Small sample gear fault degree identification method, product, medium and equipment

Through the improved fitness optimization algorithm of the crown porcupine optimization algorithm, the hyperparameters of the convolutional neural network are optimized, combined with the convolutional neural network and batch normalization layer, the accuracy and adaptability of gear fault recognition under small sample conditions are solved, and the problem of gear failure recognition is achieved in Efficient identification effect in industrial production.

CN119089267APending Publication Date: 2024-12-06HAI AN & TAIYUAN UNIV OF TECH ADVANCED MFG & INTELLIGENT EQUIP IND RES INST +1
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202411183136.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Under small sample conditions, the gear fault degree identification algorithm is easy to overfit, and the adaptive gear fault degree identification algorithm has problems such as complex signal preprocessing and large data demand, which makes it difficult for the existing technology to effectively identify gear faults in industrial production degree.

Method used

The Crown Porcupine optimization algorithm with improved fitness optimizes the convolutional neural network hyperparameters, combines the convolutional neural network and batch normalization layer to design the adaptive convolutional neural network, optimize the convolutional neural network hyperparameters to improve the training convergence speed and Smoothness, build a small sample gear fault identification model.

Benefits of technology

The accuracy and adaptability of gear failure degree recognition under small sample conditions are improved, and accurate recognition under small sample conditions is achieved, reducing labor costs and algorithm complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089267A_ABST
    Figure CN119089267A_ABST
Patent Text Reader

Abstract

The invention discloses a small sample gear fault degree identification method, a product, a medium and equipment, and relates to the technical field of gear fault degree identification, and the method comprises the steps: obtaining vibration signals of a gear in different states, and carrying out the preprocessing, and obtaining a vibration signal data set; constructing a convolutional neural network and initializing hyper-parameters of the convolutional neural network; optimizing the convolutional neural network hyper-parameters by adopting a crown porcupine optimization algorithm based on improved fitness to obtain optimized convolutional neural network hyper-parameters; reconstructing the convolutional neural network based on the optimized convolutional neural network hyper-parameters, and training by using the vibration signal data set to obtain a gear fault recognition model; and identifying the fault degree of the gear by adopting the gear fault identification model. According to the small sample gear fault degree identification method, the product, the medium and the equipment, the accuracy of gear fault degree identification under the small sample condition and the self-adaptability of the identification method can be effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of gear fault degree identification, and in particular to a small sample gear fault degree identification method, product, medium and equipment. Background Art

[0002] Continuous and smooth transmission can be achieved through the meshing of gears. When the degree of gear failure is too large, it will have a huge impact on the mechanical transmission system, thereby threatening production safety and economic benefits. In industrial production, there are problems such as difficulty in collecting gear fault data and high labeling costs, which leads to small training samples. Training the gear fault degree recognition algorithm under small sample conditions will overfit, making it difficult to accurately identify the degree of gear failure. In addition, the construction of traditional gear fault degree recognition algorithms relies on expert experience and requires multiple debugging to obtain good performance, which will cost huge manpower costs. Therefore, it is of great engineering significance to develop an adaptive gear fault degree recognition algorithm suitable for small sample conditions.

[0003] In recent years, the research on small sample gear fault degree identification algorithm and adaptive gear fault degree identification algorithm has achieved certain results. However, the small sample gear fault degree identification algorithm still has problems such as complex model, large number of parameters, difficulty in parameter setting, difficulty in training, and reliance on expert experience, while the adaptive gear fault degree identification algorithm has problems such as complex signal preprocessing and large data demand, which makes the two algorithms unable to be effectively integrated. Therefore, it is urgent to improve the adaptability of the small sample gear fault degree identification algorithm and simplify the algorithm structure to integrate the strengths of the two algorithms, so as to develop an adaptive gear fault degree identification algorithm suitable for small sample conditions. Summary of the invention

[0004] The purpose of this application is to provide a method, product, medium and equipment for identifying the degree of gear faults in a small sample, which can effectively improve the accuracy of gear fault degree identification under small sample conditions and the adaptability of the identification method.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for identifying the degree of gear faults in a small sample, comprising:

