Hub motor fault diagnosis method based on RLMD improved by SBOA

Abstract

The hub motor fault diagnosis method based on RLMD improved by SBOA belongs to the field of fault diagnosis based on acoustic vibration signals. It is used to solve the problems of low noise source recognition, high cost and reliance on manual judgment in the existing motor fault diagnosis methods based on acoustic vibration signals, resulting in low accuracy of fault diagnosis results. The present invention can capture the hub motor acoustic vibration signals generated under extreme conditions by designing the acquisition position and parameter settings of the acoustic vibration sensor; at the same time, the optimized robust local mean decomposition algorithm is used to process the hub motor acoustic vibration signals to obtain the feature set of the best IMF components of the hub motor acoustic vibration signals, and finally the enhanced random forest classifier is used to calculate the diagnostic results of the feature set, thereby improving the accuracy of the fault diagnosis results. The present invention improves the efficiency and reliability of early fault diagnosis of hub motors, and provides strong technical support for the subsequent structural optimization design of hub motors.

Description

Technical Field

[0001] The present invention belongs to the field of fault diagnosis based on acoustic vibration signals, and in particular relates to a hub motor fault diagnosis method and system based on robust local mean decomposition technology improved by a meander optimization algorithm. Background Art

[0002] The motor is a common electric device. As a driving device, it provides power for various devices and is widely used in various fields such as household appliances, transportation, aerospace, etc. Take electric vehicles and electric-assisted bicycles as an example. As an emerging means of transportation, their core power is usually provided by hub motors. The quality of hub motors affects the safety and comfort of drivers. Therefore, hub motors must be diagnosed before they leave the factory. Common factors that affect the quality of hub motors include: abnormal friction caused by improper mechanical assembly and rotor imbalance produces mechanical noise; improper magnetic circuit design leads to abnormal electromagnetic noise during operation. These two factors cause abnormal noise during the operation of the motor, which can be used for intelligent diagnosis of motor faults.

[0003] At present, the research and status of related industries in motor fault diagnosis based on acoustic vibration signals have the following problems: order analysis, transfer path analysis, frequency sweep analysis, acoustic holography and other methods require the noise signal to have a high signal-to-noise ratio, are sensitive to environmental changes, and are easily interfered by other noises, thus affecting the accurate identification of noise sources. At the same time, the equipment used is expensive and relies on manual judgment, and the accuracy rate is not high. Summary of the invention

[0004] The present invention is used to solve the problems of low noise source recognition, high cost and reliance on manual judgment in the existing hub motor fault diagnosis method based on acoustic vibration signals, resulting in low accuracy of fault diagnosis results.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] The present invention provides a hub motor fault diagnosis method based on RLMD improved by SBOA, and the fault diagnosis method specifically comprises the following steps:

[0007] Step S1: two vibration sensors are respectively set at both ends of the hub motor shaft, and two acoustic sensors are respectively set at positions 10 cm and 1 m away from the hub motor;

[0008] Step S2: setting parameters for the two vibration sensors and the acoustic sensor respectively;

[0009] Step S3: acquiring data collected by two vibration sensors and two acoustic sensors as acoustic vibration signal data of the wheel hub motor;

[0010] Step S4: determining the optimization parameters of the robust local mean decomposition algorithm according to the acoustic vibration signal data of the wheel hub motor: bandwidth threshold and B-spline order;

[0011] Step S5: optimizing the robust local mean decomposition algorithm, using the optimized robust local mean decomposition algorithm to calculate the acoustic and vibration signal data of the hub motor, obtaining the best IMF component of the acoustic and vibration signal data of the hub motor, and extracting the feature set of the best IMF component;

[0012] Step S6: Input the extracted feature set into the enhanced random forest classifier to obtain the diagnosis result.

[0013] Furthermore, there is a preferred embodiment, the vibration data of the X-axis and Z-axis of the two vibration sensors are used as the vibration signal data of the hub motor;

[0014] The two acoustic sensors must be kept in the same straight line or placed at 90°.

[0015] Furthermore, there is another preferred embodiment, the above step S2 is specifically as follows:

[0016] The X, Y, and Z axis sensitivities of the first vibration sensor of the two vibration sensors are set to 100.5 mV / g, 96.6 mV / g, and 103.5 mV / g, respectively, and the X, Y, and Z axis sensitivities of the second vibration sensor are set to 100.4 mV / g, 88.2 mV / g, and 97 mV / g, respectively;

[0017] The sensitivity of the two acoustic sensors mentioned above was set to 50 mV / g;

[0018] The bandwidth of the two acoustic sensors mentioned above is set to 12800 Hz.

[0019] Furthermore, there is a preferred embodiment, the optimization method of the above-mentioned robust local mean decomposition algorithm is:

[0020] Step 1: Initial setting of the meander optimization algorithm;

[0021] Step 2: The position of each individual in the search space of the meander optimization algorithm is represented as the optimization parameter of the robust local mean decomposition algorithm, thereby obtaining multiple parameter combinations;

[0022] Step 3: Substitute multiple parameter combinations into the robust local mean decomposition algorithm respectively, and determine the optimal parameters using the fitness function of the robust local mean decomposition algorithm;

[0023] Step 4: Determine whether the optimization reaches the termination condition of the meandering optimization algorithm. If so, proceed to the next step; if not, update the population position of the meandering optimization algorithm and return to step 2;

[0024] Step 5: Save the optimal parameter combination and substitute it into the robust local mean decomposition algorithm to obtain the optimized robust local mean decomposition algorithm.

[0025] Furthermore, there is another preferred embodiment, the above step 1 is specifically as follows:

[0026] Set the number of iterations of the meandering optimization algorithm, the population size, and define the parameter optimization range of the robust local mean decomposition algorithm.

[0027] Furthermore, in a preferred embodiment, the above-mentioned multiple parameter combinations are used to perform fitness evaluation on the hub motor sound and vibration signal data.

[0028] Furthermore, there is a preferred embodiment, the method for obtaining the above fitness function is:

[0029] The robust local mean decomposition algorithm is used to decompose the wheel hub motor acoustic and vibration signal data, and the envelope entropy of each IMF component is calculated. The minimum value of the envelope entropy is used as the fitness function.

