Permanent magnet hub motor demagnetization fault diagnosis method considering multi-source signal fusion

Abstract

The invention discloses a permanent magnet wheel hub motor demagnetization fault diagnosis method considering multi-source signal fusion, belongs to the technical field of motor fault diagnosis, and aims to perform permanent magnet wheel hub motor demagnetization fault diagnosis by using a multi-source heterogeneous signal feature fusion method and solve the problem of dependence on additional sensors. The method mainly comprises the following steps: collecting multi-source signals such as current, back electromotive force and vibration under different conditions; performing symmetric point mode processing on the multi-source signal preliminarily, performing adaptive optimization and fusion on SDP parameters according to image fitness, and performing imaging processing again; the image database of the multi-source signals is used for fault diagnosis model training, and the diagnosis accuracy of each signal is obtained; a dynamic decision-making environment is established by integrating environmental conditions, image fitness and fault diagnosis accuracy, and a model is trained to realize optimal selection of multi-source heterogeneous signals under different conditions. According to the invention, through signal selection and fusion under different conditions, the accuracy and efficiency of motor fault diagnosis are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of motor fault diagnosis, and particularly relates to a permanent magnet wheel hub motor demagnetization fault diagnosis method considering multi-source signal fusion. BACKGROUND

[0002] Permanent magnet synchronous motor has the advantages of high energy density, strong overload capacity, fast response, low cost, simple maintenance and the like, so it becomes the most widely used driving motor for electric vehicles, and also experiences a change from centralized, wheel edge to wheel hub driving mode, and the formed permanent magnet wheel hub motor gradually becomes the mainstream of the development of electric vehicle driving systems. Then, in such a new driving form, the influence of complex working conditions such as road impact, high temperature environment and load mutation induces electrical faults such as winding short circuit, turn-to-turn short circuit and permanent magnet demagnetization, which directly affects the use performance, and even endangers the reliability and safety of the electric vehicle driving. Therefore, it is a very necessary work to monitor the operation state of the permanent magnet wheel hub motor and diagnose the motor fault in various working conditions in time. When the motor fails, the state signal will also change, and the signals used for motor fault diagnosis are often vibration signals, back electromotive force signals and current signals and the like.

[0003] CN114264953A proposes a permanent magnet synchronous motor demagnetization fault diagnosis method and system and a diagnosis device. The method first acquires three-phase current signals, secondly, the signals are two-dimensional image preprocessed, and the processed image is recorded in the sample library. Then, the sample library is divided into training samples and test samples, and the convolutional neural network model is trained using the training samples, thereby obtaining a model that can diagnose demagnetization faults. However, this method only relies on a single three-phase current signal for fault diagnosis, and since a single signal is easily disturbed by environmental factors such as electromagnetic interference and temperature changes, this may cause the diagnostic accuracy to decrease in different environments. To solve the above problem, CN116597167A proposes a permanent magnet synchronous motor small sample demagnetization fault diagnosis method, which acquires the vibration and current signals of the motor, converts the processing results into gray images and fuses them, and then uses an improved meta-learning network to construct a fault classifier. This method can realize multi-sensor information fusion without increasing the data samples, and improve the stability and accuracy of diagnosis. However, this method also brings some new problems when fusing multiple signals. On the one hand, the fusion of multiple signals greatly increases the complexity of the picture information, which brings greater challenges to the fault diagnosis process and increases the difficulty of diagnosis. On the other hand, the noise in the signal has not been effectively filtered out, and after the fusion of multiple signals, there are often multiple noise signals in a picture, which will further interfere with the accuracy of fault diagnosis. In addition, although this method has made some optimization and improvement on the selection of the optimal signal between multiple signals and the parameter matching adjustment of SDP, the parameter adjustment process requires a lot of time cost. Moreover, in order to ensure high applicability under different data and condition changes, the parameters need to be continuously adjusted and optimized, which undoubtedly increases the difficulty and cost in practical application. SUMMARY

[0004] In view of the above problems existing in the prior art, the present application proposes a permanent magnet wheel hub motor demagnetization fault diagnosis method considering multi-source signal fusion. First, it is converted into an SDP image. Second, the parameters of the SDP are optimized and adjusted using the whale optimization algorithm to obtain a high-adaptability image. The single signal and signal combination are selected and fused using reinforcement learning. The design is reasonable, solves the shortcomings of the prior art, and has good effects.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] A permanent magnet wheel hub motor demagnetization fault diagnosis method considering multi-source signal fusion includes the following steps:

