A Motor Fault Diagnosis Method for a Lorenz-like Stochastic Resonance System Based on Particle Swarm

Abstract

The present invention proposes a method for diagnosing motor faults based on a particle swarm-based Lorenz-like stochastic resonance system, which includes collecting stator current data of a motor under different fault types by using a current sensor as the original signal; inputting the original signal data into the particle swarm-based Lorenz-like stochastic resonance system, taking the signal-to-noise ratio as the objective function, adaptively adjusting the parameters of the Lorenz-like system, and outputting the data with the optimal signal-to-noise ratio under the optimal parameters; and inputting the data into a well-trained extreme learning machine for fault diagnosis. The Lorenz-like stochastic resonance system proposed by the present invention has advantages such as multiple adjustable parameters and strong plasticity, providing a large adjustment space for the system to adaptively adjust parameters for different types of fault signals. The optimal signal-to-noise ratio data output by the system can highlight the signal characteristics, and thus can accurately perform fault diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent diagnosis of motor faults, and specifically provides a method for diagnosing motor faults based on a particle swarm-based Lorenz-like stochastic resonance system. Background Art

[0002] Motors are one of the most important achievements in the modern energy conversion industry and are widely used in fields such as medical and health, transportation, and information communication. While motors provide fast services, comfort, and strong guarantees for humans, due to the limitations of the lifespan of their own materials, performance degradation, manufacturing defects, pollution, or being affected by human factors and the external environment during operation, their final performance will degenerate or even lead to failure, which may endanger human lives and cause serious economic losses, such as in medicine, military operations, or transportation. Although the occurrence of motor faults cannot be completely avoided, if the fault type can be detected and diagnosed at an early stage and a response can be made accurately, quickly, and in a timely manner, serious accidents can be effectively avoided and the stable operation of equipment and systems can be maintained.

[0003] Early faults of motors often reflect certain characteristic changes through different forms such as current signal changes. Parameters such as current and temperature that can reflect the operating state of the motor are collected through sensors. However, the data collected by the sensors contains not only useful information reflecting the operating state of the motor but also error information such as measurement errors, transmission errors, and environmental noise. At this time, it is particularly important to analyze the data through signal processing techniques.

[0004] With the rapid development of nonlinear dynamics theory, a method for detecting and processing weak signals based on stochastic resonance has been proposed. Stochastic resonance is a phenomenon in a nonlinear system where noise is used to enhance weak signals. The stochastic resonance method has the following significant characteristics compared with current methods: different weak signal detection mechanisms, the ability to detect signals with lower signal-to-noise ratios, and fast and real-time applications. Summary of the Invention

[0005] Based on the above background problems, the present invention provides a method for diagnosing motor faults based on a particle swarm-based Lorenz-like stochastic resonance system. The parameters of a Lorenz-like stochastic resonance system with strong plasticity and many adjustable parameters are optimized through a particle swarm algorithm to ensure that fault signals with different characteristics can output signals with the best signal-to-noise ratio under different optimal parameters. The output signal data is filled with fault diagnosis labels and then input into an extreme learning machine for fault diagnosis.

[0006] The specific technical solution of the present invention is as follows:

[0007] A method for diagnosing motor faults based on a particle swarm-based Lorenz-like stochastic resonance system, characterized in that the method comprises the following steps:

[0008] S1. Collect the stator current data of the motor under different fault types using a current sensor as the input signal;

[0009] S2. Input the data of the input signal into the Lorenz-like stochastic resonance system based on particle swarm optimization. Using the signal-to-noise ratio as the objective function, search for the optimal parameters within the parameter range of the Lorenz-like stochastic resonance system and output the signal with the optimal signal-to-noise ratio under the optimal parameters;

[0010] S3. Combine the data of the signals of different fault types output in step S2, and fill different diagnostic labels for different faults to form a data set;

[0011] S4. Randomly divide the data set into a training set and a test set. First, input the data of the training set into the extreme learning machine for fault diagnosis learning, and then input the data of the test set into the learned extreme learning machine for fault diagnosis;

[0012] S5. Integrate the data processing of the Lorenz-like stochastic resonance system based on particle swarm optimization and the diagnosis of the extreme learning machine algorithm. After inputting the newly collected data into the integrated system, obtain the fault type corresponding to the input data.