[0007] The vibration signals of the gears in different states are obtained and preprocessed to obtain a vibration signal data set; the vibration signals include vibration signals of the gears in a normal state and in a peeling state with fault degrees of 0.5, 1, 2 and 4 respectively;

[0008] Construct a convolutional neural network and initialize the convolutional neural network hyperparameters;

[0009] The crown porcupine optimization algorithm based on improved fitness is used to optimize the hyperparameters of the convolutional neural network, and the optimized hyperparameters of the convolutional neural network are obtained;

[0010] The convolutional neural network is reconstructed based on the optimized convolutional neural network hyperparameters and trained using the vibration signal dataset to obtain a gear fault recognition model.

[0011] The gear fault identification model is used to identify the fault degree of the gear.

[0012] Optionally, the step of acquiring vibration signals of gears in different states and preprocessing them to obtain a vibration signal data set specifically includes:

[0013] The vibration signals of the gears in different states are acquired and preprocessed by data division, DC component removal, normalization and format conversion in sequence to obtain the preprocessed vibration signals;

[0014] A vibration signal dataset is constructed based on the preprocessed vibration signal.

[0015] Optionally, the step of acquiring vibration signals of gears in different states and sequentially performing data division, DC component removal, normalization, and format conversion preprocessing to obtain preprocessed vibration signals specifically includes:

[0016] The vibration signals of the gears in different states are divided at equal intervals to obtain the divided vibration signals a;

[0017] Using formula The divided vibration signal a is preprocessed to remove the DC component, and a vibration signal h with the DC component removed is obtained; wherein, is the mean value of the vibration signal a after division;

[0018] Using formula The vibration signal h with the DC component removed is normalized and preprocessed to obtain the vibration signal h with the DC component removed and normalized. * ; Among them, h min is the minimum value of the vibration signal h after removing the DC component; h max is the maximum value of the vibration signal h after removing the DC component;

[0019] The vibration signal h after removing the DC component and normalizing * Perform format conversion preprocessing to obtain a preprocessed vibration signal.

[0020] Optionally, the convolutional neural network specifically includes: a first convolutional layer, a first batch normalization layer, a first ReLU activation function layer, a first maximum pooling layer, a second convolutional layer, a second batch normalization layer, a second ReLU activation function layer, a second maximum pooling layer, a fully connected layer and a Softmax layer connected in sequence.

[0021] Optionally, the convolutional neural network hyperparameters include a learning rate, an L2 regularization coefficient, a convolution kernel size of the first convolution layer and the second convolution layer, and a number of convolution kernels.

[0022] Optionally, the method of optimizing the hyperparameters of the convolutional neural network using the crested porcupine optimization algorithm based on improved fitness to obtain the optimized hyperparameters of the convolutional neural network specifically includes:

[0023] The improved fitness is used as the fitness index of the crested porcupine optimization algorithm and the minimization of the improved fitness is used as the optimization target of the crested porcupine optimization algorithm, and the crested porcupine optimization algorithm based on the improved fitness is obtained;

[0024] The crown porcupine optimization algorithm based on improved fitness is used to optimize the hyperparameters of the convolutional neural network, and the optimized hyperparameters of the convolutional neural network are obtained.

[0025] Optionally, the specific formula for improving fitness is:

[0026]

[0027] Where fitness is the improved fitness; Loss is the cross entropy loss function; Z is the number of training iterations, i = 1, 2, 3, …, Z; I is an indicator for measuring the stability and convergence speed of training.

[0028] In a second aspect, the present application provides a computer program product, including a computer program, which implements the small sample gear fault degree identification method when executed by a processor.

[0029] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying the degree of fault of a small sample gear.

[0030] In a fourth aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the small sample gear fault degree identification method.