[0007] Step 1: Under different environmental conditions, collect the multi-source signals of the permanent magnet wheel hub motor under different working conditions, including current, back electromotive force, and vibration signals;

[0008] Step 2: SDP preliminary image processing is performed on each signal, the image fitness is calculated, and the whale algorithm is used to optimize the SDP parameters, and the signals are re-imaged according to the optimization results;

[0009] Step 3: Based on the image results of the multi-source signals, the fault diagnosis models corresponding to different signals are established respectively, and the accuracy is tested respectively;

[0010] Step 4: A dynamic decision environment is built based on environmental condition factors and accuracy, and the optimal solution of signal selection under different environmental conditions is obtained according to the image fitness and fault diagnosis accuracy for fault diagnosis.

[0011] Further, in step 2, the calculation of the image fitness is specifically:

[0012] Firstly, the gray-scale images of different fault types are converted into pixel matrices G, and the expression is:

[0013]

[0014] In the formula, L is the number of rows of the pixel matrix; W is the number of columns of the pixel matrix, G(h,y) is the pixel value corresponding to the point (h,y), h=1,...,L, y=1,...,W;

[0015] Let and respectively represent the feature values of any two images in the position (h,y) within the range N×N; the Euclidean distance Diff between the kth image and the uth image k,u is expressed as:

[0016]

[0017] Considering the influence of noise on the image, a smoothness function is introduced to neutralize the image discriminant function, assuming that Ω is the neighborhood of the pixel (h,y) within the range N×N, the smoothness value at the pixel point (sh,sy) is represented as S(sh,sy), and the smoothness formula of the image at this point is:

[0018]

[0019] Let be the smoothness matrix corresponding to the neighborhood Ω, and the total number of elements in the matrix is (L-N+1)×(W-N+1), then the calculation formula of the image used to quantify the intra-class smoothness S of the image is recorded as:

[0020]

[0021] The fitness Diff(·) of the whole image is calculated, and the calculation formula is recorded as:

[0022]

[0023] wherein a s is a complex factor, N z is the number of images, S k and S u are the intra-class smoothness of the kth image and the uth image respectively.

[0024] Further, in step 2, the whale algorithm is specifically:

[0025] The whale predation search behavior is represented as:

[0026]

[0027] In the formula, X rand (t) is the target position of the whale individual randomly selected from the current population, X(t) is the position of the current whale individual, w rand is the surrounding step length, which represents the distance between the current whale individual and the randomly selected individual, and the mathematical expressions of the coefficients A and C are:

[0028]

[0029] In the formula, the control parameter a gradually decreases from 2 to 0 as the number of iterations increases; r1 and r2 are both random numbers distributed in [0, 1], and the smaller the algebraic value of A represents the shorter the step length;

[0030] The mathematical model of the shrinkage surrounding mechanism is represented as:

[0031]

[0032] In the formula, X best (t) is the current whale optimal predation position, w best is the best surrounding step length;

[0033] The mathematical model of the spiral predation position updating stage is represented as:

[0034] X(t+1) = |X best (t) - X(t)|·e bl ·cos(2πl) + X best (t); (9)

[0035] In the formula, |X best (t) - X(t)| is the distance between the current whale position and the optimal predation position; b is a constant coefficient of the spiral surrounding route; and l is a random number between [-1, 1];

[0036] The whale search surrounding and spiral hunting behavior in the algorithm occurs simultaneously, and the search surrounding mechanism includes two parts of shrinkage surrounding and spiral hunting. Here, the probabilities of the two behaviors are both set to 0.5, and the position updating formula of the whale individual is expressed as:

[0037]

[0038] In addition, the value of |A| is the main factor determining the whale behavior. When |A|>1, global search is performed, and the target position is updated by formula (6); when |A|<1, local search is performed, and the position is updated by formula (8) by referring to the optimal individual position, and p is a random probability factor.