[0030] Furthermore, there is another preferred embodiment, the above step S5 is specifically as follows:

[0031] Step S51: Calculate the instantaneous amplitude and phase of the wheel hub motor acoustic vibration signal data using an optimized robust local mean decomposition algorithm;

[0032] Step S52: determining the local maximum and local minimum of the instantaneous amplitude;

[0033] Step S53: Calculate the local cutoff frequency according to the local maximum and local minimum of the instantaneous amplitude;

[0034] Step S54: filter the signal at the local cutoff frequency, and judge and iteratively decompose the filtering result until all signals are decomposed to obtain the best IMF component.

[0035] Furthermore, there is another preferred embodiment, the above step S6 is specifically as follows:

[0036] Step S61: setting an initial sample weight distribution, and using the sample weight distribution to train a weak classifier in the random forest classifier, and calculating the classification error rate of the weak classifier;

[0037] Step S62: Calculate the weight of the weak classifier according to the classification error rate, and update the sample weight distribution according to the weight until multiple weak classifiers are trained;

[0038] Step S63: combining all weak classifiers into a strong classifier;

[0039] Step S64: Evaluate the performance of the strong classifier and determine the weights of different weak learners until a final classifier is obtained as an enhanced random forest classifier;

[0040] Step S65: input the extracted feature set into the enhanced random forest classifier to obtain a diagnosis result.

[0041] The present invention also provides a hub motor fault diagnosis system based on SBOA-improved RLMD, the system is used to implement any one of the above-mentioned hub motor fault diagnosis methods based on SBOA-improved RLMD, the diagnosis system includes an acoustic vibration signal acquisition module, a parameter optimization module, a signal decomposition module and a fault diagnosis identification module;

[0042] The sound and vibration signal acquisition module is used to collect the sound and vibration signal data of the hub motor;

[0043] The parameter optimization module is used to determine the optimization parameters of the robust local mean decomposition algorithm according to the sound and vibration signal data of the hub motor, and is also used to optimize the parameters of the robust local mean decomposition algorithm stored in the signal decomposition module by using the meander optimization algorithm;

[0044] The signal decomposition module is used to decompose the wheel hub motor sound and vibration signal data using the optimized robust local mean decomposition algorithm to obtain the feature set of the best IMF component;

[0045] The fault diagnosis and identification module is used to use the enhanced random forest classifier to perform fault diagnosis and identification on the feature set of the best IMF component to obtain the diagnosis result.

[0046] The beneficial effects of the present invention are:

[0047] 1. Aiming at the problem that the early fault signals of the hub motor are weak and it is difficult to directly extract useful features of the fault signal from the original signal. The present invention designs the acquisition position and parameter settings of the acoustic vibration sensor so that the acoustic vibration signal data of the hub motor generated under extreme conditions can be captured; at the same time, the optimized robust local mean decomposition algorithm (RLMD) is used to process the acoustic vibration signal data of the hub motor to obtain the feature set of the best IMF component of the acoustic vibration signal data of the hub motor, and finally the enhanced random forest classifier is used to calculate the diagnostic results of the feature set, thereby improving the accuracy of the fault diagnosis results.

[0048] Furthermore, the present invention designs the collection positions of vibration sensors and acoustic sensors and optimizes the parameter settings of vibration sensors and acoustic sensors, so that signals generated under extreme conditions can be captured, more subtle vibration and sound changes can be extracted, and it can adapt to different operating conditions and environments, thereby improving the accuracy and reliability of fault diagnosis.

[0049] Furthermore, the present invention formulates optimization parameters of the robust local mean decomposition algorithm (RLMD) algorithm based on the characteristics of the wheel hub motor acoustic and vibration signals, including defining the optimization range of parameters such as bandwidth threshold and B-spline order. The formulation of the optimization parameters ensures that the robust local mean decomposition algorithm (RLMD) algorithm can adapt to the specific characteristics of the acoustic and vibration signals, such as nonlinearity and non-stationarity. At the same time, the RLMD algorithm optimized by the meander optimization algorithm (SBOA) can process acoustic and vibration signal data of different magnitudes. The adaptability of the algorithm is reflected in the ability to extract key features from a large amount of acoustic and vibration signal data, which are crucial for the fault diagnosis of the wheel hub motor. Therefore, the optimized robust local mean decomposition algorithm (RLMD) described in the present invention can more accurately identify and diagnose the faults of the wheel hub motor, rather than being used as a general signal processing tool.

[0050] Furthermore, in order to address the problem that early fault signals are weak and difficult to classify correctly, the present invention adopts a reinforcement learning algorithm (AdaBoost) to iteratively focus on samples that are misclassified by the previous classifier, giving these samples greater weights, thereby improving the classification ability of subsequent classifiers for these samples, thereby significantly improving the accuracy of the classifier, so that the wheel hub motor sound and vibration signals can be correctly classified, which is helpful for the extraction of fault signals.

[0051] The present invention further improves the efficiency and reliability of early fault diagnosis of hub motors through advanced data processing and analysis technology, and provides strong technical support for subsequent optimization design of hub motor structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 It is a flow chart of a hub motor fault diagnosis method based on RLMD improved by SBOA according to the present invention;

[0054] Figure 2 is a schematic diagram of the test environment of the present invention;

[0055] Figure 3 is a top view of the arrangement of the sensors in the test environment of the present invention;

[0056] Figure 4 It is a flow chart of processing wheel hub motor acoustic vibration signal data using optimized robust local mean decomposition algorithm according to the present invention;

[0057] Figure 5 It is a flow chart of the enhanced random forest classifier of the present invention processing the optimal IMF component feature set. DETAILED DESCRIPTION

[0058] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0059] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings. The following implementation will help those skilled in the art to further understand the present invention, but will not limit the present invention in any form.

[0060] Implementation method 1, see Figure 1 This embodiment is described. This embodiment aims at the problem that the early fault signal of the hub motor is usually weak, the traditional signal preprocessing method has limitations in accuracy, and the selection of key parameters often depends on artificial prior knowledge. An innovative hub motor fault diagnosis method based on RLMD improved by SBOA is proposed. The fault diagnosis method specifically includes the following steps:

[0061] Step S1: two vibration sensors are respectively set at both ends of the hub motor shaft, and two acoustic sensors are respectively set at positions 10 cm and 1 m away from the hub motor;

[0062] Step S2: setting parameters for the two vibration sensors and the acoustic sensor respectively;

[0063] Step S3: acquiring data collected by two vibration sensors and two acoustic sensors as acoustic vibration signal data of the wheel hub motor;

[0064] Step S4: determining the optimization parameters of the robust local mean decomposition algorithm according to the acoustic vibration signal data of the hub motor: bandwidth threshold and B-spline order;

[0065] Step S5: optimizing the robust local mean decomposition algorithm, using the optimized robust local mean decomposition algorithm to calculate the acoustic and vibration signal data of the hub motor, obtaining the best IMF component of the acoustic and vibration signal data of the hub motor, and extracting the feature set of the best IMF component;

[0066] Step S6: Input the extracted feature set into the enhanced random forest classifier to obtain the diagnosis result.

