Fault Detection Method for Traction Motor of High-speed Train

Abstract

The present invention relates to the field of fault diagnosis technology, and specifically to a method for fault detection of a traction motor of a high-speed train. The steps of the present invention are: establishing a state space model of the traction motor in continuous time, constructing a traction motor fault model that reflects the actual fault situation, establishing a discrete fault model of the traction motor in discrete time, establishing a local T-S fuzzy model of the traction motor, establishing a global T-S fuzzy system of the traction motor, constructing a residual generator, constructing a performance index, solving the gain matrix of the fault detection filter, constructing a residual evaluation function and determining a threshold, and judging whether a fault occurs. The present invention can better describe the dynamic behavior of the traction motor and can detect the fault of the traction motor; using H i / H ∞ The optimized fault detection filter method is used to design the residual generator, which effectively suppresses the influence of multi-source interference while ensuring that the residual is sensitive to faults, thereby achieving effective fault detection.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a fault detection method for a traction motor of a high-speed train. Background Art

[0002] High-speed trains, as an efficient mode of transportation, not only carry people's travel needs, but also are an important engine for promoting economic development. Their status is becoming increasingly prominent in modern society. As the power core of high-speed trains, the reliability and stability of the traction drive system are undoubtedly the key to ensuring the safe and efficient operation of trains. Since the traction motor plays a pivotal role, its operating status is directly related to the safety and work efficiency of high-speed trains. However, due to long-term high-load operation and complex and changeable working environment, traction motors face many potential risks such as overheating, voltage power supply imbalance, insulation failure, etc., all of which may cause motor failure. Once the traction motor fails, the high-speed train operation control system will lose the real-time and accurate regulation of the train operation status, which will not only affect the performance of the motor, but also pose a serious threat to the safety of the entire train. Therefore, when the traction motor fails, it is crucial to quickly and accurately detect and locate the fault to ensure the safe operation of the high-speed train.

[0003] At present, although the model-based fault detection method has been widely used in the field of traction motor fault detection, it still faces the difficulty of fault detection under multi-source interference. These interferences cause the residual signal to contain unknown input information that is irrelevant to the fault, making it difficult to accurately decouple and detect the fault. ∞ The development of robust control theory, H ∞ Filtering is widely used in the field of fault detection. However, this method can only ensure that the robustness of the residual to unknown input is as large as possible, but it is difficult to take into account the sensitivity of the residual to faults. In addition to this method, H i / H ∞ The optimization method is another widely used method for robust fault detection. In order to balance the robustness and sensitivity of the fault detection system, this method transforms the fault detection filter design problem in the form of an observer into an H ∞ / H ∞ and H - / H ∞ Optimization (i.e. H i / H ∞ The optimization problem makes the residual achieve the best balance between sensitivity to faults and robustness to unknown inputs. ∞ Compared with H i / H ∞The optimized robust fault detection method can better suppress the influence of unknown inputs and show more excellent robust performance. However, the traction motor is a complex nonlinear system with strong coupling and multiple variables. The traditional observer-based method can usually only detect a certain nonlinear system with low linearization degree, which makes the H-based method i / H ∞ The application of optimized robust fault detection methods in traction motor fault detection is limited to a certain extent. With the continuous development of intelligent methods, some technologies such as fuzzy logic, neural networks, genetic algorithms, etc. have been introduced into observer-based fault detection methods, which can make up for the limitations of traditional observer methods in nonlinear systems to a certain extent. However, in the field of high-speed train traction motor fault detection, no relevant research has yet been conducted to apply H i / H ∞ The optimized robust fault detection method is combined with the intelligent method, so this direction needs further research. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for detecting faults of a traction motor of a high-speed train.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for detecting a fault of a traction motor of a high-speed train comprises the following steps:

[0007] S1. Based on the dynamic mathematical model of the traction motor and the sensor measurement equation, a state space model of the traction motor is established in continuous time;

[0008] S2. Based on the traction motor state space model established in step S1, multi-source interference and speed sensor fault are introduced to construct a traction motor fault model reflecting the actual fault condition;