[0031] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0032] The present application provides a method, product, medium and equipment for identifying the degree of gear faults in a small sample, and realizes the identification of the degree of gear faults in a small sample by designing an adaptive convolutional neural network with improved fitness. Specifically, the present application designs an improved fitness that comprehensively considers the influence of convergence speed and stability during the training process of the convolutional neural network, and combines the improved fitness with the crown porcupine optimization algorithm with strong global optimization strength and accelerated convergence to obtain a crown porcupine optimization algorithm based on improved fitness, and uses the crown porcupine optimization algorithm based on improved fitness to optimize and select convolutional neural network hyperparameters, thereby improving the optimization effect of convolutional neural network hyperparameters; after obtaining the optimized convolutional neural network hyperparameters, the convolutional neural network is reconstructed and trained to obtain a gear fault identification model; the gear fault identification model can be used to accurately identify the degree of gear faults under small sample conditions. The method, product, medium and equipment for identifying the degree of gear faults in a small sample of the present application can effectively improve the accuracy of gear fault degree identification under small sample conditions and the adaptability of the identification method. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A schematic diagram of the process flow of the method for identifying the degree of gear fault with a small sample provided for this application;

[0035] Figure 2 The time domain waveform diagram of the vibration signal of the gear provided in this application under different states;

[0036] Figure 3 A schematic diagram of the structure of the convolutional neural network provided for this application;

[0037] Figure 4 A schematic diagram of the workflow of the crested porcupine optimization algorithm provided for this application;

[0038] Figure 5 A comparison chart of the training effects of the adaptive convolutional neural network based on improved fitness and cross entropy loss function fitness provided in this application;

[0039] Figure 6 A comparison chart of the recognition effects of the adaptive convolutional neural network based on improved fitness and cross entropy loss function fitness provided in this application;

[0040] Figure 7A fitness optimization comparison diagram of the crested porcupine optimization algorithm based on improved fitness provided in this application and other optimization algorithms;

[0041] Figure 8 A comparison chart of the training effect of the small sample gear fault degree identification method provided in this application and other algorithms;

[0042] Fig. 9 A comparison chart of the recognition effect of the small sample gear fault degree recognition method provided in this application and other algorithms;

[0043] Fig.10 A schematic diagram of a confusion matrix for a test data set identified by a small sample gear fault degree identification method provided in this application;

[0044] Fig.11 This is a t-SNE visualization of the small sample gear fault degree identification method provided in this application for the test data set. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0047] like Figure 1 As shown, a method for identifying the degree of gear fault with a small sample disclosed in the present application includes:

[0048] Step 1: Obtain the vibration signals of the gears in different states and preprocess them to obtain a vibration signal data set.

[0049] Specifically, a piezoelectric acceleration sensor is used to collect vibration signals of gears in normal state and in peeling state with fault degrees of 0.5, 1, 2 and 4. The vibration signal comes from the HD-CL-012C rotor gearbox test bench, with a sampling frequency of 10KHz, a motor speed of 900rpm, and a load of 5N·m. The time domain waveform of the vibration signal is shown in the figure below: Figure 2 As shown. Then the collected vibration signal is preprocessed, including data division, DC component removal, normalization and format conversion, as follows:

[0050] (1) Data Division

[0051] The vibration signals of the gears in different states are equally spaced and divided. Specifically, the collected vibration signals are divided into samples of 1500 equally spaced data points.

[0052] (2) Remove DC component

[0053] Remove the DC component of the divided vibration signal to eliminate the influence of the DC component and improve the data quality. The formula for removing the DC component is as follows:

[0054]

[0055] Where a is the vibration signal after division; is the mean value of the vibration signal a after division; h is the vibration signal with the DC component removed.

[0056] (3) Normalization

[0057] The vibration signal with the DC component removed is normalized to [0, 1] to eliminate the dimension effect and improve the training speed of the convolutional neural network. The normalization formula is as follows:

[0058]

[0059] In the formula, h min is the minimum value of the vibration signal after removing the DC component; h max is the maximum value of the vibration signal after removing the DC component; h * is the vibration signal with the DC component removed and normalized.

[0060] (4) Format conversion

[0061] The data format of the vibration signal with the DC component removed and normalized is converted so as to be input into the convolutional neural network. Specifically, the data format of the vibration signal with the DC component removed and normalized is converted into [1500, 1, 1] so as to be input into the convolutional neural network for processing.

[0062] Step 2: Build a convolutional neural network and initialize the convolutional neural network hyperparameters.