[0067] In view of the fact that the early fault signals of the hub motor are usually weak, the traditional signal preprocessing method has limitations in accuracy, and the selection of key parameters often depends on artificial prior knowledge. This embodiment proposes an innovative hub motor early fault diagnosis method based on sound and vibration signals. This method designs the collection position of the vibration sensor and the acoustic sensor, and optimizes the parameter settings of the vibration sensor and the acoustic sensor, so that the signals generated under extreme conditions can be captured, and more subtle vibration and sound changes can be extracted. It can adapt to different operating conditions and environments, and improve the accuracy and reliability of fault diagnosis. At the same time, the method also formulates the optimization parameters of the robust local mean decomposition algorithm (RLMD) algorithm according to the characteristics of the hub motor sound and vibration signals, including the optimization range of parameters such as bandwidth threshold and B-spline order. The formulation of the optimization parameters ensures that the robust local mean decomposition algorithm (RLMD) algorithm can adapt to the specific characteristics of the sound and vibration signals, such as nonlinearity and non-stationarity. At the same time, the RLMD algorithm optimized by the meander optimization algorithm (SBOA) can process sound and vibration signal data of different magnitudes.

[0068] like Figure 1 As shown in the figure, the parameter optimization range of the robust local mean decomposition algorithm (RLMD) is determined: the wide threshold α and the B-spline order k, the robust local mean decomposition algorithm (RLMD) is optimized to form an optimized robust local mean decomposition algorithm (RLMD), and the early fault signals of the hub motor are processed by the optimized robust local mean decomposition algorithm (RLMD), and the feature set of the best IMF component of the hub motor sound and vibration signal is determined, thereby significantly improving the accuracy of the original signal decomposition. In addition, the feature set is processed by the enhanced random forest classifier (RF) to obtain the diagnosis result, which can not only show higher accuracy and robustness, but also enable the hub motor sound and vibration signal to be correctly classified, which is helpful for the extraction of fault signals.

[0069] Implementation Method 2: See Figure 2 and Figure 3 This embodiment is described. This embodiment specifically describes the collection position and parameter settings of the vibration sensor and the acoustic sensor described in the first embodiment.

[0070] The vibration data of the two vibration sensors on the X-axis and Z-axis are used as the vibration signal data of the wheel hub motor;

[0071] The two acoustic sensors must be kept in the same straight line or placed at 90°;

[0072] The X, Y, and Z axis sensitivities of the first vibration sensor of the two vibration sensors are set to 100.5 mV / g, 96.6 mV / g, and 103.5 mV / g, respectively, and the X, Y, and Z axis sensitivities of the second vibration sensor are set to 100.4 mV / g, 88.2 mV / g, and 97 mV / g, respectively;

[0073] The sensitivity of both acoustic sensors was set to 50 mV / g;

[0074] The bandwidth of both acoustic sensors was set to 12800 Hz.

[0075] In actual application, when the hub motor has an early fault, the difference between different fault signals is not particularly obvious. If the data collection environment is arranged arbitrarily, the key characteristic information of the fault signal will be missed. If the sensitivity of the sensor is not set correctly, the accuracy of the signal collection process will be too low. Therefore, if Figure 2 and Figure 3 As shown, in this embodiment, two vibration sensors are arranged at both ends of the wheel hub motor shaft, wherein the X-axis direction is horizontal, the Y-axis direction is along the axial direction, and the Z-axis direction is perpendicular to the ground, and the vibration data of the X-axis and Z-axis are used as the vibration signal data of the wheel hub motor. Among them, the X, Y, and Z axis sensitivities of the first vibration sensor of the two vibration sensors are set to 100.5mV / g, 96.6mV / g, and 103.5mV / g, respectively, and the X, Y, and Z axis sensitivities of the second vibration sensor are set to 100.4mV / g, 88.2mV / g, and 97mV / g, respectively.

[0076] The two acoustic sensors are placed 10cm and 1m away from the wheel hub motor respectively, and the two sensors need to be placed in the same straight line or at 90°. When the two acoustic sensors are in the same straight line, they need to be at the same horizontal height as the motor to measure the noise at different distances and capture more subtle sound changes. The sensitivity of the two acoustic sensors is set to 50mV / g. The sampling window is designed as a Hanning window, and the sound level meter type is designed as A-weighted.

[0077] By correctly arranging the positions of the sensors, this embodiment can collect higher-resolution acoustic and vibration signal data, capture more subtle vibration and sound changes, and thus provide more detailed information. High-resolution acoustic and vibration signal data not only improves the accuracy of the data, but also increases the amount of data, providing more information for subsequent signal processing and fault diagnosis. This high-sensitivity setting enables the system to capture weak signal changes, which may be early signs of faults.

[0078] This embodiment also sets the acquisition parameters of the vibration sensor and the acoustic sensor so that a wider dynamic range of acoustic vibration signals can be captured. This means that signals ranging from very weak to very strong can be processed, thereby providing more comprehensive fault diagnosis information. In terms of data volume, this means that more signal data needs to be processed and analyzed, including those generated under extreme conditions. This wide dynamic range capture capability makes it possible to adapt to different operating conditions and environments, and improves the accuracy and reliability of fault diagnosis.

[0079] For example: the bandwidth of the acoustic sensor is usually designed to be 6400 Hz, but the frequency bands of some high-order electromagnetic noise and carrier frequency noise exceed this bandwidth. This information may be motor fault information. Therefore, this embodiment designs the bandwidth of the acoustic sensor to be 12800 Hz, so that high-order electromagnetic noise and carrier frequency noise can be captured, thereby improving the accuracy of fault diagnosis.

[0080] Implementation method three, see Figure 4 This embodiment is described. This embodiment specifically describes the optimization method of the robust local mean decomposition algorithm described in the first embodiment.