[0009] S3, discretizing the traction motor fault model established in step S2 to establish a discrete fault model of the traction motor under discrete time;

[0010] S4, select s fuzzy rule antecedent variables, , , , ,Establish IF THEN fuzzy rules, fuzzy linearize the discrete fault model of the traction motor obtained in step S3, and establish a local TS fuzzy model of the traction motor;

[0011] S5, performing weighted average defuzzification on the local TS fuzzy model to establish a global TS fuzzy model of the traction motor;

[0012] S6, constructing a residual generator for the global TS fuzzy model established in step S5 by using a fault detection filter method;

[0013] S7, construct performance indicators reflecting the robustness of the residual to multi-source interference and initial estimation error and the sensitivity to faults;

[0014] S8, adopting the Hi / H∞ optimized fault detection filter method, transforming the design problem of the residual generator into solving the gain matrix of the fault detection filter;

[0015] S9, constructing a residual evaluation function within the sliding time window and determining a threshold;

[0016] S10. Compare the residual evaluation function and the threshold according to the fault detection logic to determine whether a fault occurs.

[0017] This technical solution applies TS fuzzy theory to process the nonlinear model of the traction motor, which can better describe the dynamic behavior of the traction motor, and can apply the more mature linear system robust fault detection theory to detect the faults of the traction motor; the Hi / H∞ optimized fault detection filter method is used to design the residual generator, which effectively suppresses the influence of multi-source interference while ensuring that the residual is sensitive to faults. This enables the fault detection system to effectively reflect the faults contained in the residuals, thereby realizing effective fault detection.

[0018] According to an embodiment of the present invention, in step S1, a traction motor state space model is established in continuous time, which is described as follows:

[0019]

[0020] Where: is continuous time; is the state vector; is the input vector; is the measurement output vector;

[0021] , They are the state matrix and input matrix of the continuous fault model of the traction motor, describing the motion form of the traction motor;

[0022] is the measurement output matrix, describing the measurement output form of the traction motor;

[0023] , , It is determined by the dynamic equation of the traction motor and the output form of the sensor, and the specific parameters are determined by the traction motor model.

[0024] According to an embodiment of the present invention, in step S2, a traction motor fault model constructed by introducing multi-source interference and speed sensor fault is described as follows:

[0025]

[0026] Where: , They are the stresses that the traction motor is subjected to under complex operating conditions. The norm-bounded multi-source interference, the former is the system noise, and the latter is the measurement noise;

[0027] for Norm-bounded speed sensor fault vector;

[0028] The process noise input matrix of the continuous fault model of the traction motor is determined by the form of multi-source interference;

[0029] is the sensor fault input matrix, determined by the form of the fault.

[0030] According to an embodiment of the present invention, in step S3, the traction motor discrete fault model established by discretizing the traction motor fault model is described as follows:

[0031]

[0032] Where: is discrete time, is the discretized sampling period;

[0033] , , .

[0034] According to an embodiment of the present invention, in step S4, s fuzzy rule antecedent variables are selected, where: each fuzzy rule has the same form but is independent of each other, Fuzzy rules , described as follows:

[0035]

[0036] Where: The guiding statement is the antecedent of the fuzzy rule, which represents the condition of each rule;

[0037] The guided statement is the conclusion of the fuzzy rule, which means the nonlinear model established in step S3 is obtained after fuzzy linearization. Locally Linearized Subsystem Model, superscript , which represents the fuzzy linearization of the nonlinear fault model The first of the fuzzy rules strip;

[0038] Antecedent variable In its field The membership function on , Domain The minimum value that can be obtained, Domain The maximum value that can be obtained, subscript Corresponds to The first of the antecedent variables indivual; , , , , Respectively Limited Subsystem The state matrix, input matrix, process noise input matrix, measurement output matrix, and fault input matrix.