[0063] Specifically, the specific structure of the established convolutional neural network is as follows Figure 3As shown in the figure, the first layer is the first convolutional layer, the convolution kernel size is [K1, 1], the number of convolution kernels is N1, and the moving step is [1, 1], where K1 and N1 are the hyperparameters of the convolutional neural network to be optimized, and their optimization ranges are [1, 100] and [1, 50] respectively; the second layer is the first batch normalization layer; the third layer is the first ReLU activation function layer; the fourth layer is the first maximum pooling layer, the pooling kernel size is [2, 1], and the moving step is [1, 1]; the fifth layer is the second convolutional layer, the convolution kernel size is [2, 1], and the moving step is [1, 1]; The kernel size is [K2, 1], the number of convolution kernels is N2, and the moving step is [1, 1], where K2 and N2 are the hyperparameters of the convolutional neural network to be optimized, and their optimization ranges are [1, 100] and [1, 50] respectively; the sixth layer is the second batch normalization layer; the seventh layer is the second ReLU activation function layer; the eighth layer is the second maximum pooling layer, the pooling kernel size is [2, 1], and the moving step is [1, 1]; the ninth layer is a fully connected layer with 5 nodes; the tenth layer is a Softmax layer.

[0064] This application uses a maximum pooling layer with a step size of [1, 1] and a pooling kernel size of [2, 1] to retain as much feature information as possible and highlight strong features; the batch normalization layer is introduced to suppress the overfitting problem, thereby improving the versatility of the convolutional neural network.

[0065] The training parameters of the constructed convolutional neural network are set as follows: the optimizer is Adam; the loss function is cross entropy; the batch size and the number of training rounds are 64 and 30 respectively, and the total number of iterations is 210; the learning rate decay factor and the learning rate decay period are 0.1 and 10 respectively; the L2 regularization coefficient is l2, which is the convolutional neural network hyperparameter to be optimized, and the range is [1e-4, 1e-2]; the learning rate is lr, which is the convolutional neural network hyperparameter to be optimized, and the range is [1e-4, 1e-2].

[0066] The purpose of setting the L2 regularization coefficient, learning rate decay factor and learning rate decay period in this application is to accelerate convergence, reduce oscillation and suppress overfitting during the training process of the convolutional neural network so as to improve the versatility of the convolutional neural network and the ability of the convolutional neural network to identify the degree of gear faults under small sample conditions.

[0067] In summary, the convolutional neural network hyperparameters to be optimized in this application include the learning rate, the L2 regularization coefficient, the convolution kernel size of the first convolution layer and the second convolution layer, and the number of convolution kernels.

[0068] Step 3: Use the crown porcupine optimization algorithm based on improved fitness to optimize the hyperparameters of the convolutional neural network to obtain the optimized hyperparameters of the convolutional neural network.

[0069] Specifically, the present application first sets the parameters of the crested porcupine optimization algorithm, including: the initial population size is 20; the minimum population size is 16; the maximum number of iterations is 100; the loop control variable is 2; the convergence speed factor is 0.1; and the trade-off factor is 0.5. Subsequently, an improved fitness is designed as the fitness index of the crested porcupine optimization algorithm and the minimization of the improved fitness is used as the optimization target of the crested porcupine optimization algorithm to obtain the crested porcupine optimization algorithm based on the improved fitness. Finally, the crested porcupine optimization algorithm based on the improved fitness is used to optimize and select the hyperparameters of the convolutional neural network to obtain the optimized hyperparameters of the convolutional neural network. Among them, the improved fitness comprehensively considers the influence of convergence speed and stability during the training process of the convolutional neural network.

[0070] The construction principle of improved fitness is as follows: first, the loss function of the entire training process is solved on average to reduce the impact of training randomness; second, a reasonable loss function threshold is set based on experience, and the number of iterations is counted from the last training iteration until the loss function value is higher than the set loss function threshold. The larger the number of iterations counted, the higher the training stability and training convergence speed; finally, the above two parts are combined to improve fitness.