[0081] The method comprises the following steps:

[0082] Step 1: Initial setting of the meander optimization algorithm;

[0083] Step 2: The position of each individual in the search space of the meander optimization algorithm is represented as the optimization parameter of the robust local mean decomposition algorithm, thereby obtaining multiple parameter combinations;

[0084] Step 3: Substitute multiple parameter combinations into the robust local mean decomposition algorithm respectively, and determine the optimal parameters using the fitness function of the robust local mean decomposition algorithm;

[0085] Step 4: Determine whether the optimization reaches the termination condition of the meandering optimization algorithm. If so, proceed to the next step; if not, update the population position of the meandering optimization algorithm and return to step 2;

[0086] Step 5: Save the optimal parameter combination and substitute it into the robust local mean decomposition algorithm to obtain the optimized robust local mean decomposition algorithm.

[0087] In practical application, if Figure 4As shown in the figure, before the optimization process starts, the initial settings of the meander optimization algorithm (SBOA) need to be made, including: setting the number of iterations and population size of the meander optimization algorithm (SBOA); and the optimization parameters of the robust local mean decomposition algorithm (RLMD) need to be determined, including: defining the optimization range of parameters such as bandwidth threshold and B-spline order, which are specially selected for the characteristics of the wheel hub motor sound and vibration signal. The position of each individual in the meander optimization algorithm in the search space represents the optimization parameters of the robust local mean decomposition algorithm, thereby obtaining multiple parameter combinations. These parameter combinations will be used for the subsequent fitness evaluation of the wheel hub motor sound and vibration signal to ensure that the algorithm can optimize the specific structure of the wheel hub motor sound and vibration signal. Each parameter combination is substituted into the robust local mean decomposition algorithm (RLMD), and the performance of these parameters is evaluated using the fitness function. The fitness function is obtained by decomposing the wheel hub motor sound and vibration signal data, calculating the envelope entropy of each IMF component, and taking the minimum value of the envelope entropy as the fitness function to ensure that the specific structural properties of the wheel hub motor sound and vibration signal can be optimized. The types and values ​​of key parameters required for each type of data are different. Therefore, the algorithm used in this embodiment is designed according to the characteristics of the hub motor sound and vibration signal. According to the evaluation result of the fitness function, the optimal parameter combination is selected. At the same time, it is also necessary to determine whether the optimization reaches the termination condition of the meander optimization algorithm. If so, proceed to the next step; if not, update the population position of the meander optimization algorithm until the optimization is completed. Save the optimal parameter combination and substitute it into the robust local mean decomposition algorithm, and optimize the nonlinear and non-stationary characteristics of the hub motor sound and vibration signal, and use the robust local mean decomposition algorithm (RLMD) with the optimal parameter combination to decompose the hub motor sound and vibration signal data.

[0088] Among them, the meander optimization algorithm (SBOA) is specifically described:

[0089] The SBOA specifically includes: Initialization phase: The position of each griffon (candidate solution) is randomly generated and located between the upper and lower bound constraints of the given problem. The position represents a candidate solution to the problem, and the objective function value is evaluated based on the problem variable values ​​proposed by the griffon, and the results are compiled into a vector. Determine the hunting strategy: Hunting behavior is usually divided into three stages: finding prey, consuming prey, and attacking prey. The SBOA simulates these three stages and explores the search space through different update strategies.

[0090] Among them, the prey search phase: In this phase, the griffon searches for potential prey in a wide area. The algorithm updates the current griffon position by randomly selecting two different griffon positions, calculating the difference between them, and then multiplying it by a randomly generated coefficient vector. This strategy helps the algorithm to explore widely in the search space and avoid falling into a local optimal solution.

[0091] Consuming prey: When the vulture approaches prey, it will adopt a more cautious strategy to consume the prey. In the algorithm, this corresponds to using the best solution currently found and gradually approaching the optimal solution through smaller steps.

[0092] Attacking prey: When the vulture approaches its prey, it will also adopt an evasive strategy to avoid being discovered by the predator. In the algorithm, this corresponds to avoiding overexploitation by taking smaller steps when approaching the optimal solution.

[0093] In each iteration, the meander optimization algorithm (SBOA) checks the positions of all meanders and updates the global optimal solution. If the current meander position is better than the global optimal solution, the global optimal solution is updated.

[0094] The multiple parameter combinations obtained by the meandering optimization algorithm (SBOA) are substituted into the robust local mean decomposition algorithm, and the optimal parameters are determined by the fitness function; among them, the robust local mean decomposition algorithm (RLMD) is used to decompose the acoustic and vibration signals of the hub motor, and the envelope entropy of each IMF component is calculated, and the minimum value of the envelope entropy is used as the fitness function.

[0095] The envelope entropy calculation formula is:

[0096]

[0097] in, p i It is i The probability distribution of the envelope entropy of the IMF components. Determine whether the optimization reaches the termination condition of the algorithm. If so, proceed to the next step; if not, update the population position and reset the hunting strategy parameters.

[0098] The snake-snaking optimization algorithm (SBOA) iterates until a termination condition is met, such as reaching a maximum number of iterations or the quality of the solution meets the requirements.

[0099] The optimal parameter combination is saved and substituted into the robust local mean decomposition (RLMD) algorithm.

[0100] Implementation method 4: See Figure 4 This embodiment is described. This embodiment is a specific description of step S5 described in the above embodiment 1.

[0101] Step S51: Calculate the instantaneous amplitude and phase of the wheel hub motor acoustic vibration signal data using an optimized robust local mean decomposition algorithm;

[0102] Step S52: determining the local maximum and local minimum of the instantaneous amplitude;

[0103] Step S53: Calculate the local cutoff frequency according to the local maximum and local minimum of the instantaneous amplitude;

[0104] Step S54: filter the signal at the local cutoff frequency, and judge and iteratively decompose the filtering result until all signals are decomposed to obtain the best IMF component.

[0105] In practical application, if Figure 4 As shown in the figure, according to the evaluation result of the fitness function, the optimal parameter combination is selected. And the nonlinear and non-stationary characteristics of the wheel hub motor acoustic and vibration signals are optimized, and the robust local mean decomposition algorithm (RLMD) with the optimal parameter combination is used to decompose the acoustic and vibration signal data of the wheel hub motor. This step includes calculating the instantaneous amplitude and phase of the signal, determining the local maximum and minimum values ​​of the instantaneous amplitude, calculating the local cutoff frequency based on these values, and performing signal filtering and iterative decomposition calculations until the optimal IMF component is obtained.