[0039] According to an embodiment of the present invention, in step S5, a global TS fuzzy model of the traction motor is established, which is described as follows:

[0040]

[0041] Where: , , , , They are the global TS fuzzy models obtained after weighted defuzzification. The state matrix, input matrix, process noise input matrix, measurement output matrix, and fault input matrix are calculated by the following equations:

[0042] , ,

[0043] ,

[0044] Where: For the A normalized fuzzy membership function is calculated as follows:

[0045]

[0046] Where: Antecedent variable In its field For fuzzy subsets The membership function of is calculated as follows:

[0047]

[0048] Where: is a time-varying binary parameter, that is, each antecedent variable Corresponding to two different fuzzy subsets and .

[0049] According to an embodiment of the present invention, in step S6, a fault detection filter method is used to construct a residual generator, which is described as follows:

[0050]

[0051] Where: is the initial state estimate, for State estimation at the moment;

[0052] for The output estimation error at the moment is the new information; is the generated residual signal;

[0053] , They are the observer gain matrix and post-filter parameter matrix to be designed respectively;

[0054] Then define the state estimation error variable:

[0055]

[0056] The error dynamic system is described as:

[0057]

[0058] Where: is the initial estimation error, , , .

[0059] According to an embodiment of the present invention, in step S7, constructing a performance indicator reflecting the robustness of the residual to multi-source interference and initial estimation error and the sensitivity to faults includes the following steps:

[0060] S71, using H∞ norm As a performance indicator reflecting the robustness of the residual to multi-source interference and initial estimation error, it is described as follows:

[0061]

[0062] Where: is the residual signal in the absence of fault; The maximum value of a; superscript is the transpose symbol; is the continuous summation symbol, for ; Given by:

[0063]

[0064] S72, using H∞ norm and H-index As the best and worst case performance indicators reflecting the sensitivity of residual to faults, they are described as follows:

[0065]

[0066]

[0067] Where: is the residual signal without multi-source interference and zero initial estimation error; is the maximum value of a under condition b; is the minimum value of a under condition b.

[0068] According to one embodiment of the present invention, in step S8, the design problem of the residual generator is transformed into solving the gain matrix of the fault detection filter, then the observer gain matrix and the post-filter parameter matrix :

[0069] or

[0070] Where: is the maximum value of a under condition b;

[0071] The process noise and measurement noise from multiple sources are expanded into the following form:

[0072]

[0073] The process noise input matrix and the measurement noise input matrix are expanded into the following forms:

[0074]

[0075]

[0076] In the formula, Represents the identity matrix, with an order of 4;

[0077] The residual generator is designed by using the Hi / H∞ optimized fault detection filter method. Then the observer gain matrix is and the post-filter parameter matrix Calculated recursively by the following equation:

[0078]

[0079] Where: is the estimated error covariance needle, For new information The Gram matrix.

[0080] According to one embodiment of the present invention, in step S9 and step S10, the residual evaluation function constructed within the sliding time window n is described as follows:

[0081]

[0082] The threshold is the tolerance of unknown disturbances and uncertainties in the operation of a fault-free system, so the threshold is defined as follows:

[0083]

[0084] Where: is the residual evaluation function of the fault-free system;

[0085] According to the fault detection logic, the residual evaluation function and the threshold are compared to determine whether a fault has occurred:

[0086]

[0087] When the evaluation function exceeds the set threshold, it indicates that a fault has occurred and an alarm is triggered;

[0088] When the evaluation function does not exceed the set threshold, it indicates that no fault occurs and no alarm is triggered.

[0089] Compared with the prior art, the present invention has the following beneficial effects:

[0090] The present invention adopts TS fuzzy H i / H ∞The optimized fault detection method realizes the fault detection of high-speed train traction motor. In the formula: TS fuzzy model can realize accurate fuzzy modeling of nonlinear system online through actual input and output data. Compared with the traditional modeling method, the description of TS fuzzy model is closer to the actual system, and the currently mature linear system fault detection theory can be extended to solve the fault detection problem of nonlinear system; based on H i / H ∞ Compared with the traditional fault detection method based on fault detection filter, the optimized fault detection method can effectively suppress the influence of unknown inputs such as multi-source interference on the residual while ensuring that the residual is as sensitive as possible to faults, and can more effectively reflect all the fault information contained in the residual. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 It is a flow principle block diagram of the fault detection method of the present invention.