[0039] Further, in step 2, the whale algorithm is used to optimize the SDP parameters, including the following sub-steps:

[0040] Step 2.1: input is the partial diagnostic signal Z=[E1, E2,..., E n ];

[0041] Step 2.2: randomly generate multiple sets of initial parameters, each set including two parameters τ, κ, and perform SDP image processing on the input diagnostic signal to obtain an initial feature map Z p =[P1, P2,..., P n ];

[0042] Step 2.3: calculate the fitness Diff() of each image, and calculate the current optimal solution;

[0043] Step 2.4: perform position updating according to formula (6) to obtain the global optimal solution, and perform local optimization according to formulas (8) and (9), with l as a random number for local optimal iteration solution;

[0044] Step 2.5: substitute the optimal solution of this time into the next optimization process;

[0045] Step 2.6: repeat steps 2.2-2.5 multiple times to obtain the optimal parameters τ' and κ';

[0046] Based on the solution results of step 2.6, the diagnostic signal is image processed to obtain a database with high image distinguishability.

[0047] Further, in step 3, a picture classification algorithm based on convolutional neural network CNN is used to perform fault classification processing on the database. The activation function in the convolutional neural network uses ReLU activation function, the Softmax activation function is used for multi-classification, the Adam optimizer is used for weight update, and the loss function is sparse classification cross-entropy.

[0048] Further, in step 4, the overall state environment S t , including temperature T, load L, speed R and battery supply condition B, is represented by a vector, S t = [T t , L t , R t , B t ]; the action A t is to select a signal from a candidate signal set, specifically including: motor no-load back electromotive force signal EMF, vibration signal Vib, motor current signal I, back electromotive force signal fusion vibration signal [EMF, Vib], vibration signal fusion motor current signal [Vib, I], back electromotive force signal fusion motor current signal [EMF, I]; the action space is defined as the combination of signals A t ∈ {EMF, Vib, I, [EMF, Vib], [Vib, I], [EMF, I]}; the reward function R t considers the accuracy Acc of signal selection and the picture fitness function Fit, and is defined as R t = α·Acc + β·Fit, where α and β are weight coefficients for balancing the influence of the two indicators;

[0049] The goal of reinforcement learning is to learn the optimal signal selection strategy by maximizing the cumulative reward, expressed as:

[0050]

[0051] where G t is the cumulative reward from step 1 to step n d , γ ∈ [0, 1] is the discount factor, and K is the number of iterations;

[0052] The deep Q network DQN approximates the action value function Q(S t , A t ) through a neural network, which realizes the value evaluation of state-action. The Q function update formula is:

[0053] Q(S t , A t ) ← Q(S t , A t ) + δ· [R t + γ·max A′ Q(S t+1 , A′) - Q(S t , A t )]; (12)

[0054] where δ is the learning rate, R t is the current reward, and max A′ Q(St+1 A' is the maximum Q value of the next state.

[0055] The beneficial technical effects brought by the present application are:

[0056] The present application can dynamically learn the optimal signal selection strategy, and adapt to different environmental conditions (temperature, load, speed, battery power supply). The DQN algorithm combined with the deep neural network can efficiently approximate Q(S t , A t ), and realize signal optimization under complex state space. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is a flowchart of the permanent magnet wheel hub motor fault diagnosis method based on data analysis of the present application.

[0058] Figure 2 It is a data processing schematic diagram based on reinforcement learning of the present application.

[0059] Figure 3 It is a database processing schematic diagram based on WOA algorithm optimization of the present application. DETAILED DESCRIPTION

[0060] The specific embodiments of the present application will be further described below in combination with specific embodiments:

[0061] A permanent magnet wheel hub motor demagnetization fault diagnosis method considering multi-source signal fusion, as shown in the figure, comprises the following steps: Figure 1

[0062] Step 1: Under different environmental conditions, collect multi-source signals of the permanent magnet wheel hub motor under different working conditions, including current, back electromotive force, vibration signal;

[0063] Step 2: Perform SDP preliminary image processing on each signal, calculate the image fitness, and use the whale algorithm to optimize the SDP parameters, and re-image each signal according to the optimization result, as shown in the figure; Figure 3

[0064] SDP image processing is a mapping algorithm from Cartesian coordinate system to polar coordinate system, which can intuitively describe the amplitude change and frequency characteristics of time series signal. The signal between the i-th moment and the i+τ moment in the time domain time series x can be converted into a point with corresponding radius and rotation angle in the polar coordinate system through simple calculation. The specific conversion formula can be expressed as:

[0065]

[0066] In the formula, x max and x min ​​α represents the maximum and minimum values ​​in the time series x; α0 is the initial angle for plotting; κ is the angular domain gain factor; and τ is the time delay coefficient.