[0106] Specifically:

[0107] The optimized robust local mean decomposition algorithm (RLMD) decomposes the acoustic and vibration signal data of the hub motor to obtain the best IMF component, and selects the IMF component with the largest envelope spectrum fault feature energy ratio for envelope demodulation analysis to extract fault features. Figure 4 As shown,

[0108] (1) Calculate the instantaneous amplitude and phase using the following formula:

[0109]

[0110] For the actual wheel hub motor acoustic and vibration signal x t , calculate its instantaneous amplitude A(t) and instantaneous phase ϕ ( t ),in, x t yes x t Hilbert transform.

[0111] (2) Find the local minimum and maximum and calculate the local cutoff frequency. A ( t ) is the local minimum A tmin and local maximumA tmax . The local cutoff frequency is calculated using the following formula:

[0112]

[0113] in, and It is estimated based on local minima and maxima.

[0114] (3) Signal filtering, judgment and iteration. The nodes as local extreme points are further interpolated using B-spline approximate filters. τ ( k ), and get the filtering result C 1( k ).

[0115] if ,but σ ( k ) is considered as an IMF component; otherwise, σ ( k )− C 1( k ) as the new input signal and repeat the above steps. ζ is the bandwidth threshold, σ avg ( k ) is the weighted average instantaneous frequency, BL ( k ) is the instantaneous bandwidth.

[0116] (4) Signal decomposition, finally, the original input signal σ ( k ) is decomposed by the robust local mean decomposition (RLMD) algorithm to obtain S Subsequence C i ( k );

[0117] in, C i ( k ) is the i subsequences, and satisfy:

[0118]

[0119] The optimized robust local mean decomposition (RLMD) algorithm is used to decompose the acoustic and vibration signal data of the hub motor to obtain the optimal IMF component, and the eigenvalue of the optimal IMF component is extracted. The mean, variance, peak value, kurtosis, effective value, peak factor, pulse factor, waveform factor and margin factor of the motor sound signal component are calculated as the eigenvector.

[0120] In summary, this embodiment formulates the optimization parameters of the robust local mean decomposition algorithm (RLMD) algorithm based on the characteristics of the wheel hub motor acoustic and vibration signals, including defining the optimization range of parameters such as bandwidth threshold and B-spline order. The formulation of the optimization parameters ensures that the robust local mean decomposition algorithm (RLMD) algorithm can adapt to the specific characteristics of the acoustic and vibration signals, such as nonlinearity and non-stationarity. At the same time, the RLMD algorithm optimized by the meander optimization algorithm (SBOA) can process acoustic and vibration signal data of different magnitudes. The adaptability of the algorithm is reflected in the ability to extract key features from a large amount of acoustic and vibration signal data, which are crucial for the fault diagnosis of the wheel hub motor. Therefore, the optimized robust local mean decomposition algorithm (RLMD) described in this embodiment can more accurately identify and diagnose the faults of the wheel hub motor, rather than as a general signal processing tool.

[0121] Implementation method 5: See Figure 5 This embodiment is described. This embodiment specifically describes step S6 described in the first embodiment.

[0122] S61: Setting an initial sample weight distribution, specifically:

[0123] Initialize the sample weights. The initial weight of each sample is the same and is usually set to N1, where N is the total number of samples.

[0124]

[0125]

[0126] Using the sample weight distribution to train a weak classifier in the random forest classifier, and calculating the classification error rate of the weak classifier;

[0127] Specifically:

[0128] Iteratively train the weak classifier, for each round m(m= 1 , 2 , …… ,M）， Use the current weight distribution D m Train a weak classifier G m (x) and calculate the classification error rate. m Classification error rate of weak classifiers e m .

[0129]

[0130] in, I is the indicator function, if G m(xi) and the true label y i Different, then I is 1 if the value is true, otherwise it is 0.

[0131] Step S62: Calculate the weight of the weak classifier according to the classification error rate, and update the sample weight distribution according to the weight until multiple weak classifiers are trained;

[0132] Specifically:

[0133] Calculate the weight of the weak classifier and calculate the weight of the weak classifier according to the classification error rate α m .

[0134]

[0135] Then update the sample weights and update the sample weight distribution D m+1 , prepare for the next round of iteration.

[0136]

[0137] in, Z m is the normalization factor, ensuring D m+1 is a probability distribution.

[0138] Step S63: combining all weak classifiers into a strong classifier;

[0139] Specifically:

[0140] Construct the final classifier and combine all weak classifiers into a strong classifier.

[0141]

[0142]

[0143] in, sign Represents the symbol of the function return parameter. If the parameter is less than 0, it outputs -1, and if it is greater than 0, it outputs 1. Output the final classification result. For the new input sample x , using the final classifier G(x) to classify.

[0144] Step S64: Evaluate the performance of the strong classifier and determine the weights of different weak learners until a final classifier is obtained as an enhanced random forest classifier;

[0145] Step S65: input the extracted feature set into the enhanced random forest classifier to obtain a diagnosis result.

[0146] In practical application, if Figure 5 As shown, each sample has the same weight to form an initial sample weight distribution; use the current sample weight distribution to train a weak classifier, and this implementation uses a random forest classifier. If the sample is correctly classified, its weight will be reduced when constructing the next training set, otherwise it will be increased; use the updated sample set to train the next classifier. For the misclassified training sample set, the weight will increase, and the weight of the misclassified sample will be amplified; combine all weak classes into a strong classifier. After the training process of each weak classifier is completed, increase the weight of the weak classifier with a small classification error rate, and reduce the weight of the weak classifier with a large classification error rate; evaluate the performance of the final strong classifier and determine the weights of different weak learners.

[0147] Embodiment 6. This embodiment provides a hub motor fault diagnosis system based on SBOA-RLMD-AdaBoost-RF, which is used to implement the hub motor fault diagnosis method based on RLMD improved by SBOA described in the above embodiment. The system includes an acoustic vibration signal acquisition module, a parameter optimization module, a signal decomposition module and a fault diagnosis identification module;

[0148] The sound and vibration signal acquisition module is used to collect the sound and vibration signal data of the hub motor;

[0149] The parameter optimization module is used to determine the optimization parameters of the robust local mean decomposition algorithm according to the wheel hub motor sound and vibration signal data, and is also used to optimize the parameters of the robust local mean decomposition algorithm stored in the signal decomposition module by using the meander optimization algorithm;

[0150] The signal decomposition module is used to decompose the hub motor sound and vibration signal data using an optimized robust local mean decomposition algorithm to obtain a feature set of the best IMF component;

[0151] The fault diagnosis and identification module is used to use an enhanced random forest classifier to perform fault diagnosis and identification on a feature set of an optimal IMF component to obtain a diagnosis result.