[0013] A method for diagnosing motor faults based on a Lorenz-like stochastic resonance system with particle swarm optimization, characterized in that the Lorenz-like stochastic resonance system in step S2 is:

[0014] (1)

[0015] where is the system parameter, is the state variable, is the input signal, composed of the signal and noise .

[0016] A method for diagnosing motor faults based on a Lorenz-like stochastic resonance system with particle swarm optimization, characterized in that the method for determining the parameter range of the Lorenz-like stochastic resonance system in step S2 is:

[0017] When the Lorenz-like stochastic resonance system is at the equilibrium point, there is always , so the system equilibrium curve equation can be obtained as

[0018] (2)

[0019] Then the poles of the equilibrium curve are ; when the system is at the equilibrium point, there is always established, the curves and The intersection value is the equilibrium point of the system; the occurrence of stochastic resonance is manifested as the input signal driving the system to transition between steady states; therefore, the amplitude of the pole of the equilibrium curve affected by the parameters should be less than the amplitude of the input signal, that is, it satisfies The system parameters constitute a set ;

[0020] The class Lorenz stochastic resonance system is developed from the Lorenz chaotic system. Only under appropriate initial values and parameters, when the Lorenz chaotic system is a fixed point or a periodic state, the class Lorenz system will generate stochastic resonance; therefore, the Lyapunov exponent spectrum of the chaotic system under different parameters After sorting from large to small, the parameters that satisfy the maximum Lyapunov exponent The set of is ;

[0021] The parameter range of the class Lorenz stochastic resonance system .

[0022] A method for diagnosing motor faults of a class Lorenz stochastic resonance system based on particle swarm, characterized in that the steps of finding the optimal parameters of the class Lorenz stochastic resonance system based on particle swarm in the system parameter range in step S2 are as follows:

[0023] S21. Initialize the system parameters, iteration step size, system initial value, population size and learning rate, iteration times and maximum iteration times, and inertia weight;

[0024] S22. Calculate the signal-to-noise ratio of the output signal of the class Lorenz stochastic resonance system, and find the individual optimal value and the group optimal value;

[0025] S23. Update the particle velocity and position within the parameter range and within the iteration step size range;

[0026] S24. Calculate the signal-to-noise ratio of the output signal of the class Lorenz stochastic resonance system, and update the individual optimal value and the group optimal value;

[0027] S25. When the termination condition is not satisfied, repeat steps S23 and S24. When the termination condition is satisfied, output the signal data with the optimal signal-to-noise ratio under the optimal parameters.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] The class Lorenz stochastic resonance system proposed by the present invention has many adjustable parameters and strong plasticity, providing an adjustment space for adaptively finding the optimal parameters for different types of input signals.

[0030] Based on the particle swarm, adaptively search for the optimal parameters of the Lorenz-like stochastic resonance system for the motor current signals of different fault types. The output signal data with the optimal signal-to-noise ratio can highlight different fault characteristics. When the processed data is used for fault diagnosis, the diagnostic effect is good. Brief Description of the Drawings

[0031] Figure 1 It is a flow chart of a motor fault diagnosis method based on a Lorenz-like stochastic resonance system using particle swarm;

[0032] Figure 2 It is a time-domain diagram of a group of data in different types of motor fault data collected by a current sensor;

[0033] Figure 3 It is a time-domain diagram of the fault signal data output under the optimal parameters of the Lorenz-like stochastic resonance system based on particle swarm;

[0034] Figure 4 It is a diagnostic result diagram of the extreme learning machine after the fault signal is processed by the Lorenz-like stochastic resonance system based on particle swarm;

[0035] Figure 5 It is a diagnostic result diagram of the extreme learning machine without processing the fault signal by the Lorenz-like stochastic resonance system based on particle swarm; Detailed Embodiment