[0067] To quantify the discriminative power of SDP images, a fitness function that reflects this discriminative power needs to be established. First, grayscale images of different fault types are converted into pixel matrices. The pixel matrix features at any point in the image are represented by the average value, and the representational differences between different images are defined based on Euclidean distance. Simultaneously, considering the impact of noise on the image, a smoothness function is introduced for control. Finally, the image discriminative power and smoothness are combined to obtain the image fitness.

[0068] The specific steps for calculating image fitness are as follows:

[0069] First, the grayscale images of different fault types are converted into a pixel matrix G, expressed as:

[0070]

[0071] In the formula, L is the number of rows in the pixel matrix; W is the number of columns in the pixel matrix; G(h,y) is the pixel value corresponding to point (h,y), h=1,...,L,y=1,...,W;

[0072] make and Let H and Y represent the feature values ​​of any two images within a range of N×N at position (h, y); then the feature value of the k-th image is... It can be represented as:

[0073]

[0074] Image feature values ​​of the u-th image The corresponding formula can be expressed as:

[0075]

[0076] The pixel difference between the k-th image and the u-th image is calculated as follows:

[0077]

[0078] use Reference The elements in the matrix are... Reference The elements in the matrix. Then the Euclidean distance Diff between corresponding points in the two images. k,u It can be represented as:

[0079]

[0080] Since the SDP method is used to convert images, the outermost basic blank pixels are discarded when calculating the Euclidean distance. The total sum of differences between (L-N+1)×(L-N+1) pixel pairs is calculated. The Euclidean distance between the kth image and the uth image represents the difference, which is defined according to the Euclidean distance, and its formula is:

[0081]

[0082] The formula quantifies the overall difference between images by calculating the eigenvalue of each pixel position. In image processing, a smaller Euclidean distance value indicates a higher similarity between the features of two images, while a larger value indicates a greater difference in features between the two images. Ideally, the greater the difference between different types of images, the better, because a greater difference means that different fault modes can be better distinguished. However, in practical applications, noise is an unavoidable factor in image processing. The presence of noise can cause disturbances in image features, affecting the calculation results of the Euclidean distance, and thus reducing the distinguishability between different fault modes. Therefore, how to effectively suppress the influence of noise when calculating the Euclidean distance has become a problem that needs to be solved in image difference measurement. Therefore, in addition to calculating the Euclidean distance, the influence of noise on the image is also considered, and a smoothing function is introduced to neutralize the image distinguishability function. Assuming that Ω is the neighborhood of pixel (h, y) within the N×N range, the smoothing value of pixel (sh, sy) is represented by

[0083]

[0084] The smoothing value of all points in the entire image can be written as

[0085]

[0086] Let be the smoothing matrix corresponding to the neighborhood Ω. The total number of elements in the matrix is (L-N+1)×(W-N+1). The formula for calculating the image in-class smoothness S of the image is

[0087]

[0088] The image adaptability calculation formula considering the image difference and the image internal smoothness between the kth image and the uth image can be represented as

[0089]

[0090] where Diff k,u (·) is the adaptability calculation function of the kth image and the uth image, Sk and S u are the intra-class smoothness of the kth image and the uth image, respectively; a s is a composite factor, which adjusts the proportion between the image discrimination and the smoothness function. The greater the composite factor, the greater the proportion of smoothness, and the more sensitive the image to data noise; the smaller the composite factor, the more emphasis on image discrimination, and the easier to distinguish between different types of images.

[0091] After the motor signal is image processed by the SDP algorithm, the fitness of the overall picture is calculated by the above formula, and the calculation formula can be recorded as:

[0092]

[0093] Where N z is the number of images.

[0094] Through the calculation of the image fitness, the quality of the angular domain gain factor κ and the time delay coefficient τ in the SDP parameter can be quantitatively processed. When the original motor signal is converted into a polar coordinate pattern, if the angular domain gain factor κ and the time delay coefficient τ are too large or too small, the image symmetry will be poor, the fitness result will be poor, and the image features will not be obvious.

