JPH011914A - Sensor abnormality diagnosis method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a sensor abnormality diagnosis method that can be applied to a preprocessing device of a plant abnormality diagnosis system.

[Prior Art] Conventionally, in sensor abnormality diagnosis, three or more identical sensors are arranged and an abnormal sensor is extracted by majority vote. However, this method increases costs, so its application is limited. For this reason, two sensors are often used to detect abnormalities by checking their deviations, but it is not possible to identify which sensor is abnormal.
Furthermore, it was not possible to estimate the abnormal state of the sensor.

[Problems to be solved by the invention] In the conventional sensor abnormality diagnosis method using two sensors, as mentioned above, only the deviation between the output signals of the two sensors is focused, and abnormality or normality is determined based on the magnitude of the deviation. Therefore, it is difficult to detect an abnormal sensor, and it is also difficult to determine an abnormal state.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above conventional problems and provide a sensor abnormality diagnosis method that can detect an abnormal sensor and simultaneously determine the cause of the abnormality even when two sensors are arranged.

According to the present invention, the above function can be achieved by finding a linear regression equation between the output signals of two sensors and utilizing the fact that its slope, intercept, and center of gravity show independent changes depending on the states of both sensors. This is what I did.

However, since changes in the slope, intercept, and center of gravity vary depending on the characteristics of the sensor and the characteristics of the placement process, fuzzy inference is used for processing. Here, fuzzy inference means defining a membership function called a membership subfunction to represent the state of a signal as a fuzzy set, and performing set operations to infer changes in the state.

[Means for solving the problem] The sensor abnormality diagnosis method according to the present invention expresses the relationship between the output signals of two sensors by a linear regression equation, and calculates the slope of the equation,
Identify abnormal sensors by processing changes in the intercept and center of gravity using methods such as fuzzy inference,
It is also characterized by inferring the cause of the abnormality.

[Operation] According to the present invention, the time term of the output signals of the two sensors can be eliminated using a linear regression equation, and the difference in characteristics between the two can be reexpressed using three parameters. This makes it possible to judge changes in parameters by taking into account the characteristics of sensors and processes, and to process even ambiguous stages of normality and abnormality, compared to normal size judgments.

[Example] FIG. 1 is a block diagram showing an example of a device used to implement the method of the present invention, in which 1 is a sensor A, 2 is a sensor B, 3 is a linear regression analyzer, and 4 is a fuzzy inference device. , 5 is a sensor abnormality diagnosis device, 6 is a process wheat seedling, 7 is an output A, 8
indicates the output B, and 9 indicates the diagnosis result.

In FIG. 1, sensor A1 and sensor B2 send outputs A7 and B8 to sensor abnormality diagnosis device 5 for the same process variable 6. In the sensor abnormality diagnosis device 5,
A linear regression equation B-α·A+β is determined by the linear regression analyzer 3 and sent to the fuzzy inference unit 4. The fuzzy reasoner 4 calculates the center of gravity position (m A ).

mB), intercept β, and slope α, the abnormal sensor is extracted and its cause is diagnosed, and a diagnosis result 9 is output.

Fig. 2 shows the principle of abnormality check, Fig. 2 (A) shows the case of gain change, Fig. 2 (B) shows the case of drift occurrence,
FIG. 3 defines membership functions. Although the shape of the function is determined by the object, a normal distribution function is used here. FIG. 4 and FIG. 5 show rule tables for fuzzy inference, and the symbols indicate the names of membership functions that correspond to each one.

”-Description The principle and method of one embodiment of the present invention will be explained.

Fig. 2 is a diagram showing the principle of sensor abnormality check. If the two sensors are sensor A and sensor B, their individual outputs are the same signal, so the AB two-dimensional plane of the output within a certain M1 period is The linear regression equation for is B −1 if both are normal.
,0*A + 0.0. Therefore, when either sensor A or B becomes abnormal, that is, when the gain or zero point of the sensor changes, the slope, intercept, and center of gravity of the linear regression equation exhibit specific changes depending on the abnormal state. For example, when the gain of the B sensor increases, the center of gravity shifts upward and the slope exceeds 1.0. If drift also occurs at the same time, the intercept also changes to a value other than 0.0.

Since all of these occur independently, it is possible to extract an abnormal sensor and determine its condition by examining the conditions of the three parameters, namely, the center of gravity, slope, and intercept.

However, the above judgment is only possible when the following prerequisites are met.

(i) The extent and time of occurrence of abnormality in sensors A and B can be treated as independent events.

(11) There is no sticking to a constant value. When sticking occurs, the condition is not determined by linear regression. Abnormal sensor extraction can be determined by moving the center of gravity.

It is possible to judge the state by tracking changes in the three parameters mentioned above, but if you track changes by determining the size of the values, which is usually done, there is a risk that the characteristics of the judgment will change significantly depending on the quality of the reference value settings. There is. In particular, it is difficult to simultaneously detect the gradual loss of sensor health and exclude malfunctions due to noise or the like.

In this example, changes were tracked using fuzzy inference and statistical processing to solve the problem.

FIG. 3 defines the form of the membership function, and uses a normal distribution function. The set of tables is determined sequentially using each average value and standard deviation value, and it is possible to follow changes over time in the process. Further, although fixed values are used for the slope and the set of intercepts, these values are determined empirically in consideration of the characteristics of the target sensor.

[Effects of the Invention] According to the present invention, sensor abnormality diagnosis can be carried out inexpensively and reliably, and it can be used in various fields as a pre-processing device when constructing a plant abnormality diagnosis system, etc., and is beyond practical use. Some excellent effects can be achieved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of a device used to carry out the method of the present invention, FIG. 2 is a principle diagram of sensor abnormality check in the method of the present invention, FIG. 3 is a diagram defining membership functions, and FIG. FIG. 5 and FIG. 5 are diagrams showing a rule table for fuzzy inference. DESCRIPTION OF SYMBOLS 1...Sensor A, 2...Sensor B, 3...Linear regression analyzer, 4...Fuzzy inference device, 5...Sensor abnormality diagnosis device. Figure 1 N-no-(LZ α) Castle

Claims (1)

[Claims] A linear regression equation between the output signals of the two related sensors is determined, and from the change in the slope, intercept, and center of gravity of the equation, the above 2.
1. A sensor abnormality diagnosis method characterized by detecting an abnormal sensor among two sensors and simultaneously determining whether the cause of the abnormality is a change in sensor gain or occurrence of drift.