[0092] Figure 2 is a graph of the residual output and threshold value when the speed sensor is not faulty.

[0093] Figure 3 It is a curve diagram of the residual evaluation results when a constant deviation fault occurs in the speed sensor.

[0094] Figure 4 It is a curve diagram of the residual evaluation results when a speed sensor has a gradual failure. DETAILED DESCRIPTION

[0095] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0096] Example 1

[0097] like Figure 1 As shown, this embodiment provides a method for detecting a fault of a traction motor of a high-speed train, comprising the following steps:

[0098] S1. Based on the dynamic mathematical model of the traction motor and the sensor measurement equation, a traction motor state space model is established in continuous time:

[0099]

[0100] Where: is continuous time; is the state vector; is the input vector; is the measurement output vector; , They are the state matrix and input matrix of the continuous fault model of the traction motor, describing the motion form of the traction motor; is the measurement output matrix, which describes the measurement output form of the traction motor; , , Determined by the dynamic equation of the traction motor and the output form of the sensor, they are as follows:

[0101] , ,

[0102]

[0103] Where: , , They are the number of magnetic pole pairs, moment of inertia, and load torque of the traction motor; the other parameters , , , , , They are defined as follows:

[0104] , , , , ,

[0105] In the formula, , are the resistance values ​​of the stator and rotor of the traction motor respectively; is the mutual inductance between the stator and the rotor, , are the self-inductances of the stator and rotor of the traction motor respectively.

[0106] S2. Based on the traction motor state space model established in step S1, multi-source interference and speed sensor fault are introduced to construct a traction motor fault model reflecting the actual fault condition as follows:

[0107]

[0108] Where: and The traction motor is subjected to The norm-bounded multi-source interference, the former is the system noise and the latter is the measurement noise; for Norm-bounded speed sensor fault vector; The process noise input matrix of the continuous fault model of the traction motor is determined by the form of multi-source interference; is the sensor fault input matrix, determined by the form of the fault.

[0109] S3, discretize the traction motor fault model in step S2, and obtain the traction motor fault model in discrete time as follows:

[0110]

[0111] Where: is discrete time, is the discretized sampling period;

[0112] , , .

[0113] S4, perform local fuzzy linearization processing on the traction motor fault model established in step S3 under discrete time. Considering the convenience of data acquisition, the measurable output of the traction motor is selected as the antecedent variable of the fuzzy rule, that is, , , , , thus establishing 16 IF THEN fuzzy rules, each fuzzy rule has the same form but is independent of each other. Fuzzy rules The description is as follows:

[0114]

[0115] Where: The guiding statement is the antecedent of the fuzzy rule, which represents the condition of each rule; The guided statement is the conclusion of the fuzzy rule, which means the nonlinear model established in step S3 is obtained after fuzzy linearization. Locally Linearized Subsystem Model, superscript , which represents the fuzzy linearization of the nonlinear fault model The first of the fuzzy rules strip; Represents the antecedent variable In its field The membership function on , Domain The minimum value that can be obtained, Domain The maximum value that can be obtained, subscript Corresponds to The first of the antecedent variables indivual; , , , , Respectively Limited Subsystem The state matrix, input matrix, process noise input matrix, measurement output matrix and fault input matrix are as follows:

[0116]

[0117] , , ,

[0118] Where: , ,parameter It is expressed as:

[0119]

[0120]

[0121]

[0122]

[0123] S5. Perform weighted average defuzzification on the local TS fuzzy system and establish the global TS fuzzy system of the traction motor as follows:

[0124]

[0125] Where: , , , , They are the TS global fuzzy systems obtained after weighted defuzzification. The state matrix, input matrix, process noise input matrix, measurement output matrix and fault input matrix are calculated by the following equations:

[0126] , ,

[0127] ,

[0128] Where: For the A normalized fuzzy membership function is calculated as follows:

[0129]

[0130] Where: Antecedent variable In its field For fuzzy subsets The membership function of is calculated as follows:

[0131]

[0132] Where: is a time-varying binary parameter, that is, each antecedent variable Corresponding to two different fuzzy subsets and .