[0095] The whale algorithm is specifically:

[0096] The whale predation search behavior is represented as:

[0097]

[0098] In the formula, X rand (t) is the target position of a randomly selected whale individual from the current population, X(t) is the position of the current whale individual, w rand is the surrounding step length, which represents the distance between the current whale individual and the randomly selected individual, and the mathematical expressions of coefficients A and C are:

[0099]

[0100] In the formula, the control parameter a gradually decreases from 2 to 0 as the number of iterations increases; r1 and r2 are both random numbers distributed in [0, 1], and the smaller the algebraic value of A represents the shorter the step length;

[0101] The mathematical model of the shrinkage surrounding mechanism is represented as:

[0102]

[0103] In the formula, X best (t) is the current whale optimal predation position, w best is the best surrounding step length;

[0104] The mathematical model of spiral predation update position stage is expressed as:

[0105] X(t+1) = |X best (t) - X(t) | · e bl · cos(2πl) + X best (t); (16)

[0106] wherein, |X best (t) - X(t) | is the distance between the current whale position and the optimal predation position; b is a constant coefficient of spiral hunting route; l is a random number between [-1, 1];

[0107] The behaviors of whale search enclosure and spiral predation in the algorithm occur simultaneously, and the search enclosure mechanism includes two parts of shrinkage enclosure and spiral predation, and the probabilities of the two behaviors are both set to 0.5, and the position update formula of the whale individual is expressed as:

[0108]

[0109] In addition, the value of |A| is the main factor to determine the behavior of the whale, when |A| > 1, global search is performed, and the target position is updated by formula (13); when |A| < 1, local search is performed, and the position is updated by formula (15) by referring to the optimal individual position, and p is a random probability factor.

[0110] The optimization of SDP parameters by using the whale algorithm includes the following sub-steps:

[0111] Step 2.1: input is part of the diagnostic signal Z = [E1, E2,..., E n ];

[0112] Step 2.2: randomly generate multiple groups of initial parameters, each group including two parameters τ, κ, and perform SDP image processing on the input diagnostic signal to obtain an initial feature map Z p = [P1, P2,..., P n ];

[0113] Step 2.3: calculate the fitness Diff() of each group of images, and calculate the current optimal solution;

[0114] Step 2.4: perform position update according to formula (13) to obtain the global optimal solution, and perform local optimization according to formulas (15) and (16), and l is a random number for local optimal iteration solution;

[0115] Step 2.5: the optimal solution of this time is substituted into the next optimization process;

[0116] Step 2.6: Steps 2.2-2.5 are repeated multiple times to obtain optimal parameters τ' and κ';

[0117] Based on the results of step 2.6, the diagnostic signal is image processed to obtain a database with high image resolution, and a dynamic decision environment is established for reinforcement learning training.

[0118] Step 3: Based on the image results of the multi-source signal, corresponding fault diagnosis models are established and tested for accuracy, as shown in Figure 2 ;

[0119] The picture classification algorithm based on convolutional neural network (CNN) is used for fault classification processing of the database. The convolutional neural network used is a basic convolutional network structure, the activation function in the convolutional neural network is ReLU activation function, the Softmax activation function is used for multi-classification, the Adam optimizer is used for weight update, and the loss function is sparse classification cross-entropy, which is suitable for integer type labels.

[0120] Step 4: Based on the environmental condition factors and accuracy, a dynamic decision environment is built, and the optimal solution of signal selection under different environmental conditions is obtained according to the image fitness and fault diagnosis accuracy for fault diagnosis.

[0121] The overall state environment S t is established, including temperature T, load L, speed R and battery power supply condition B, represented by a vector, S t =[T t ,L t ,R t ,B t ]; the action A t is to select signals from the candidate signal set, including: motor no-load back electromotive force signal EMF, vibration signal Vib, motor current signal I, back electromotive force signal fusion vibration signal [EMF, Vib], vibration signal fusion motor current signal [Vib, I], back electromotive force signal fusion motor current signal [EMF, I]; the action space is defined as the combination of signals A t ∈{EMF,Vib,I,[EMF,Vib],[Vib,I],[EMF,I]}; the reward function R t considers the accuracy of signal selection Acc and the picture fitness function Fit, defined as R t = α·Acc + β·Fit, where α and β are weight coefficients for balancing the influence of the two indicators.