[0071] The specific formula for improving fitness is as follows:

[0072]

[0073] Where fitness is the improved fitness, the smaller its value is, the better the comprehensive training effect of the convolutional neural network is; Loss is the cross entropy loss function; Z is the number of training iterations, i = 1, 2, 3, ..., Z; I is an indicator for measuring the stability and convergence speed of training, and its formula is as follows:

[0074]

[0075] Where D is the loss function threshold set based on experience and is set to 0.1.

[0076] The crested porcupine optimization algorithm is an optimization algorithm developed based on the bionics principle and the various defensive behaviors of crested porcupines. It has the advantages of strong global optimization and accelerated convergence. The population initialization formula of the crested porcupine optimization algorithm is as follows:

[0077] X i =L+r×(UL)|i=1,2...,N' (5)

[0078] Among them, N' is the initial population size; U is the search upper limit; L is the search lower limit; r is a random vector with each element between 0 and 1; X i is the i-th initialized individual in the population.

[0079] Inspired by the fact that only threatened crested porcupine individuals in a population will activate their defense mechanisms, the crested porcupine optimization algorithm developed a population reduction technique with the following formula:

[0080]

[0081] Where T is the loop control variable; T max is the maximum number of iterations; % is the remainder symbol; t is the current number of iterations; N min is the minimum value of the newly generated population size; N is the newly generated population size.

[0082] The crested porcupine optimization algorithm has four defense strategies corresponding to the crested porcupine's visual, sound, smell, and physical attack defense behaviors:

[0083] (1) The first defense strategy: When a crested porcupine is aware of an enemy, it will flap its quills, and the enemy will either approach it or move away from it. When the enemy approaches it, it will prioritize exploring the area between the two parties, and when the enemy moves away from it, it will prioritize exploring the distant area. The formula is as follows:

[0084]

[0085] in, is the optimal solution accumulated to the tth iteration; is a vector representing the enemy position at the tth iteration; τ1 is a random number based on normal distribution; τ2 is a random number between 0 and 1; r is a random number between 1 and N; and are the positions of the i-th individual at the t-th iteration and the t+1-th iteration respectively; is the position of the rth individual at the tth iteration.

[0086] (2) Second defense strategy: Crested porcupine individuals will make noise to threaten the enemy. The closer the enemy gets to it, the louder the noise it makes. If the noise is small, the enemy will continue to approach; if the noise is large, the enemy will stop where it is; if the noise is very loud, the enemy will be afraid and leave. The formula is as follows:

[0087]

[0088] in, and are the positions of the r1th and r2th individuals at the tth iteration respectively; r1 and r2 are random numbers between 1 and N; τ3 is a random number between 0 and 1; y is a vector representing the enemy's position; U1 is a randomly generated binary vector.

[0089] (3) The third defense strategy: Crested porcupines will secrete malodorous substances and spread the odor in the surrounding area to prevent enemies from approaching. The formula is as follows:

[0090]

[0091] in, is the position of the r3th individual at the tth iteration; r3 is a random number between 1 and N; τ3 is a random number between 0 and 1; δ is a parameter used to control the search direction; r4 is a random vector with each element between 0 and 1; rand is a random number between 0 and 1; γ t is the defense factor; max is the maximum number of iterations; is the odor diffusion factor; f is the fitness function; is a minimum value used to avoid division by zero; is the position of the jth individual at the tth iteration.

[0092] (4) The fourth defense strategy: When the enemy is very close to the crested porcupine individual, it will use its quills to attack the enemy. The formula is as follows:

[0093]

[0094] Among them, α is the convergence speed factor; τ4 and τ5 are random numbers between 0 and 1; F i t is the average force exerted by the ith individual on the predator at the tth iteration, which is derived from the law of inelastic collision and appropriately simplified. The specific formula is as follows:

[0095]

[0096] Among them, m i is the mass of the ith individual; is the final velocity of the i-th individual at the t+1th iteration; is the initial velocity of the i-th individual at the t-th iteration; τ is a random vector with each element between 0 and 1.