[0133] S6. For the TS fuzzy model established in step S5, a residual generator is constructed using a method based on a fault detection filter as follows:

[0134]

[0135] Where: is the initial state estimate, for State estimation at the moment; for The output estimation error at the moment is the new information; is the generated residual signal; and are the observer gain matrix and post-filter parameter matrix to be designed respectively; define the state estimation error variable:

[0136]

[0137] The error dynamic system can be described as:

[0138]

[0139] Where: is the initial estimation error, , , .

[0140] S7. Construct performance indicators that reflect the robustness of the residual to multi-source interference and initial estimation error and the sensitivity to faults. The specific contents are as follows:

[0141] Using H ∞ Norm As a performance indicator reflecting the robustness of the residual to multi-source interference and initial estimation errors, it is defined as follows:

[0142]

[0143] Where: is the residual signal in the absence of fault; The maximum value of a; superscript is the transpose symbol; is the continuous summation symbol, for ; Given by:

[0144]

[0145] Using H ∞ Norm and H - index As the best and worst case performance indicators reflecting the sensitivity of residual to faults, they are defined as follows:

[0146]

[0147]

[0148] Where: is the residual signal without multi-source interference and zero initial estimation error; is the maximum value of a under condition b; is the minimum value of a under condition b.

[0149] S8, using H i / H ∞ The optimized fault detection filter method transforms the design problem of the residual generator into solving the fault detection filter gain matrix. The specific contents are as follows:

[0150] The design problem of the residual generator is transformed into solving the observer gain matrix under the following problem and the post-filter parameter matrix :

[0151] or

[0152] Where: is the maximum value of a under condition b;

[0153] The process noise and measurement noise from multiple sources are expanded into the following form:

[0154]

[0155] The process noise input matrix and the measurement noise input matrix are expanded into the following forms:

[0156]

[0157]

[0158] In the formula, Represents the identity matrix, with an order of 4;

[0159] Furthermore, using H i / H ∞ The optimized fault detection filter method designs the residual generator, where the observer gain matrix and the post-filter parameter matrix Calculated recursively by the following equation:

[0160]

[0161] Where: is the estimated error covariance needle, For new information The Gram matrix.

[0162] S9. The residual evaluation function constructed in the sliding time window n is as follows:

[0163]

[0164] The threshold is usually regarded as the tolerance of unknown disturbances and other uncertainties in the operation of the fault-free system, so the threshold is defined as follows:

[0165]

[0166] Where: is the residual evaluation function of the fault-free system.

[0167] S10, comparing the residual evaluation function and the threshold according to the following fault detection logic to determine whether a fault has occurred:

[0168]

[0169] The above formula shows that when the evaluation function exceeds the set threshold, it indicates that a fault has occurred and an alarm is triggered; when the evaluation function does not exceed the set threshold, it indicates that no fault has occurred and no alarm is triggered.

[0170] The present invention takes the MT205 three-phase squirrel cage asynchronous motor used in China Railway High-speed No. 2 (CRH2) EMU as an example to detect the constant deviation fault and gradual fault of the speed sensor. The relevant parameters of the traction motor are given in Table 1.

[0171] Table 1 Traction motor related parameters

[0172] Set the simulation time to 100s and the sampling period to s, the occurrence time of two different types of faults are set at s is the step length The fault detection method proposed in the present invention is used for fault detection, and the threshold is selected by Figure 2 The test results are given by Figure 3 and Figure 4 Given.