[0122] The goal of reinforcement learning is to learn the optimal signal selection strategy by maximizing the cumulative reward (expected discounted return), expressed as:

[0123]

[0124] wherein G t is the cumulative reward from step 1 to step n d , γ ∈ [0, 1] is the discount factor, and K is the number of iterations.

[0125] Deep Q-Network (DQN) approximates the action-value function Q(S t ,A t ) by neural network to evaluate the value of state-action.

[0126] Q(S t ,A t ) ← Q(S t ,A t ) + α· [R t + γ· max A′ Q(S t+1 ,A′) - Q(S t ,A t )]; (19)

[0127] wherein δ is the learning rate, R t is the current reward, and max A′ Q(S t+1 ,A′) is the maximum Q value of the next state.

[0128] Through the above reinforcement learning scheme, the optimal signal selection strategy can be dynamically learned systematically to adapt to different environmental conditions (temperature, load, speed, and battery power supply). The DQN algorithm combined with the deep neural network can efficiently approximate Q(S t ,A t ) to realize signal optimization in a complex state space.

[0129] The trained diagnostic model and classifier are used to extract features and classify the target domain data set to verify the reliability and accuracy of the trained model in detecting and diagnosing motor demagnetization faults under various conditions.

[0130] Of course, the above description is not a limitation of the present application, and the present application is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present application should also be within the scope of the present application.

Claims

1. A method for diagnosing demagnetization faults in permanent magnet hub motors considering multi-source signal fusion, characterized in that, Includes the following steps: Step 1: Under different environmental conditions, collect multi-source signals of the permanent magnet hub motor under different operating conditions, including current, back electromotive force, and vibration signals; Step 2: Perform preliminary SDP image processing on each signal, calculate the image fitness, and use the whale algorithm to optimize the SDP parameters. Based on the optimization results, perform image processing on each signal again. Step 3: Based on the image results of multi-source signals, establish fault diagnosis models corresponding to different signals respectively, and test the accuracy of each model. Step 4: Based on environmental conditions and accuracy, build a dynamic decision-making environment. Based on image fitness and fault diagnosis accuracy, obtain the optimal solution for signal selection under different environmental conditions for fault diagnosis.

2. The method for diagnosing demagnetization faults in a permanent magnet hub motor considering multi-source signal fusion as described in claim 1, characterized in that, In step 2, the image fitness is calculated as follows: First, the grayscale images of different fault types are converted into a pixel matrix G, expressed as: In the formula, L is the number of rows in the pixel matrix; W is the number of columns in the pixel matrix; G(h,y) is the pixel value corresponding to point (h,y), h=1,...,L,y=1,...,W; make and Let H and Y represent the feature values ​​of any two images within the range N×N at position (h, y); and let Diff represent the Euclidean distance between the k-th image and the u-th image. k,u Represented as: Considering the impact of noise on the image, a smoothness function is introduced to neutralize the image discrimination function. Let Ω be the neighborhood of the pixel at point (h,y) within the N×N range. Let (sh, sy) represent the smoothness value at pixel (sh, sy). Then, the formula for the smoothness of the image at that point is: make Let Ω be the smoothness matrix corresponding to the neighborhood Ω, and the total number of elements in the matrix is ​​(L-N+1)×(W-N+1). Then, the formula for calculating the intra-class smoothness S of the image used to quantize the image is denoted as: The fitness Diff(·) of the overall image is calculated using the following formula: Among them, a s As a composite factor, N z S represents the number of images. k and S u denoted as the intra-class smoothness of the k-th image and the u-th image, respectively.

