阅读说明：本技术 一种基于数据驱动的高速列车悬挂系统早期故障检测方法 (Early fault detection method for high-speed train suspension system based on data driving ) 是由 吴云凯 周扬 苏宇 于 2021-07-02 设计创作，主要内容包括：一种基于数据驱动的高速列车悬挂系统早期故障检测方法,步骤是：1)利用位移传感器、陀螺仪获取高速列车运行时的垂向位移、角速度信号及转向架构架质心的垂向位移信号；2)依据高速列车悬挂系统的离散空间状态方程进行数据建模,得到隐含悬挂系统的输入输出关系的输入输出数据模型；3)利用所得输入输出数据模型,设计基于数据驱动的故障检测残差,构造评价指标；4)实时获取系统输出,更新输入输出数据矩阵,依据新的输入输出数据矩阵更新构造的评价指标；5)判断更新的评价指标是否超出设定阈值,当评价指标J超过阈值,则给出故障报警。本发明在悬挂系统发生早期故障时能实现实时的检测报警,对悬挂系统的早期故障具有较高的灵敏度。(A method for detecting early faults of a high-speed train suspension system based on data driving comprises the following steps: 1) acquiring a vertical displacement and an angular velocity signal of the high-speed train in operation and a vertical displacement signal of the bogie frame mass center by using a displacement sensor and a gyroscope; 2) performing data modeling according to a discrete space state equation of the suspension system of the high-speed train to obtain an input and output data model which implies the input and output relation of the suspension system; 3) designing a fault detection residual error based on data driving by using the obtained input and output data model, and constructing an evaluation index; 4) acquiring system output in real time, updating an input and output data matrix, and updating a constructed evaluation index according to the new input and output data matrix; 5) and judging whether the updated evaluation index exceeds a set threshold value, and giving a fault alarm when the evaluation index J exceeds the threshold value. The invention can realize real-time detection and alarm when the suspension system has early failure, and has higher sensitivity to the early failure of the suspension system.)

1. A method for detecting early failure of a suspension system of a high-speed train based on data driving is characterized by comprising the following steps:

acquiring a vertical displacement of a train carriage mass center, an angular velocity signal and a vertical displacement signal of a bogie frame mass center when a high-speed train runs by using a displacement sensor and a gyroscope, wherein the signals of the sensor and the gyroscope are system output;

step two, carrying out data modeling according to a discrete space state equation of the high-speed train suspension system to obtain a hidden suspensionA suspension system input and output data model of the input and output relation of the suspension system; constructing an input/output data matrix according to the data model by the data obtained in the step oneAnd Y_k,l；

Thirdly, designing a fault detection residual error based on data driving by using the input and output data model obtained in the second step, and constructing an evaluation index J;

step four, acquiring the system output in the step one in real time, updating the input and output data matrix in the step two, and then updating the evaluation index J constructed in the step three according to the new input and output data matrix;

and step five, judging whether the evaluation index J updated in the step four exceeds a set threshold value in real time, and giving a fault alarm when the evaluation index J exceeds the threshold value:

the alarm threshold value of the evaluation index J is J_thThe fault detection mechanism is obtained by historical data calculation and comprises the following steps:

2. the method for detecting the early failure of the suspension system of the high-speed train based on the data driving as claimed in claim 1, wherein in the step two, the method for modeling the input and output data model comprises the following specific steps and contents:

step 1, according to the train suspension system operation mechanism, a discrete system model is expressed as follows:

in the formula, A, B, D, C, E and F are coefficient matrixes corresponding to a space state equation respectively; x is a system state and comprises vertical displacement of the vehicle body and the bogie and pitch angles of the vehicle body and the bogie; u is a system input, including a control output quantity of an actuator in the active suspension; d is track disturbance excitation; f characterizing all possible faults; y is the output signal of the system, including the vertical displacement of the vehicle body and the bogie and the pitch angle of the vehicle body and the bogie; w, v are process noise and measurement noise, respectively;

step 2, when the train suspension system has no fault, obtaining a data model of the high-speed train suspension system which implies the input-output relation of the train suspension system from the train suspension system discrete system model as follows:

in the formula, the raw materials are mixed,is a matrix compounded by A, B, D, C, E and F,respectively a disturbance data matrix and a noise data matrix; input data matrixAnd the output data matrix Y_k,lMeasurable/known in practical application, the form is as follows:

wherein u (k), y (k) represent input and output vectors of the system at time k.