[0097] Among them, Figure 4 As shown in the figure, the optimization of the hyperparameters of the convolutional neural network using the crested porcupine optimization algorithm based on improved fitness mainly includes population initialization, random indicator generation, defense strategy trade-off, optimal solution (optimal convolutional neural network hyperparameters) update and population reduction, as follows:

[0098] S1: Initialize the population according to the optimization range of the convolutional neural network hyperparameters and the parameter settings of the crested porcupine optimization algorithm.

[0099] S2: Randomly generate τ6, τ7, τ8, τ9 and τ between 0 and 1 10 index.

[0100] S3: Select and execute corresponding defense strategies for the hyperparameter optimization problem of convolutional neural networks.

[0101] Specifically, when τ8<τ9, the exploration phase begins, otherwise, the development phase begins. In the exploration phase, when τ6<τ7, the first defense strategy is executed, otherwise, the second defense strategy is executed. In the development phase, τ 10 <T f The third defense strategy will be executed when the error is 0, and the fourth defense strategy will be executed when the error is 0. f is a trade-off factor between 0 and 1. This application sets T f This application sets the minimization of improved fitness as the execution goal of the defense strategy.

[0102] S4: Update the optimal convolutional neural network hyperparameters based on the execution results of the defense strategy.

[0103] S5: Update the population size using cyclic population reduction technique.

[0104] S6: Enter the next optimization iteration.

[0105] S7: Repeat S2 to S6 until the number of optimization iterations reaches the upper limit.

[0106] Step 4: Reconstruct the convolutional neural network based on the optimized convolutional neural network hyperparameters and train it using the vibration signal dataset to obtain the gear fault recognition model.

[0107] After obtaining the vibration signal data set through step 1, the present application also divides the vibration signal data set into a training data set and a test data set in a ratio of 5:2, wherein each state in the training data set contains 100 samples, and each state in the test data set contains 40 samples, so as to construct a data set under small sample conditions. Subsequently, based on the training data set under small sample conditions, the convolutional neural network is reconstructed and trained using the optimized convolutional neural network hyperparameters to obtain a gear fault recognition model. The gear fault recognition model is used to identify the test data set, and a gear fault degree recognition test is performed under small sample conditions to verify the accuracy of the gear fault recognition model.

[0108] As a specific embodiment, by Figure 5It can be seen that the training accuracy curve of the adaptive convolutional neural network based on improved fitness in this application reached the highest accuracy for the first time in the 8th round of training, and it no longer fluctuated after reaching the highest accuracy; the training accuracy curve of the adaptive convolutional neural network based on the fitness of the cross entropy loss function reached the highest accuracy for the first time in the 18th round of training, and it still fluctuated after reaching the highest accuracy. It can be seen that compared with the fitness of the cross entropy loss function, the improved fitness of this application can improve the training convergence speed and training stability of the adaptive convolutional neural network.

[0109] pass Figure 6 It can be seen that the recognition accuracy of the adaptive convolutional neural network based on improved fitness in this application for the test data set reached 99%, while the recognition accuracy of the adaptive convolutional neural network based on the cross entropy loss function fitness for the test data set only reached 97%. It can be seen that compared with the cross entropy loss function fitness, the improved fitness of this application can improve the recognition effect of the adaptive convolutional neural network.

[0110] pass Figure 7 It can be seen that compared with other optimization algorithms, such as the whale optimization algorithm, the particle swarm optimization algorithm and the sparrow search algorithm, the crown porcupine optimization algorithm based on improved fitness (referred to as the crown porcupine optimization algorithm) of this application has the best optimization effect on the improved fitness of the adaptive convolutional neural network. It has a faster convergence speed, and the optimization curve has converged at the 22nd iteration. It has a strong optimization strength, and the final fitness is 0.00367, which is significantly better than other optimization algorithms.

[0111] pass Figure 8 It can be seen that compared with other algorithms, such as traditional convolutional neural network (CNN), CNN-BiLSTM and CNN-LSTM, the training effect of the small sample gear fault degree identification method of the present application is the best. Among them, CNN-BiLSTM is a joint algorithm of CNN and bidirectional long and short-term memory neural network (Bi-directional Long and Short-Term Memory, BiLSTM); CNN-LSTM is a joint algorithm of CNN and long and short-term memory neural network (Long and Short-Term Memory, LSTM). The present application has a very fast training convergence speed and strong training stability. Its training accuracy curve reached the highest accuracy in the 8th round of training, and it no longer fluctuated after reaching the highest accuracy, which is significantly better than other algorithms.