[0152] In the actual application of this embodiment, the acoustic and vibration signal acquisition module is used to obtain the sound and vibration signal data of the hub motor; the parameter optimization module is specially designed to determine the optimization parameters of the robust local mean decomposition algorithm, such as bandwidth threshold and B-spline order, based on the sound and vibration signal data of the hub motor, so as to improve the effect of signal decomposition. Through this key step, the module can effectively calculate the optimal parameters of the robust local mean decomposition algorithm (RLMD), ensuring the accuracy and reliability of subsequent data processing steps, and providing a strong guarantee for the early fault diagnosis of the hub motor. The signal decomposition module uses the optimized robust local mean decomposition algorithm (RLMD) to decompose the signal and calculate the envelope entropy of each IMF (Intrinsic Mode Functions) component. The minimum value of the envelope entropy is used as the fitness function to obtain the optimal IMF component, and the optimal IMF is used instead of the original signal as the input signal for feature extraction. The fault diagnosis and identification module uses the reinforcement learning algorithm (AdaBoost) to train the random forest classifier (RF) to obtain the enhanced RF strong classifier. The trained random forest classifier is used to classify the status of the hub motor, determine whether there is a fault and the type of fault, and evaluate the performance of the fault diagnosis system, including accuracy, recall rate, F1 score and other indicators. In actual application, a user interface module will also be added to provide a user interaction interface, allowing users to enter parameters, start the diagnosis process, and display the diagnosis results.

[0153] Implementation method 7: See Figure 2 and Figure 3 This embodiment is described. This embodiment is a hub motor fault diagnosis system and method thereof for a hub motor fault diagnosis method based on RLMD improved by SBOA described in the above embodiment.

[0154] The main steps include:

[0155] (1) Test environment layout and signal acquisition: Adjust the height of the acoustic sensor and vibration sensor, use the Testlab software to collect the motor sound and vibration signals, and set the sensor sensitivity, sampling frequency, sampling bandwidth, sampling time and other parameters in the test analysis software Testlab. The system uses two acoustic sensors and two vibration sensors.

[0156] Considering that the radial vibration of the motor is more significant during actual riding, two vibration sensors are placed at both ends of the motor shaft, with the X-axis direction being horizontal, the Y-axis direction being axial, and the Z-axis direction being perpendicular to the ground. After the test is completed, the vibration data of the X-axis and Z-axis are mainly focused.

[0157] like Figure 2 and Figure 3As shown, the two acoustic sensors are placed at 10cm and 1m away from the hub motor respectively to measure the noise conditions at different distances, which can capture more subtle sound changes. To ensure the accuracy of the test, the positions of the two sensors need to be kept in the same straight line or placed at a 90° angle. In this test, the two sensors are located in the same straight line and maintain the same horizontal height as the motor. Among them, the X, Y, and Z axis sensitivities of the first vibration sensor are set to 100.5mV / g, 96.6mV / g, and 103.5mV / g, respectively. The response sensitivities of the second vibration sensor are 100.4mV / g, 88.2mV / g, and 97mV / g, respectively. The sensitivity of both acoustic sensors is set to 50mV / g. The sampling window uses the Hanning window, and the sound level meter type selects A weighting. After the parameters are set, the acoustic and vibration signals in various states are collected when the motor is running;

[0158] (2) Parameter optimization module, which is specially designed to optimize the parameters of the RLMD algorithm using the meander optimization algorithm (SBOA), such as the bandwidth threshold and the B-spline order, to improve the effect of signal decomposition. The specific steps are as follows:

[0159] ①: Set the number of iterations and population size of the Scorpion Bird Optimization Algorithm (SBOA), and define the parameter optimization range of the Robust Local Mean Decomposition (RLMD) algorithm. Randomly initialize the position of the secretary bird in the search space, and each secretary bird represents a candidate solution to the optimization problem. The initialization formula is as follows:

[0160]

[0161] in, Xi represents the position of the i-th secretary bird, lb and ub are the lower and upper bounds of the solution space, respectively. r is a random number between 0 and 1.

[0162] ②: Set up hunting strategy. This stage is divided into three steps:

[0163] Finding prey: Simulate the behavior of secretary birds in finding prey and use differential evolution strategy for global search. The formula is as follows:

[0164]

[0165] in, X new,i Indicates i The new state of the secretary bird in the stage of searching for prey, X random_1 and X random_2 Randomly selected secretary bird positions,R 1 is a randomly generated array in the interval [0,1].

[0166] Consuming prey: Simulate the behavior of secretary birds consuming prey and use Brownian motion for local search. The update formula is as follows:

[0167]

[0168] in, x best Indicates the current optimal solution.

[0169] Attacking prey: Simulate the behavior of secretary birds attacking prey and introduce Levy flight strategy for global search. The Levy flight formula is as follows:

[0170]

[0171] in, u and v are random numbers from a standard normal distribution, β are the parameters of Levy flight.

[0172] ③: Use the robust local mean decomposition (RLMD) algorithm to decompose the signal and calculate the envelope entropy of each IMF component, with the minimum envelope entropy as the fitness function. Envelope entropy calculation formula:

[0173]

[0174] in, p i It is i The envelope entropy probability distribution of the IMF components. Determine whether the optimization reaches the termination condition of the algorithm. If so, proceed to the next step; if not, update the population position and return to step ② to reset the hunting strategy parameters.

[0175] (3) Signal decomposition module: Decompose the signal using the optimized RLMD algorithm and calculate the envelope entropy of each IMF (Intrinsic Mode Functions) component. The minimum value of the envelope entropy is used as the fitness function to obtain the optimal IMF component. The optimal IMF is used instead of the original signal as the input signal for feature extraction. The specific steps are as follows:

[0176] ①: Calculate the instantaneous amplitude and phase, the formula is as follows:

[0177]

[0178] For the actual signal x t , calculate its instantaneous amplitude A(t) and instantaneous phase ϕ ( t ),inx t yes x t Hilbert transform.

[0179] ②: Find the local minimum and maximum and calculate the local cutoff frequency. A ( t ) is the local minimum A tmin and local maximum A tmax . The local cutoff frequency is calculated using the following formula:

[0180]

[0181] in, and It is estimated based on local minima and maxima.