[0173] Figure 2 In the figure, the horizontal axis represents the residual evaluation function and the threshold value when there is no fault, the vertical axis represents the simulation step, the solid line represents the residual evaluation function of fault detection, and the dotted line represents the threshold selected for the residual evaluation. Figure 3 and Figure 4 In the figure, the horizontal axis represents the size of the residual evaluation function and the threshold, the vertical axis represents the simulation step, the solid line represents the residual evaluation function of fault detection, the dotted line represents the threshold selected for residual evaluation, and the dotted line represents the moment when the fault detection system detects a fault and issues an alarm. Figure 2 It can be seen that the threshold is selected as the maximum value of the residual evaluation function when there is no fault. Figure 3 The residual evaluation results show that the present invention can The alarm is always sounded, that is, the fault is detected quickly 0.22s after the fault occurs; Figure 4 It can be seen that the present invention can The alarm is always issued, that is, the fault is detected 6.7s after the fault occurs, and the ideal detection effect is also achieved. Figure 3 and Figure 4 The detection effect proves that the fault detection method proposed in the present invention can effectively detect the fault of the high-speed train traction motor speed sensor under the influence of multi-source interference, and has high real-time performance. At the same time, the detection accuracy of the present invention for constant deviation faults reaches 99.920%, and the detection accuracy for gradual faults reaches 93.761%; the fault detection accuracy results show that the fault detection method proposed in the present invention has high accuracy. Among them, the detection accuracy Calculated by the following formula:

[0174]

[0175] Where: is the total number of samples, is the number of false positive samples, is the number of samples that were missed.

[0176] Although the present invention has been described in detail by referring to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for detecting a fault of a high-speed train traction motor, characterized in that: The steps include: S1. Based on the dynamic mathematical model of the traction motor and the sensor measurement equation, a state space model of the traction motor is established in continuous time; S2. Based on the traction motor state space model established in step S1, multi-source interference and speed sensor fault are introduced to construct a traction motor fault model reflecting the actual fault condition; S3, discretizing the traction motor fault model established in step S2 to establish a discrete fault model of the traction motor under discrete time; S4, select s fuzzy rule antecedent variables, , , , ,Establish IF THEN fuzzy rules, fuzzy linearize the discrete fault model of the traction motor obtained in step S3, and establish a local TS fuzzy model of the traction motor; S5, performing weighted average defuzzification on the local TS fuzzy model to establish a global TS fuzzy model of the traction motor; S6, constructing a residual generator for the global TS fuzzy model established in step S5 by using a fault detection filter method; S7, construct performance indicators reflecting the robustness of the residual to multi-source interference and initial estimation error and the sensitivity to faults; S8, adopting the Hi / H∞ optimized fault detection filter method, transforming the design problem of the residual generator into solving the gain matrix of the fault detection filter; S9, constructing a residual evaluation function within the sliding time window and determining a threshold; S10, comparing the residual evaluation function and the threshold according to the fault detection logic to determine whether a fault occurs; In step S1, a state space model of the traction motor is established in continuous time, which is described as follows: Where: is continuous time; is the state vector; is the input vector; is the measurement output vector; , They are the state matrix and input matrix of the continuous fault model of the traction motor, describing the motion form of the traction motor; is the measurement output matrix, describing the measurement output form of the traction motor; , , Determined by the dynamic equation of the traction motor and the output form of the sensor, the specific parameters are determined by the traction motor model; In step S2, a traction motor fault model constructed by introducing multi-source interference and speed sensor fault is described as follows: Where: , They are the stresses that the traction motor is subjected to under complex operating conditions. The norm-bounded multi-source interference, the former is the system noise, and the latter is the measurement noise; for Norm-bounded speed sensor fault vector; The process noise input matrix of the continuous fault model of the traction motor is determined by the form of multi-source interference; is the sensor fault input matrix, which is determined by the form of the fault; In step S3, the traction motor fault model is discretized to establish a traction motor discrete fault model, which is described as follows: Where: is discrete time, is the discretized sampling period; ， ， 。 2. The method for detecting a fault of a high-speed train traction motor according to claim 1, characterized in that: In step S4, s fuzzy rule antecedent variables are selected, where: each fuzzy rule has the same form but is independent of each other. Fuzzy rules , described as follows: Where: The guiding statement is the antecedent of the fuzzy rule, which represents the condition of each rule; The guided statement is the conclusion of the fuzzy rule, which means the nonlinear model established in step S3 is obtained after fuzzy linearization. Locally Linearized Subsystem Model, superscript , which represents the fuzzy linearization of the nonlinear fault model The first of the fuzzy rules strip; Antecedent variable In its field The membership function on , Domain The minimum value that can be obtained, Domain The maximum value that can be obtained, subscript Corresponds to The first of the antecedent variables indivual; , , , , Respectively Limited Subsystem The state matrix, input matrix, process noise input matrix, measurement output matrix, and fault input matrix.