3. The method for detecting the early failure of the suspension system of the high-speed train based on the data driving as claimed in claim 1, wherein in the third step, the method for designing the failure detection residual error based on the data driving comprises the following design steps and contents:

step 1, introducing an orthogonal projection matrix

Step 2, in the step two, establishing an orthogonal projection matrix for simultaneously right-riding on two sides of an input and output data model of the high-speed train suspension systemSince the noise is independent of the system input,the following residuals based on the input and output data are obtained:

4. the method for detecting the early failure of the suspension system of the high-speed train based on the data driving as claimed in claim 1, wherein in the third step, the method for constructing the evaluation index J comprises the following steps:

step 1, in order to eliminate the influence of system noise on residual error quantity, introducing an auxiliary variable z independent from the noise_kIs provided with

Step 2, constructing a symmetric matrix

And 3, selecting the evaluation index J ═ tr (R).

5. The method for detecting the early failure of the suspension system of the high-speed train based on the data driving as claimed in claim 1, wherein in the fourth step, the method for updating the input-output data matrix comprises the following updating steps:

step 1, collecting online data

Step 2, updating the data matrix by adding a new data column to the last column of the data matrix and deleting the first column of the data matrixAnd Z_k：

6. The method for detecting the early failure of the suspension system of the high-speed train based on the data driving as claimed in claim 1, wherein in the fourth step, the method for updating the evaluation index comprises the following steps:

step 1, inputting the updated input data matrixThe following calculations were made:

step 2, updating the orthogonal projection matrix:

and 3, updating the evaluation index according to the evaluation index construction method in the step three by using the updated data matrix and the orthogonal projection matrix.

7. The method for detecting the early failure of the suspension system of the high-speed train based on the data driving as claimed in claim 1, wherein in the fifth step, the method for calculating the failure detection threshold value comprises the following steps:

J_th＝max{J_i}

wherein J_iThe method is an evaluation index of the suspension system at different moments when the suspension system has no fault.

Technical Field

The invention discloses a data-driven early fault detection method for a high-speed train suspension system, and belongs to the technical field of high-speed train suspension system fault detection.

Background

The high-speed train suspension system plays a role in supporting a train body and a bogie, simultaneously isolates wheel-rail acting force caused by track irregularity, and ensures the stability and safety of a train in high-speed operation, so that the high-speed train suspension system has high reliability requirements. Since the opening of a CRH (China Railway High-speed) motor train unit train with a speed of more than 200 kilometers per hour on a main trunk line in 2008, China has built a highway network with the largest scale and the fastest operation speed all over the world through the development of more than 10 years. The method has important significance for improving the running safety of high-speed rails in China and aiming at the research of early fault detection of the suspension system.

The high-speed train suspension system comprises an active suspension and a semi-active suspension and adopts a closed-loop control structure. The suspension system adopts a primary suspension device which is arranged between the axle box and the bogie frame, and a secondary suspension device which is arranged between the bogie frame and the vehicle body. Contains a large number of components including coil springs, lateral/vertical dampers, air springs, active actuators and sensors. The active actuator is used as an important actuator part and is important for the safe running and riding comfort of a high-speed train; active actuators calculate the active control force required by the active suspension system from the vehicle output signal measured by the sensor, and therefore the sensor is also of high importance.

Suspension system failures can be classified as actuator failures, sensor failures, and mechanical component failures. As the train operation time increases, some parts in the suspension system, such as the coil spring, the shock absorber, the air spring, the active/semi-active actuator and the sensor, will have a certain performance degradation, which may induce early failures such as slight crack of the coil spring, slight oil leakage/leakage of the shock absorber, slight air leakage of the air spring, small-amplitude loss of the actuation performance of the actuator, and deviation drift of the sensor, and thus bring a potential danger to the train operation safety.

Disclosure of Invention

The invention aims to solve the technical problems and defects of the prior art and provide a method for detecting early faults of a suspension system of a high-speed train based on data driving.

The method consists of an off-line design part and an on-line detection part. In the off-line design part, the method mainly comprises the steps of constructing a residual signal by utilizing the collected fault-free data, establishing an evaluation index system and estimating a threshold value according to training data. The on-line detection part updates the orthogonal matrix by using the on-line measurement value, and then compares the calculated evaluation index with a threshold value to make a fault judgment.

In order to achieve the purpose, the invention adopts the following technical scheme to realize the purpose:

a method for detecting early faults of a high-speed train suspension system based on data driving comprises the following steps:

acquiring a vertical displacement of a train carriage mass center, an angular velocity signal and a vertical displacement signal of a bogie frame mass center when a high-speed train runs by using a displacement sensor and a gyroscope, wherein the signals of the sensor and the gyroscope are system output;

step two, carrying out data modeling according to a discrete space state equation of the suspension system of the high-speed train to obtain a suspension system input and output data model which implies the input and output relation of the suspension system; constructing an input/output data matrix according to the data model by the data obtained in the step oneAnd Y_k,l；

