CN112180312A

CN112180312A - Current sensor composite fault diagnosis method

Info

Publication number: CN112180312A
Application number: CN202010856198.9A
Authority: CN
Inventors: 陈复扬; 金帆; 王克杰; 李子恒
Original assignee: Nanjing University of Aeronautics and Astronautics
Current assignee: Nanjing University of Aeronautics and Astronautics
Priority date: 2020-08-24
Filing date: 2020-08-24
Publication date: 2021-01-05
Anticipated expiration: 2040-08-24
Also published as: CN112180312B

Abstract

The invention discloses a composite fault diagnosis method for a current sensor, which comprises the following steps: simulating two-phase current when the current sensor fails through simulation software to obtain corresponding current waveforms; extracting the characteristics of the current waveform from a time domain angle to obtain fault characteristics; screening and reconstructing fault characteristics by using a gradient lifting tree to obtain corresponding one-hot codes; taking the one-hot code as a training sample, and training by adopting a logistic regression algorithm to obtain a parameter-optimized combination model; and inputting the current sensor sample to be tested into the combined model, extracting the fault characteristics of the sample to be tested by the combined model according to the optimized parameters, diagnosing and outputting the fault condition. According to the invention, the fault characteristics are extracted from the angles of a plurality of time domain characteristic values, so that the omission of fault information is avoided; the dependence on an accurate physical model is low; the method can more accurately diagnose gain faults, bias faults and composite faults of the gain faults and the bias faults in the current sensor, and is simple and easy to implement.

Description

Current sensor composite fault diagnosis method

Technical Field

The invention relates to current sensor fault diagnosis, in particular to a current sensor composite fault diagnosis method.

Background

The traction transmission system of the high-speed train is connected with the personal safety and the property safety of passengers in a myriad ways, so that strict requirements are imposed on the reliability of the traction transmission system. The traction transmission system converts electric energy into mechanical energy, which is a core part of the whole high-speed train, if one of the components is damaged and cannot be processed in time, chain reaction may be caused, other components are influenced, composite obstacles are caused, and finally, the operation of the high-speed train is influenced.

The current sensor contained in the three-level inverter in the traction transmission system is most prone to failure, and an accurate and reliable failure diagnosis method is needed for supervision. The traditional fault diagnosis method is based on an analytical model method, but the method relies on an accurate mathematical model. At present, a sliding-mode observer method, a load current analysis method and the like are commonly used as methods for fault diagnosis of a current sensor, but the methods mainly consider single fault conditions, lack of extraction and analysis of different fault characteristics and are difficult to realize accurate diagnosis under the condition of compound faults; under the condition of compound faults, the representation of a single fault can be changed and cannot be described by simple linear superposition, so that the effect of a common data driving method is weakened.

Disclosure of Invention

The purpose of the invention is as follows: in view of the above problems, an object of the present invention is to provide a current sensor composite fault diagnosis method for diagnosing a gain fault, a bias fault, and a composite fault of both occurring in a current sensor.

The technical scheme is as follows: the invention provides a composite fault diagnosis method for a current sensor, which comprises the following steps:

(1) simulating two-phase current when the current sensor fails through simulation software to obtain corresponding current waveforms;

(2) extracting the characteristics of the current waveform from a time domain angle to obtain fault characteristics;

(3) screening and reconstructing fault characteristics by using a gradient lifting tree to obtain corresponding unique hot codes, and normalizing and converting the fault characteristics of different magnitudes into the same magnitude so as to avoid the influence of the different magnitudes of values on a model;

(4) taking the one-hot code as a training sample, and training by adopting a logistic regression algorithm to obtain a parameter-optimized combination model;

(5) inputting a current sensor sample to be detected into a combined model, extracting fault characteristics of the sample to be detected by the combined model according to optimized parameters, respectively sending the characteristic samples to a gain fault classifier and a bias fault classifier for fault classification, and respectively outputting detection results by the two classifiers to finish fault diagnosis of the detected sample.

The fault types in the step (1) comprise a bias fault, a gain fault and a composite fault when two faults occur simultaneously.

The fault characteristics in the step (2) comprise mean values, maximum values, minimum values, range differences, standard deviations, mean square values, root mean square, skewness, kurtosis factors, wave form factors, pulse factors and margin factors of four conditions of no fault, gain fault, bias fault and composite fault.

