults in a system having at least one sensor providing a sensor signal for monitoring or controlling system parameters, the method comprising the following steps:

(a) generating reference sensor signals of the system;

(b) recording the value of each sensor signal and the corresponding reference sensor signals at regular intervals;

(c) comparing the shape characteristics of a series of values of each sensor signal with the shape characteristics of a series of corresponding modified reference signals by means of a correlation coefficient;

(d) continuously updating the respective series of signals with the most recent values and

(e) declaring a fault when the correlation coefficient falls outside a prescribed range, wherein the modified reference signal is generated by;

(f) multiplying the reference signal by a scaling factor to convert each sensor signal and its corresponding reference signal;

(g) calculating the mean values of each sensor signal and its corresponding reference signal;

(h) subtracting the respective mean values from each sensor signal and its corresponding reference signal to obtain new values for each sensor signal and its corresponding reference signal;

(i) applying deviation thresholds to either side of the new values for each sensor signal;

(j) superimposing the value of each reference signal over the thresholds around the new value of its corresponding sensor signal, and

(k) formulating a modified reference signal from a hybrid of values of the new sensor signal and the new reference signal, wherein said correlate coefficient is of the form ##EQU4## Where x_n is the nth sample of the sensor signal;

y_n the nth sample of the reference signal;

x is the mean-value of the sensor signal taken over N samples and

y is the mean value of the reference signal taken over N samples

a is a constant.