[0036] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and actual experiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0037] As Figure 1 shown in the flow chart of a motor fault diagnosis method based on a Lorenz-like stochastic resonance system using particle swarm of the present invention, it includes the following steps:

[0038] S1. Use a current sensor to collect the stator current data of the motor under different fault types as the input signal;

[0039] S2. Input the data of the input signal into the Lorenz-like stochastic resonance system based on particle swarm. Taking the signal-to-noise ratio as the objective function, search for the optimal parameters within the parameter range of the Lorenz-like stochastic resonance system and output the signal with the optimal signal-to-noise ratio under the optimal parameters;

[0040] S3. Combine the data of the different fault type signals output in step S2, and fill different diagnostic labels for different faults to form a data set;

[0041] S4. Randomly divide the dataset into a training set and a test set. First, input the training set data into the extreme learning machine for fault diagnosis learning, and then input the test set data into the well-trained extreme learning machine for fault diagnosis.

[0042] S5. Integrate the data processing of the particle swarm-based Lorenz-like stochastic resonance system and the diagnosis algorithm of the extreme learning machine. After inputting the newly collected data into the integrated system, obtain the fault type corresponding to the input data.

[0043] Furthermore, the data in step S1 is collected through an asynchronous motor experimental platform. The specific experimental motor is a YSP90L-4 type asynchronous motor, and its main parameters are: rated power 1.5 kW, rated speed 1400 r / min, rated voltage 380 V, number of pole pairs 2, rated frequency 50 Hz, connection method Y, power factor 0.79, and ambient temperature 40 o C.

[0044] Furthermore, different fault types in step S1 are simulated by continuously replacing the motor model to obtain different faulty motors, and the current acquisition experiments under various faults are completed. The fault types collected in the experiment include: stator winding fault (SWF), air gap eccentricity (AGE), no fault (NORMAL), bearing failure (BF), shaft bending (SB), and rotor broken bar (RBB). The collected current data is 500 groups with 2000 continuous sampling points extracted for each type of fault.

[0045] Furthermore, the time-domain diagrams of 1 group of data containing 2000 continuous sampling points in different types of motor fault data in step S1 are as Figure 2 shown, Figure 2 (a) is 1 group of data containing 2000 continuous sampling points in the stator winding fault, Figure 2 (b) is 1 group of data containing 2000 continuous sampling points in the air gap eccentricity fault, Figure 2 (c) is 1 group of data containing 2000 continuous sampling points in the no-fault condition, Figure 2 (d) is 1 group of data containing 2000 continuous sampling points in the bearing failure, Figure 2 (e) is 1 group of data containing 2000 continuous sampling points in the shaft bending fault, Figure 2 (f) is 1 group of data containing 2000 continuous sampling points in the rotor broken bar fault.

[0046] Further, in the Lorenz-like stochastic resonance system based on particle swarm in step S2, the Lorenz-like stochastic resonance system is as follows:

[0047] (3)

[0048] where are system parameters, are state variables, is the input signal, which is composed of the signal and noise .

[0049] Further, in step S2, the objective function is determined by the signal-to-noise ratio of the system output, specifically: (4)

[0050] where, is the data length of the input signal, is the one-sided spectral amplitude of the input signal corresponding to the characteristic frequency.

[0051] Further, the method for determining the parameter range of the Lorenz-like stochastic resonance system in step S2 is as follows:

[0052] When the Lorenz-like stochastic resonance system is at the equilibrium point, there is always , so the system equilibrium curve equation can be obtained as

[0053] (3)

[0054] Then the poles of the equilibrium curve are ; when the system is at the equilibrium point, there is always established, and the intersection value of the curves and is the equilibrium point of the system; the occurrence of stochastic resonance is manifested as the input signal driving the system to transition between steady states; therefore, the amplitude of the poles of the equilibrium curve affected by the parameters should be less than the amplitude of the input signal, that is, the system parameters satisfying constitute the set ;