Thirdly, designing a fault detection residual error based on data driving by using the input and output data model obtained in the second step, and constructing an evaluation index J;

step four, acquiring the system output in the step one in real time, updating the input and output data matrix in the step two, and then updating the evaluation index constructed in the step three according to the new input and output data matrix;

and step five, judging whether the evaluation index updated in the step four exceeds a set threshold value in real time, and giving a fault alarm when the evaluation index J exceeds the threshold value:

the alarm threshold value of the evaluation index J is J_thThe fault detection mechanism is obtained by historical data calculation and comprises the following steps:

further preferably, in step two, the method for modeling the input/output data model specifically comprises the following steps and contents:

step 1, according to the train suspension system operation mechanism, a discrete system model can be expressed as follows:

a, B, D, C, E and F are coefficient matrixes corresponding to a space state equation respectively; x is a system state and comprises vertical displacement of the vehicle body and the bogie and pitch angles of the vehicle body and the bogie; u is a system input, including a control output quantity of an actuator in the active suspension; d is track disturbance excitation; f characterizing all possible faults; y is the output signal of the system, including the vertical displacement of the vehicle body and the bogie and the pitch angle of the vehicle body and the bogie; w, v are process noise and measurement noise, respectively.

And 2, when the system has no fault, obtaining the following data model of the high-speed train suspension system which implies the input-output relationship of the suspension system according to the discrete state space equation established in the step 1:

wherein a data matrix is inputAnd the output data matrix Y_k,lKnown in practical application.

Further preferably, in step three, the method for designing the fault detection residual based on the data driving includes the following steps:

step 1, introducing an orthogonal projection matrix

Step 2, establishing an orthogonal projection matrix for simultaneously right-riding on two sides of an input and output data model of the high-speed train suspension system in the step twoDue to the fact thatAndare independent of each other and can be used for,the following residuals based on the input and output data can be obtained:

further preferably, the method for constructing the evaluation index J in step three comprises the steps of:

step 1, in order to eliminate the influence of system noise on residual error quantity, introducing an auxiliary variable z independent from the noise_kIs provided with

Step 2, constructing a symmetric matrix

And 3, selecting the evaluation index J ═ tr (R).

Further preferably, in step four, the method for updating the input/output data matrix includes the following updating steps:

step 1,Collecting online data

Step 2, updating the data matrix by adding a new data column to the last column of the data matrix and deleting the first column of the data matrixAnd Z_k：

As a further preferable aspect of the present invention, the method for updating an evaluation index in the fourth step includes the steps of:

step 1, inputting the updated input data matrixThe following calculations were made:

step 2, updating the orthogonal projection matrix:

and 3, updating the evaluation index according to the evaluation index construction method in the step three by using the updated data matrix and the orthogonal projection matrix.

Further preferably, in step five, the method for calculating the fault detection threshold value includes:

J_th＝max{J_i}

wherein J_iThe method is an evaluation index of the suspension system at different moments when the suspension system has no fault.

The invention has the following technical effects:

the invention aims at the suspension system of the high-speed train closed-loop control structure, and takes the active actuator, the sensor and the mechanical part as fault detection objects, and the detection method is very sensitive to the early fault of the suspension system, so that the dynamic characteristic of the fault can be detected in real time and accurately estimated when the suspension system has a small fault. The method can be used for executing mechanism fault analysis and system reliability analysis based on the Simpack-Matlab/Simulink simulation architecture.

Drawings

FIG. 1 is a flow chart of a fault detection method of the present invention.

Fig. 2 is a block diagram of a high speed train suspension fault detection architecture.

Fig. 3 is a schematic diagram of installation positions of sensors and gyroscopes of the suspension system of the high-speed train.

FIG. 4 is a graph of suspension actuator early fault detection simulation.

FIG. 5 is a graph of suspension sensor early failure detection simulation.

FIG. 6 is a graph of a simulation of early failure detection of a suspended mechanical component.

Detailed Description

The technical scheme of the invention is further explained in detail by combining the attached drawings:

as shown in fig. 1-3, the method for detecting the early failure of the suspension system of the high-speed train based on data driving comprises the following specific steps:

the method comprises the following steps: the method comprises the following steps of establishing an input and output data model of a high-speed train suspension system:

step 1, according to the train suspension system operation mechanism, a discrete system model can be expressed as follows:

a, B, D, C, E and F are coefficient matrixes corresponding to a space state equation respectively; x is a system state and comprises vertical displacement of the vehicle body and the bogie and pitch angles of the vehicle body and the bogie; u is a system input, including a control output quantity of an actuator in the active suspension; d is track disturbance excitation; f characterizing all possible faults; y is the output signal of the system, including the vertical displacement of the vehicle body and the bogie and the pitch angle of the vehicle body and the bogie; w, v are process noise and measurement noise, respectively.