And (4) the combined model in the step (4) consists of the probability of occurrence of the gain fault and the polarization fault.

And (5) the fault conditions comprise no fault, single fault and compound fault.

Has the advantages that: compared with the prior art, the invention has the following remarkable advantages:

1. the fault diagnosis method is based on data, and has low dependence on an accurate physical model;

2. fault characteristics are extracted from the angles of the time domain characteristic values, and omission of fault information is avoided;

3. screening fault features by using a gradient lifting tree, which is beneficial to extracting feature information with high discrimination;

4. training the model by adopting a logistic regression algorithm, so that the model parameter optimization is facilitated;

5. a combined model formed by the double fault classifiers is constructed, the gain fault, the bias fault and the composite fault of the gain fault and the bias fault in the current sensor can be diagnosed more accurately, and the method is simple and easy to implement.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a graph of two-phase current waveforms without fault;

FIG. 3 is a diagram of two-phase current waveforms during an A-phase current sensor offset fault

FIG. 4 is a graph of two-phase current waveforms during a gain fault for a phase A current sensor;

FIG. 5 is a graph of two-phase current waveforms at a composite fault of the A-phase current sensor;

FIG. 6 is a time domain fault signature distribution plot for different fault conditions of a current sensor;

FIG. 7 is a schematic flow chart of a process for gradient lifting tree;

FIG. 8 is a flow chart of the combinatorial model.

Detailed Description

The composite fault diagnosis method for the current sensor, disclosed by the invention, has the flow chart shown in figure 1, and comprises the following steps of:

(1) two-phase current when a bias fault, a gain fault and a composite fault occur simultaneously in the current sensor is simulated through simulation software, and corresponding current waveforms are obtained. As shown in Table 1, the simulated fault parameter setting conditions are as shown in Table 1, and the fault conditions are 81 groups, wherein the gain fault degree is 1 to 1.4, 9 degrees are set, and the degree 1 represents no gain fault; the degree of bias failure was set to 9 degrees from 0 to 40, and the degree 0 indicates no bias failure. Four exemplary sets of current waveform diagrams for the a-phase current sensor with no fault, gain fault, bias fault, and compound fault are shown in fig. 2-5.

TABLE 1 Current sensor composite Fault level settings

(2) And carrying out feature extraction on the current waveform from a time domain angle to obtain fault features. The fault characteristics include mean, maximum, minimum, range, standard deviation, mean square, root mean square, skewness, kurtosis factor, form factor, pulse factor, margin factor, as shown in fig. 6.

(3) And screening and reconstructing fault characteristics by using the gradient lifting tree to obtain corresponding one-hot codes. The gradient lifting tree processing flow diagram is shown in fig. 7.

(4) The one-hot code is used as a training sample, and training is performed by using a logistic regression algorithm to obtain a combined model with optimized parameters, as shown in fig. 8, wherein the combined model comprises a time domain feature extraction link, a gain fault classifier and a bias fault classifier.

(5) Inputting a current sensor sample to be detected into a combined model, extracting fault characteristics of the sample to be detected by the combined model according to optimized parameters, respectively sending the characteristic samples to a gain fault classifier and a bias fault classifier for fault classification, respectively outputting detection results by the two classifiers, and finally judging whether the output sample belongs to no fault, single fault or composite fault according to corresponding rules shown in table 2, namely completing fault diagnosis of the detected sample.

TABLE 2 correspondence between current sensor fault type and classifier output

Claims

1. A composite fault diagnosis method for a current sensor is characterized by comprising the following steps:

(3) screening and reconstructing fault characteristics by using a gradient lifting tree to obtain corresponding one-hot codes;

2. The current sensor composite fault diagnosis method according to claim 1, wherein the step (1) fault types include a bias fault, a gain fault, and a composite fault when two faults occur simultaneously.

3. The current sensor composite fault diagnosis method of claim 2, wherein the fault characteristics of step (2) include mean, maximum, minimum, range, standard deviation, mean square, root mean square, skewness, kurtosis factor, form factor, pulse factor and margin factor of four conditions of no fault, gain fault, bias fault and composite fault.

4. The current sensor composite fault diagnosis method according to claim 1, wherein the combined model in the step (4) is composed of a gain fault and a polarization fault occurrence probability.

5. The current sensor composite fault diagnostic method of claim 1, wherein the step (5) fault conditions include no fault, single fault, and composite fault.