[0182] ③: Signal filtering, judgment and iteration. The nodes as local extreme points are further interpolated using B-spline approximate filters. τ ( k ), and get the filtering result C 1( k ).

[0183] if ,but σ ( k ) is considered as an IMF component; otherwise, σ ( k )− C 1( k ) as the new input signal and repeat the above steps. ζ is the bandwidth threshold, σ avg ( k ) is the weighted average instantaneous frequency, BL ( k ) is the instantaneous bandwidth.

[0184] ④: Signal decomposition, finally, the original input signal σ ( k ) is decomposed by RLMD to obtain S Subsequence C i ( k ).in, C i ( k ) is the i subsequences, and satisfy:

[0185]

[0186] ⑤: Extract features from the decomposed signal to form a feature set. Calculate the mean, variance, peak value, kurtosis, effective value, peak factor, pulse factor, waveform factor, and margin factor of the motor sound and vibration signal components as feature vectors. For motor sound signals, the mean can provide information about the baseline level or DC component of the signal as shown below:

[0187]

[0188] in, x mean represents the signal mean; N Indicates the number of samples; x i Represents the sample value.

[0189] In the motor sound signal, the variance can represent the overall change of the signal, thus reflecting the stability or instability of the signal, as shown below:

[0190]

[0191] in, x var represents the signal variance.

[0192] The peak value indicates the maximum amplitude or extreme value of the signal. In motor noise, the peak information can be used to detect the maximum noise amplitude of the signal as follows:

[0193]

[0194] in, x peak Indicates the signal peak.

[0195] Kurtosis describes the sharpness or thickness of the tail of a signal. In motor noise, a high kurtosis value indicates that the signal has sharp peaks, as shown below:

[0196]

[0197] in, x kur Represents the signal kurtosis value.

[0198] The effective value reflects the effective energy or amplitude of the signal. In motor noise, the effective value is used to evaluate the overall energy level of the signal, as well as the relative strength of the noise, as shown below:

[0199]

[0200] in, x rms Indicates the effective value of the signal.

[0201] The crest factor describes the relationship between the peak value and the effective value, providing information about the peak characteristics and peak margin of the signal, as shown in the following formula:

[0202]

[0203] in, x pf Indicates the signal crest factor.

[0204] The pulse factor reflects the pulse or impact of the signal. In motor noise, it is used to evaluate the sharpness or impact of the signal, as shown in the following formula:

[0205]

[0206] in, x if Indicates the signal pulse factor.

[0207] The form factor describes the waveform shape or energy distribution of a signal. In motor noise, the form factor provides information about the waveform characteristics of the signal, as shown in the following formula:

[0208]

[0209] in, x sf Indicates signal shape factor.

[0210] The margin factor describes the dynamic range between the peak value and the effective value. In motor noise, the margin factor provides information about the signal dynamic range or peak margin as shown in the following equation:

[0211]

[0212] in, x cf Represents the signal margin factor.

[0213] (4) Fault diagnosis and identification module: Use the reinforcement learning algorithm (AdaBoost) to train the random forest classifier (RF) to obtain the enhanced RF strong classifier, and use the trained random forest classifier to classify the status of the hub motor. The specific steps are as follows:

[0214] ①: Initialize sample weights. The initial weight of each sample is the same, usually set to N 1, where N is the total number of samples.

[0215]

[0216]

[0217] ②: Iteratively train weak classifiers. For each round m (m=1,2,…,M), Use the current weight distribution D m Training a weak classifier G m (x) And calculate the classification error rate. m Classification error rate of weak classifiers e m .

[0218]

[0219] in, I is an indicator function, if G m (x i ) With the true label y i Different, then I is 1 if the value is true, otherwise it is 0.

[0220] ③: Calculate the weight of the weak classifier, and calculate the weight of the weak classifier according to the classification error rate α m ;

[0221]

[0222] Then update the sample weights and update the sample weight distribution D m+1 , prepare for the next round of iteration.

[0223]

[0224] in, Z m is the normalization factor, ensuring D m+1 is a probability distribution.

[0225] ④: Build the final classifier and combine all weak classifiers into a strong classifier.

[0226]

[0227]

[0228] in, sign The function returns the sign of the parameter. If the parameter is less than 0, it outputs -1, and if it is greater than 0, it outputs 1. Output the final classification result. For the new input sample x , using the final classifier G (x ) for classification.

[0229] In order to evaluate the reliability of the model after training, the accuracy of the classification results is calculated x cacc , unweighted average precision x uap , unweighted average recall x uar , geometric mean x gm , F1 score x F1 For performance evaluation, the formula is as follows:

[0230]

[0231]

[0232]

[0233]

[0234]

[0235] In the formula, N is the number of categories, and its value is 4. TP i , FP i 、FN i TN i are the true positive number, false positive number, false negative number, and true negative number of the classification results respectively.

[0236] The classification results shown in Table 1 are obtained:

[0237] Table 1

[0238]

[0239] The results show that the random forest (Adaboost-RF) optimized by reinforcement learning has high evaluation indicators, including 99.55±0.25% accuracy, 98.65±0.16% non-weighted average precision, 99.50±0.23% non-weighted average recall, 97.63±0.30% geometric mean, and 97.46±0.22% F1 score. It has high reliability and is suitable for use as a fault diagnosis classifier for hub motors.

[0240] (5) User interface module: provides a user interaction interface, allowing users to input parameters, start the diagnostic process, and display the diagnostic results. Before acquisition, set the sampling frequency, phase, amplitude, acquisition channel, signal link mode, sampling rate, sampling rate per channel, sampling mode, filter selection, filter type, order, cutoff frequency range, passband ripple, stopband attenuation value, window type and other parameters.

[0241] In summary, the present invention innovatively develops an early fault diagnosis and identification technology for hub motors. By designing the collection position and parameter settings of the acoustic vibration sensor, the acoustic vibration signal data of the hub motor generated under extreme conditions can be captured; at the same time, the optimized robust local mean decomposition algorithm (RLMD) is used to process the acoustic vibration signal data of the hub motor to obtain the feature set of the best IMF components of the acoustic vibration signal data of the hub motor, and finally the enhanced random forest classifier is used to calculate the diagnostic results of the feature set, thereby improving the accuracy of the fault diagnosis results.