And 2, when the system has no fault, the discrete state space equation established in the step one comprises the following steps:

wherein

Step 3, the expression obtained in the step two contains a state variable x (k), and a stack matrix can be introduced for eliminating the state variable x (k)Obtaining a discrete state space equation established in the step one

Step 4, substituting the expression obtained in the step three into the expression obtained in the step two, wherein

Is simple and easy to obtain

Wherein

Step 5. considering the time interval N, defining a stack matrix Andand simplifying the expression in the fourth step to obtain the following data model of the high-speed train suspension system with the hidden input-output relation of the suspension system:

wherein a data matrix is inputAnd the output data matrix Y_k,lKnown in practical application.

Step two: designing a data-driven fault detection residual error, wherein the specific steps and contents are as follows:

step 1. introduce the orthogonal projection matrix

Step 2, establishing an orthogonal projection matrix for simultaneously right-riding on two sides of an input and output data model of the high-speed train suspension system in step fiveDue to the fact thatAndare independent of each other and can be used for,the following residuals based on the input and output data can be obtained:

step three: the method comprises the following specific steps and contents of constructing an evaluation index:

step 1, in order to eliminate the influence of system noise on residual error quantity, an auxiliary variable z independent of the noise is introduced_kIs provided with

Step 2. orderConstructing a symmetric matrix

And 3, selecting an evaluation index J-tr (R).

Step four: the input and output data matrix is updated by the following specific steps and contents:

step 1. collecting on-line data

Step 2, updating the data matrix by adding a new data column to the last column of the data matrix and deleting the first column of the data matrixAnd Z_k：

Step five: updating the evaluation index, and the specific steps and contents are as follows:

step 1, inputting the updated input data matrixThe following calculations were made:

step 2, updating the orthogonal projection matrix:

and 3, updating the evaluation index according to the evaluation index construction method in the step three by using the updated data matrix and the orthogonal projection matrix.

Step six: the method comprises the following specific steps and contents of calculating a fault detection threshold value:

step 1, calculating n groups of evaluation indexes J without faults_iI is 1,2, L n, n is a sufficiently large integer

Step 2, calculating a fault detection threshold value: j. the design is a square_th＝max{tr(R_i)}

Step seven: and C, carrying out online detection on the fault according to the real-time evaluation index obtained in the fifth step and the fault detection threshold in the sixth step, judging whether the detection index reaches an alarm threshold, and alarming when the detection index exceeds the threshold.

In the event of a failure of the suspension system,

evaluation index

When the fault term is not zero, J is more than J_thI.e. indicating the occurrence of a fault.

The following simulation verification of the method of the present invention,

step 1, setting characteristic information of actuator faults, comprising the following steps: the fault type is the early fault of an actuator with the failure of 1%, the starting time of the fault is 30 seconds, the ending time of the fault is the simulation ending time and the fault size, the fault is injected by software through a fault injection module, and a fault model is established;

step 2, setting the characteristic information of the sensor fault, comprising the following steps: the failure type is the early failure of the sensor in the gradual change drift, and the drift amount is theta_s(t) 0.001 × (0.15+0.02sin (t)) +0.01sin (0.2t)), the fault starting time is 30 seconds, the fault ending time is the simulation ending time and the fault size, and the fault is injected by software through a fault injection module to establish a fault model;

and 3, setting the characteristic information of the mechanical part fault, comprising the following steps: the method comprises the following steps that a spring of a secondary suspension system of the trailer fails by 10%, the fault starting time is 30 seconds, and the fault finishing time is simulation finishing time and fault magnitude, wherein the fault is subjected to software injection through a fault injection module, and a fault model is established;

and 4, adopting SIMPACK and Matlab/Simulink for combined simulation, importing the track-train coupled model established in the SIMPACK into the Matlab/Simulink, and establishing a train vertical suspension system control simulation model in the Simulink. The running speed of the vehicle is set to be 250Km/h, the simulation time is often 34 seconds, 1700 groups of sample data are generated in total, and the fault occurrence time 30s is the corresponding moment of the 1000 th sampling point.

As shown in fig. 4, when the actuator of the suspension system of the high-speed train fails at the early stage of failure of 1% at the 30 th s, the method proposed by the patent can well detect the early failure of the actuator.

As shown in fig. 5, when the sensor of the suspension system of the high-speed train has the early-stage fault of the gradual drift at the 30 th time, the method proposed by the patent can well detect the early-stage fault of the gradual drift of the sensor.

As shown in fig. 6, when the mechanical part of the suspension system of the high-speed train (taking a secondary spring as an example) fails 10% at the 30 th time, the method proposed by the patent can well detect the occurrence of the early failure of the mechanical part of the suspension system of the train.

The fault detection method has higher sensitivity to the early fault of the suspension system of the high-speed train, can effectively realize the detection of the early fault of the suspension system of the high-speed train, effectively solves the detection of the early fault under a closed-loop control structure and the practical engineering problem thereof, and has important significance for early warning and real-time monitoring of the suspension fault of the high-speed train.

The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.