3. The method for detecting a fault of a high-speed train traction motor according to claim 2, characterized in that: In step S5, a global TS fuzzy model of the traction motor is established, which is described as follows: Where: , , , , They are the global TS fuzzy models obtained after weighted defuzzification. The state matrix, input matrix, process noise input matrix, measurement output matrix, and fault input matrix are calculated by the following equations: ， ， ， Where: For the A normalized fuzzy membership function is calculated as follows: Where: Antecedent variable In its field For fuzzy subsets The membership function of is calculated as follows: Where: is a time-varying binary parameter, that is, each antecedent variable Corresponding to two different fuzzy subsets and .

4. The method for detecting a fault of a high-speed train traction motor according to claim 3, characterized in that: In step S6, a residual generator is constructed using a fault detection filter method, which is described as follows: Where: is the initial state estimate, for State estimation at the moment; for The output estimation error at the moment is the new information; is the generated residual signal; , They are the observer gain matrix and post-filter parameter matrix to be designed respectively; Then define the state estimation error variable: The error dynamic system is described as: Where: is the initial estimation error, , , .

5. The method for detecting a fault of a high-speed train traction motor according to claim 4, characterized in that: In step S7, a performance index reflecting the robustness of the residual to multi-source interference and initial estimation error and the sensitivity to fault is constructed, including the following steps: S71, using H∞ norm As a performance indicator reflecting the robustness of the residual to multi-source interference and initial estimation error, it is described as follows: Where: is the residual signal in the absence of fault; The maximum value of a; superscript is the transpose symbol; is the continuous summation symbol, for ; Given by: S72, using H∞ norm and H-index As the best and worst case performance indicators reflecting the sensitivity of residual to faults, they are described as follows: Where: is the residual signal without multi-source interference and zero initial estimation error; is the maximum value of a under condition b; is the minimum value of a under condition b.

6. The method for detecting a fault of a high-speed train traction motor according to claim 1, characterized in that: In step S8, the design problem of the residual generator is transformed into solving the gain matrix of the fault detection filter, then the observer gain matrix and the post-filter parameter matrix : or Where: is the maximum value of a under condition b; The process noise and measurement noise from multiple sources are expanded into the following form: The process noise input matrix and the measurement noise input matrix are expanded into the following forms: In the formula, Represents the identity matrix, with an order of 4; The residual generator is designed by using the Hi / H∞ optimized fault detection filter method. Then the observer gain matrix is and the post-filter parameter matrix Calculated recursively by the following equation: Where: is the estimated error covariance needle, For new information The Gram matrix.

7. The method for detecting a fault of a high-speed train traction motor according to claim 6, characterized in that: In step S9 and step S10, the residual evaluation function constructed within the sliding time window n is described as follows: The threshold is the tolerance of unknown disturbances and uncertainties in the operation of a fault-free system, so the threshold is defined as follows: Where: is the residual evaluation function of the fault-free system; According to the fault detection logic, the residual evaluation function and the threshold are compared to determine whether a fault has occurred: When the evaluation function exceeds the set threshold, it indicates that a fault has occurred and an alarm is triggered; When the evaluation function does not exceed the set threshold, it indicates that no fault occurs and no alarm is triggered.

Images