[0055] The Lorenz-like stochastic resonance system is developed from the Lorenz chaotic system. Only with appropriate initial values and parameters can the Lorenz chaotic system be a fixed point or a periodic state, and then the Lorenz-like system can produce stochastic resonance; therefore, after sorting the Lyapunov exponent spectra of the chaotic system under different parameters from largest to smallest, the set of parameters satisfying the maximum Lyapunov exponent is ;

[0056] Parameter range of the Lorenz-like stochastic resonance system 。

[0057] Furthermore, the parameter settings of the Lorenz-like stochastic resonance system based on the particle swarm in step S2 are shown in the following table:

[0058]

[0059] Furthermore, in step S2, the steps for the Lorenz-like stochastic resonance system based on the particle swarm to find the optimal parameters within the system parameter range are as follows:

[0060] S21. Initialize the system parameters, iteration step size, system initial value, population size, learning rate, number of iterations and maximum number of iterations, and inertia weight;

[0061] S22. Calculate the signal-to-noise ratio of the output signal of the Lorenz-like stochastic resonance system, and find the individual optimal value and the global optimal value;

[0062] S23. Update the particle velocity and position within the parameter range and within the iteration step size range;

[0063] S24. Calculate the signal-to-noise ratio of the output signal of the Lorenz-like stochastic resonance system, and update the individual optimal value and the global optimal value;

[0064] S25. When the termination condition is not met, repeat steps S23 and S24. When the termination condition is met, output the signal data with the optimal signal-to-noise ratio under the optimal parameters.

[0065] Furthermore, after the stator current data collected by the current sensor in step S1 passes through step S2, one set of data of different fault types with the optimal signal-to-noise ratio is output. The time-domain diagram of 2000 consecutive sampling point data is as Figure 3 shown Figure 3 (a) is a diagram of one set of 2000 consecutive sampling point data in the processed stator winding fault data, Figure 3 after processing, (b) is a diagram of one set of 2000 consecutive sampling point data in the air-gap eccentricity fault data, Figure 3 after processing, (c) is a diagram of one set of 2000 consecutive sampling point data in the fault-free data, Figure 3 after processing, (d) is a diagram of one set of 2000 consecutive sampling point data in the bearing fault data, Figure 3 after processing, (e) is a diagram of one set of 2000 consecutive sampling point data in the shaft bending fault data, Figure 3 after processing, (f) is a diagram of one set of 2000 consecutive sampling point data in the rotor bar breaking fault data.

[0066] Further, in step S3, the data of different fault type signals output in step S2 are combined, and different diagnostic labels are filled for different faults. The specific labels are shown in the following table:

[0067]

[0068] Further, in step S4, the way to randomly divide the data set into a training set and a test set is as follows:

[0069] Randomly select 80% of the samples in the data set as the training set of the algorithm, and 20% of the samples as the test set.

[0070] Further, the diagnosis result graph of the extreme learning machine after the fault signal is processed by the particle swarm optimization-based Lorenz-like stochastic resonance system in step S4 is as Figure 4 shown; Figure 4 (a) is the comparison graph of the diagnosed fault categories and the true fault categories in one extreme learning machine diagnosis experiment among 20 repeated experiments, Figure 4 (b) is the diagnosis situation graph of each fault category; it can be seen from the figure that after being processed by the particle swarm optimization-based Lorenz-like stochastic resonance system, the average correct rate of the extreme learning machine for fault diagnosis is 100%.

[0071] Further, the comparison graph of the diagnosis results of the extreme learning machine when the fault signal is processed by the particle swarm optimization-based Lorenz-like stochastic resonance system and when the fault signal is not processed by the particle swarm optimization-based Lorenz-like stochastic resonance system in step S4 is as Figure 5 shown; Figure 5 (a) is the comparison graph of the diagnosed fault categories and the true fault categories in one extreme learning machine diagnosis experiment among 20 repeated experiments, Figure 5 (b) is the diagnosis situation graph of each fault category; it can be seen from the figure that the average correct rate of the extreme learning machine without being processed by the particle swarm optimization-based Lorenz-like stochastic resonance system is 97.316%.