[0112] pass Fig. 9It can be seen that compared with other algorithms, such as traditional CNN, CNN-BiLSTM, CNN-LSTM, BiLSTM, LSTM and back propagation neural network (BP), the gear fault recognition model constructed by the small sample gear fault degree recognition method of the present application has the best recognition effect. The recognition accuracy of the gear fault recognition model of the present application for the training data set and the test data set reached 100% and 99%, which is significantly better than other algorithms.

[0113] pass Fig.10 and Fig.11 From the confusion matrix diagram of the small sample gear fault degree identification method provided in the present application for the identification of the test data set and the t-distributed Stohastic Neighbor Embedding (t-SNE) visualization effect diagram for the identification of the test data set, it can be seen that the gear fault identification model of the present application has a good recognition effect on the test data set. It can fully extract the fault characteristics of the collected signal and can accurately separate and identify the collected vibration signals of different fault degrees.

[0114] Step 5: Use the gear fault identification model to identify the fault degree of the gear.

[0115] After verifying the accuracy of the gear fault identification model through the above step 4, the gear fault identification model is used to identify the fault degree of the gear. Compared with other machine learning algorithms, the training curve of the gear fault identification model constructed by the small sample gear fault degree identification method of the present application is smoother and converges faster, which can more accurately identify the gear fault degree under small sample conditions.

[0116] In summary, the small sample gear fault degree identification method of the present application first preprocesses the collected vibration signal to eliminate the dimension effect, eliminate the DC component effect and convert the data format, so as to improve the data quality and facilitate the data input into the convolutional neural network for subsequent processing; secondly, the present application constructs a convolutional neural network, uses a maximum pooling layer with a step size of [1, 1] and a pooling kernel size of [2, 1] to retain as much feature information as possible and highlight strong features, introduces a batch normalization layer and an L2 regularization mechanism to suppress the overfitting problem of the convolutional neural network and improve the performance of the convolutional neural network, so that the convolutional neural network The versatility is improved; secondly, this application designs an improved fitness that comprehensively considers the influence of convergence speed and stability during the training process of the convolutional neural network, takes the improved fitness as the fitness index of the crown porcupine optimization algorithm with strong global optimization strength and accelerated convergence, and takes the minimization of the improved fitness as the optimization goal of the crown porcupine optimization algorithm, and uses the crown porcupine optimization algorithm based on improved adaptation to optimize the selection of convolutional neural network hyperparameters, thereby improving the optimization effect of the convolutional neural network hyperparameters; finally, the convolutional neural network is reconstructed and trained using the optimized convolutional neural network hyperparameters to obtain a gear fault recognition model. Experiments show that the gear fault recognition model of this application has an accuracy rate of 99% in identifying the degree of gear faults under small sample conditions. Therefore, the small sample gear fault degree recognition method of this application can effectively improve the accuracy of gear fault degree recognition under small sample conditions and the adaptability of the recognition method, and provide a technical support for the accurate recognition of mechanical fault degrees under small sample conditions.

[0117] In some embodiments, the present application also provides a computer program product, including a computer program, which implements the small sample gear fault degree identification method when executed by a processor.

[0118] In some embodiments, the present application also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for identifying the degree of gear fault with a small sample is implemented.

[0119] In some embodiments, the present application also provides a computer device, including a processor, a memory, an input / output interface (I / O for short), a communication interface, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for identifying the degree of gear fault in a small sample.

[0120] Wherein, the processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Wherein, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the method for identifying the degree of fault of a small sample gear is implemented.