3. The method for diagnosing demagnetization faults in a permanent magnet hub motor considering multi-source signal fusion as described in claim 2, characterized in that, In step 2, the whale algorithm specifically refers to: Whale predation and search behavior is represented as: In the formula, X rand (t) represents the target location of a randomly selected whale individual from the current group, X(t) represents the current location of the whale individual, and w rand To enclose the step size, representing the distance between the current whale individual and a randomly selected individual, the mathematical expressions for coefficients A and C are: In the formula, the control parameter a gradually decreases from 2 to 0 as the number of iterations increases; r1 and r2 are random numbers distributed in [0, 1], and the smaller the algebraic value of A, the shorter the step size. The mathematical model of the contraction encirclement mechanism is expressed as follows: In the formula, X best (t) represents the current optimal feeding position for the whale, w best To achieve the optimal encirclement step size; The mathematical model for the spiral predation position update phase is expressed as: X(t+1)=|X best (t)-X(t)|·e bl ·cos(2πl)+X best (t); (9) In the formula, |X best (t)-X(t)| represents the distance between the current whale position and the optimal predation position; b is the constant coefficient of the spiral encirclement route; l is a random number between [-1, 1]; In the algorithm, the whale's search for encirclement and spiral predation behaviors occur simultaneously. The search for encirclement mechanism includes two parts: shrinking encirclement and spiral predation. Here, the probability of both behaviors is set to 0.

5. The formula for updating the position of an individual whale is expressed as follows: In addition, the value of |A| is the main factor that determines the behavior of whales. When |A|>1, a global search is performed and the target position is updated by formula (6); when |A|<1, a local search is performed and the position is updated by formula (8) by referring to the optimal individual position, where p is a random probability factor.

4. The method for diagnosing demagnetization faults in a permanent magnet hub motor considering multi-source signal fusion as described in claim 3, characterized in that, Step 2, which optimizes the SDP parameters using the whale algorithm, includes the following sub-steps: Step 2.1: Input is a partial diagnostic signal Z = [E1, E2, ..., E n ]; Step 2.2: Randomly generate multiple sets of initial parameters, each set including two parameters τ and κ. Perform SDP image processing on the input diagnostic signal to obtain a set of initial feature maps Z. p =[P1,P2...,P n ]; Step 2.3: Calculate the fitness Diff() for each group of images and calculate the current optimal solution; Step 2.4: Update the position according to formula (6), find the global optimal solution, and perform local optimization according to formulas (8) and (9). Use l as a random number to perform local optimal iterative solution. Step 2.5: Substitute the optimal solution from this step into the next optimization process; Step 2.6: Repeat steps 2.2 to 2.5 multiple times to obtain the optimal parameters τ′ and κ′; Based on the solution results of step 2.6, the diagnostic signals are processed into images to obtain a database with high image discrimination.

5. The method for diagnosing demagnetization faults in a permanent magnet hub motor considering multi-source signal fusion as described in claim 4, characterized in that, In step 3, an image classification algorithm based on a convolutional neural network (CNN) is used to classify faults in the database. The activation function in the convolutional neural network is the ReLU activation function, the Softmax activation function is used for multi-class classification, the Adam optimizer is used for weight updates, and the loss function is sparse classification cross-entropy.

6. The method for diagnosing demagnetization faults in a permanent magnet hub motor considering multi-source signal fusion as described in claim 5, characterized in that, In step 4, the overall state environment S is established. t This includes temperature T, load L, rotational speed R, and battery power status B, represented by a vector S. t =[T t ,L t ,R t B t Action A t To select signals from the candidate signal set, specifically including: motor no-load back electromotive force signal EMF, vibration signal Vib, motor current signal I, back electromotive force signal fused with vibration signal [EMF,Vib], vibration signal fused with motor current signal [Vib,I], and back electromotive force signal fused with motor current signal [EMF,I]; the action space is defined as the combination A of signals. t ∈{EMF,Vib,I,[EMF,Vib],[Vib,I],[EMF,I]};Reward function R t Taking into account both the accuracy of signal selection (Acc) and the image fitness function (Fit), it is defined as R. t =α·Acc + β·Fit, where α and β are weighting coefficients used to balance the influence of the two indicators; The goal of reinforcement learning is to learn the optimal signal selection policy by maximizing cumulative reward, expressed as: Among them, G t For steps from step 1 to step n d The cumulative reward for each step, γ∈[0,1] is the discount factor, and K is the number of iterations; Deep Q-Network (DQN) approximates the action-value function Q(S) through a neural network. t A t This enables the evaluation of the value of state-action states; the Q-function update formula is: Q(S t ,A t )←Q(S t ,A t )+δ·[R t +γ·max A′ Q(S t+1 ,A′)-Q(S t ,A t )];(12) Where δ is the learning rate, R t For the current reward, max A′ Q(S t+1 ,A′) is the maximum Q value of the next state.

Images