[0072] Further, by comparing Figure 4 with Figure 5 it can be known that the diagnosis result of the extreme learning machine processed by the particle swarm optimization-based Lorenz-like stochastic resonance system is better than that of the extreme learning machine not processed by the particle swarm optimization-based Lorenz-like stochastic resonance system, and the correct rate of fault diagnosis is improved, indicating that a motor fault diagnosis method based on a particle swarm optimization-based Loren-like stochastic system has good diagnosis effect.

Claims

1. A motor fault diagnosis method for a Lorenz-like stochastic resonance system based on particle swarm, characterized in that, The method includes the following steps: S1. Use a current sensor to collect the stator current data under different motor fault types as input signals; S2. Input the data of the input signal into the particle swarm-based Lorenz-like stochastic resonance system, use the signal-to-noise ratio as the objective function, search for the optimal parameters within the parameter range of the Lorenz-like stochastic resonance system, and output the signal with the optimal signal-to-noise ratio under the optimal parameters; the Lorenz-like stochastic resonance system is as follows: Within, find the optimal parameters and output the signal with the optimal signal-to-noise ratio under the optimal parameters; the Lorenz-like stochastic resonance system is: ， wherein is a system parameter, is a state variable, is an input signal, which consists of a signal and noise ; S3. Combine the data of different fault type signals output in step S2, and fill different diagnostic labels for different faults to form a data set; S4. Randomly divide the data set into a training set and a test set. First, input the training set data into an extreme learning machine for fault diagnosis learning, and then input the test set data into the learned extreme learning machine for fault diagnosis; S5. Integrate the data processing of a particle swarm-based Lorenz-like stochastic resonance system and the diagnosis of an extreme learning machine algorithm. After inputting the newly collected data into the integrated system, obtain the fault type corresponding to the input data.

2. The motor fault diagnosis method of a Lorenz-like stochastic resonance system based on particle swarm according to claim 1, characterized in that, Parameter range of the class Lorenz stochastic resonance system in step S2 The determination method is as follows: When the class Lorenz stochastic resonance system is at the equilibrium point, there is always , so the equilibrium curve equation of the system can be obtained as , then the poles of the equilibrium curve are ; when the system is at the equilibrium point, there is always holds, and the intersection value of the curves and is the equilibrium point of the system; the occurrence of stochastic resonance is manifested as the input signal driving the system to transition between steady states; therefore, the amplitude of the poles of the equilibrium curve affected by the parameters should be less than the amplitude of the input signal, that is, the system parameters that satisfy form a set ; The class Lorenz stochastic resonance system is developed from the Lorenz chaotic system. Only when the Lorenz chaotic system is in a fixed point or periodic state under appropriate initial values and parameters can the class Lorenz system generate stochastic resonance. Therefore, the Lyapunov exponent spectra of chaotic systems under different parameters After being sorted from large to small, the parameters that satisfy the maximum Lyapunov exponent form a set of which is ; Parameter range of the Lorenz-like stochastic resonance system .

3. The motor fault diagnosis method of a Lorenz-like stochastic resonance system based on particle swarm according to claim 1, characterized in that, In step S2, the steps for the particle swarm-based Lorenz-like stochastic resonance system to find the optimal parameters within the system parameter range are as follows: S21. Initialize system parameters, iteration step size, system initial value, population size, learning rate, number of iterations and maximum number of iterations, inertia weight; S22. Calculate the signal-to-noise ratio of the output signal of the Lorenz-like stochastic resonance system, and find the individual optimal value and the group optimal value; S23. Update the particle velocity and position within the parameter range and within the iteration step size range; S24. Calculate the signal-to-noise ratio of the output signal of the Lorenz-like stochastic resonance system, and update the individual optimal value and the group optimal value; S25. When the termination condition is not satisfied, repeat steps S23 and S24. When the termination condition is satisfied, output the signal data with the optimal signal-to-noise ratio under the optimal parameters.

Images