[0242] Furthermore, the present invention adopts an intelligent fault diagnosis method that integrates the robust local mean decomposition algorithm (RLMD) improved by the meandering vulture optimization algorithm (SBOA) and the random forest (RF) optimized by reinforcement learning (AdaBoost) to realize the fault diagnosis of the hub motor, and the early weak fault signal of the hub motor is processed by the meandering vulture optimization algorithm (SBOA) relying on its powerful global search ability and precise local search ability. By simulating the strategy of the predatory behavior of the meandering vulture and combining the Levy flight mechanism, the exploration and development ability of the algorithm is enhanced, and then the optimal parameters of the robust local mean decomposition algorithm (RLMD) are found, so that the weak fault characteristic signal in the signal is retained for the next step of processing.

[0243] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may be modified and varied in various ways.

Claims

1. A hub motor fault diagnosis method based on RLMD improved by SBOA, characterized in that: The method is: S1: Two vibration sensors are respectively set at both ends of the hub motor shaft, and two acoustic sensors are respectively set at a distance of 10 cm and 1 m from the hub motor; S2: Set parameters for two vibration sensors and an acoustic sensor respectively; S3: acquiring data collected by two vibration sensors and two acoustic sensors as acoustic vibration signal data of the wheel hub motor; S4: According to the acoustic and vibration signal data of the hub motor, the optimization parameters of the robust local mean decomposition algorithm are determined: bandwidth threshold and B-spline order; S5: Optimize the robust local mean decomposition algorithm RLMD, use the optimized robust local mean decomposition algorithm to calculate the acoustic and vibration signal data of the hub motor, obtain the best IMF component of the acoustic and vibration signal data of the hub motor, and extract the feature set of the best IMF component; The optimization and improvement method of the robust local mean decomposition algorithm RLMD is: Step 1: Initial setting of the meander optimization algorithm SBOA; Step 2: The position of each individual in the SBOA optimization algorithm in the search space is used to represent the optimization parameters of the robust local mean decomposition algorithm RLMD, thereby obtaining multiple parameter combinations; Step 3: Substitute multiple parameter combinations into the robust local mean decomposition algorithm RLMD, and determine the optimal parameters using the fitness function of the robust local mean decomposition algorithm RLMD; Step 4: Determine whether the optimization reaches the termination condition of the meander optimization algorithm SBOA. If so, proceed to the next step; if not, update the population position of the meander optimization algorithm SBOA and return to step 2; Step 5: Save the optimal parameter combination and substitute it into the robust local mean decomposition algorithm to obtain the optimized robust local mean decomposition algorithm; S6: Input the extracted feature set into the enhanced random forest classifier to obtain the diagnosis result.

2. The method for diagnosing hub motor faults based on RLMD improved by SBOA according to claim 1 is characterized in that: The vibration data of the two vibration sensors on the X-axis and Z-axis are used as the vibration signal data of the wheel hub motor; The two acoustic sensors must be kept in the same straight line or placed at 90°.

3. The method for diagnosing a hub motor fault based on RLMD improved by SBOA according to claim 2 is characterized in that: S2 is specifically: The X, Y, and Z axis sensitivities of the first vibration sensor of the two vibration sensors are set to 100.5 mV / g, 96.6 mV / g, and 103.5 mV / g, respectively, and the X, Y, and Z axis sensitivities of the second vibration sensor are set to 100.4 mV / g, 88.2 mV / g, and 97 mV / g, respectively; The sensitivity of both acoustic sensors was set to 50 mV / g; The bandwidth of both acoustic sensors was set to 12800 Hz.

4. The method for diagnosing hub motor faults based on RLMD improved by SBOA according to claim 1, characterized in that: Step 1 is as follows: Set the number of iterations of the meandering optimization algorithm, the population size, and define the parameter optimization range of the robust local mean decomposition algorithm.

5. The method for diagnosing wheel hub motor faults based on RLMD improved by SBOA according to claim 1, characterized in that: Multiple parameter combinations are used to evaluate the fitness of the hub motor sound and vibration signal data.

6. The method for diagnosing wheel hub motor faults based on RLMD improved by SBOA according to claim 1, characterized in that: The fitness function is obtained as follows: The robust local mean decomposition algorithm is used to decompose the wheel hub motor acoustic and vibration signal data, and the envelope entropy of each IMF component is calculated. The minimum value of the envelope entropy is used as the fitness function.

7. The method for diagnosing hub motor faults based on RLMD improved by SBOA according to claim 1, characterized in that: S5 is as follows: S51: Calculate the instantaneous amplitude and phase of the wheel hub motor acoustic and vibration signal data using an optimized robust local mean decomposition algorithm; S52: Determine the local maximum and local minimum of the instantaneous amplitude; S53: Calculating a local cutoff frequency according to a local maximum value and a local minimum value of the instantaneous amplitude; S54: filtering the signal at the local cutoff frequency, and judging and iteratively decomposing the filtering result until all signals are decomposed to obtain the best IMF component.

8. The method for diagnosing hub motor faults based on RLMD improved by SBOA according to claim 1, characterized in that: S6 is specifically: S61: setting an initial sample weight distribution, and using the sample weight distribution to train a weak classifier in the random forest classifier, and calculating the classification error rate of the weak classifier; S62: Calculate the weight of the weak classifier according to the classification error rate, and update the sample weight distribution according to the weight until multiple weak classifiers are trained; S63: combine all weak classifiers into a strong classifier; S64: Evaluate the performance of the strong classifier and determine the weights of different weak learners until the final classifier is obtained as an enhanced random forest classifier; S65: Input the extracted feature set into the enhanced random forest classifier to obtain a diagnosis result.

9. A hub motor fault diagnosis system based on RLMD improved by SBOA, characterized in that: The system includes an acoustic vibration signal acquisition module, a parameter optimization module, a signal decomposition module and a fault diagnosis and identification module; The sound and vibration signal acquisition module is used to collect the sound and vibration signal data of the hub motor; The parameter optimization module is used to determine the optimization parameters of the robust local mean decomposition algorithm according to the sound and vibration signal data of the hub motor, and is also used to optimize the parameters of the robust local mean decomposition algorithm stored in the signal decomposition module by using the meander optimization algorithm; The signal decomposition module is used to decompose the wheel hub motor sound and vibration signal data using the optimized robust local mean decomposition algorithm to obtain the feature set of the best IMF component; The fault diagnosis and identification module is used to use the enhanced random forest classifier to perform fault diagnosis and identification on the feature set of the best IMF component to obtain the diagnosis result.

Images