[0121] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0122] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ReadOnlyMemory, ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0123] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for identifying the degree of gear faults with a small sample, characterized in that: include: The vibration signals of the gears in different states are obtained and preprocessed to obtain a vibration signal data set; the vibration signals include vibration signals of the gears in a normal state and in a peeling state with fault degrees of 0.5, 1, 2 and 4 respectively; Construct a convolutional neural network and initialize the convolutional neural network hyperparameters; The crown porcupine optimization algorithm based on improved fitness is used to optimize the hyperparameters of the convolutional neural network, and the optimized hyperparameters of the convolutional neural network are obtained; The convolutional neural network is reconstructed based on the optimized convolutional neural network hyperparameters and trained using the vibration signal dataset to obtain a gear fault recognition model. The gear fault identification model is used to identify the fault degree of the gear.

2. The method for identifying the degree of gear fault of a small sample according to claim 1 is characterized in that: The method of obtaining vibration signals of gears in different states and preprocessing them to obtain a vibration signal data set specifically includes: The vibration signals of the gears in different states are acquired and preprocessed by data division, DC component removal, normalization and format conversion in sequence to obtain the preprocessed vibration signals; A vibration signal dataset is constructed based on the preprocessed vibration signal.

3. The method for identifying the degree of gear fault of a small sample according to claim 2 is characterized in that: The method of obtaining the vibration signals of the gears in different states and sequentially performing data division, DC component removal, normalization and format conversion preprocessing to obtain the preprocessed vibration signals specifically includes: The vibration signals of the gears in different states are divided at equal intervals to obtain the divided vibration signals a; Using formula The divided vibration signal a is preprocessed to remove the DC component, and a vibration signal h with the DC component removed is obtained; wherein, is the mean value of the vibration signal a after division; Using formula The vibration signal h with the DC component removed is normalized and preprocessed to obtain the vibration signal h with the DC component removed and normalized. * ; Among them, h min is the minimum value of the vibration signal h after removing the DC component; h max is the maximum value of the vibration signal h after removing the DC component; The vibration signal h after removing the DC component and normalizing * Perform format conversion preprocessing to obtain a preprocessed vibration signal.

4. The method for identifying the degree of gear fault of a small sample according to claim 1, characterized in that: The convolutional neural network specifically includes: a first convolutional layer, a first batch normalization layer, a first ReLU activation function layer, a first maximum pooling layer, a second convolutional layer, a second batch normalization layer, a second ReLU activation function layer, a second maximum pooling layer, a fully connected layer and a Softmax layer connected in sequence.

5. The method for identifying the degree of gear fault of a small sample according to claim 4 is characterized in that: The convolutional neural network hyperparameters include learning rate, L2 regularization coefficient, convolution kernel size of the first convolution layer and the second convolution layer, and the number of convolution kernels.

6. The method for identifying the degree of gear fault with a small sample according to claim 1, characterized in that: The convolutional neural network hyperparameters are optimized by using the crested porcupine optimization algorithm based on improved fitness to obtain the optimized convolutional neural network hyperparameters, specifically including: The improved fitness is used as the fitness index of the crested porcupine optimization algorithm and the minimization of the improved fitness is used as the optimization target of the crested porcupine optimization algorithm, and the crested porcupine optimization algorithm based on the improved fitness is obtained; The crown porcupine optimization algorithm based on improved fitness is used to optimize the hyperparameters of the convolutional neural network, and the optimized hyperparameters of the convolutional neural network are obtained.

7. The method for identifying the degree of gear fault with a small sample according to claim 6, characterized in that: The specific formula for improving fitness is: Where fitness is the improved fitness; Loss is the cross entropy loss function; Z is the number of training iterations, i = 1, 2, 3, …, Z; I is an indicator for measuring the stability and convergence speed of training.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for identifying the degree of fault of a small sample gear according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying the degree of fault of a small sample gear according to any one of claims 1 to 7 is implemented.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for identifying the degree of fault of a small sample gear according to any one of claims 1 to 7.

Citation Information

Patent Citations

  • Intelligent fault diagnosis and service life prediction method for rocker arm of coal mining machine

    CN110987396A

  • Automatic compression processing method for target detection network model

    CN112612602A

  • Hydraulic pump leakage fault diagnosis method

    CN115822943A

  • Classification method for bearing fault diagnosis

    CN116992341A

  • Motor turn-to-turn short circuit fault detection method based on variational mode decomposition fusion deep learning

    CN117572300A