Gearbox fault diagnosis method and system

Abstract

The invention relates to a gearbox fault diagnosis method and system. The method comprises the following steps: acquiring fault data of a gearbox; performing feature extraction by using the convolutional neural network fault data to obtain fault features; inputting the fault features into an extreme gradient lifting classifier to obtain a fault diagnosis result of the gearbox; wherein the extreme gradient lifting classifier adopts an improved big sugarcane-mouse algorithm to carry out parameter optimization, and the improved big sugarcane-mouse algorithm is combined with a nonlinear weight, an adaptive factor, sine and cosine and a sparrow warning mechanism. According to the invention, the features of the gearbox fault data are mined by using the convolutional neural network, accurate fault classification and identification are carried out on the basis of the features by using the extreme gradient lifting classifier, and the hyper-parameters of the extreme gradient lifting classifier are optimized by introducing the improved big sugarcane-mouse algorithm. And the optimization capacity, the stability and the convergence speed are improved, so that the overall performance of the model is effectively improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a gearbox fault diagnosis method and system. Background Art

[0002] In the industrial field, gearboxes are key components in mechanical transmission systems. Their failure may cause downtime of the entire equipment and high maintenance costs. Therefore, accurate and efficient gearbox fault diagnosis is of great significance to ensure the safe and stable operation of the equipment. With the development of machine learning and artificial intelligence technology, fault diagnosis technology is also constantly improving. However, traditional fault diagnosis methods often have problems of insufficient accuracy and poor generalization ability when dealing with complex working conditions and multiple fault states. Summary of the invention

[0003] Based on this, it is necessary to provide a gearbox fault diagnosis method and system that can improve the fault diagnosis accuracy and the generalization ability of the diagnostic model to address the above technical problems.

[0004] In a first aspect, the present application provides a gearbox fault diagnosis method. The method comprises:

[0005] Obtain gearbox fault data;

[0006] Use convolutional neural network to extract features from fault data and obtain fault features;

[0007] The fault features are input into the extreme gradient boosting classifier to obtain the fault diagnosis results of the gearbox;

[0008] Among them, the extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, which combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms.

[0009] In one embodiment, obtaining fault data of the gearbox includes:

[0010] Obtaining the one-dimensional vibration signal of the faulty gearbox in working state;

[0011] The one-dimensional vibration signal is converted into two-dimensional fault data using the Gram angular field difference.

[0012] In one embodiment, the improved cane rat algorithm includes:

[0013] Generate the spatial position of the great cane rat population; the spatial position of the great cane rat population includes the spatial positions of each great cane rat; determine the great cane rat as the dominant male according to the fitness values ​​of each great cane rat, and update the spatial position of the great cane rat population based on the position of the dominant male;

[0014] Entering the iteration step, the iteration step includes: initializing the sine-cosine strategy coefficient, the nonlinear weight factor and the adaptive factor; generating a random number, and comparing the random number with a preset value; if the random number is less than the preset value, entering the exploration phase, using the adaptive factor and the first factor to update the spatial position of the giant cane rat population; if the random number is not less than the preset value, entering the development phase, using the adaptive factor, the first factor and the second factor to update the spatial position of the giant cane rat population; successively evaluating whether the fitness value of each giant cane rat in the giant cane rat population updated in the exploration phase is better than the fitness value of the dominant male, if so, updating the current optimal fitness value and the dominant male position, if not, updating the current giant cane rat position according to the nonlinear weight factor, the sine strategy coefficient and the first factor, and updating the dominant male position at the current giant cane rat position. After the update, check whether the current position of the great cane rat meets the boundary conditions, and recalculate the fitness value of the current great cane rat. If it is better than the current best fitness value, update the current best fitness value and the dominant male position again; sequentially evaluate whether the fitness values ​​of the great cane rats in the great cane rat population updated in the development stage are better than the fitness value of the dominant male. If so, update the current best fitness value and the dominant male position. If not, update the current position of the great cane rat according to the nonlinear weight factor, the cosine strategy coefficient, the first factor and the second factor, and after the current position of the great cane rat is updated, check whether the current position of the great cane rat meets the boundary conditions, and recalculate the current fitness value of the great cane rat. If it is better than the current best fitness value, update the current best fitness value and the dominant male position again;

[0015] When the number of iterations reaches the maximum number of iterations, the current best fitness value and the dominant male position corresponding to the current iteration step are output;

[0016] The first factor is used to simulate the effect of abundant food sources, and the second factor is used to simulate the number of pups produced by cane rats each year.

[0017] In one embodiment, the sine-cosine strategy coefficient is related to the current number of iterations, the maximum number of iterations and the random number, and the nonlinear weight factor and the adaptive factor are both related to the current number of iterations and the maximum number of iterations.

[0018] In one embodiment, after updating the current spatial position of the great cane rat, the method further comprises:

[0019] Boundary conditions are used to constrain all cane rats to remain within the search space.

[0020] In one embodiment, the convolutional neural network introduces residual connections.

[0021] In a second aspect, the present application also provides a gearbox fault diagnosis system. The system comprises:

[0022] A data acquisition module, used to obtain fault data of the gearbox;

[0023] A feature extraction module is used to extract features from fault data using a convolutional neural network to obtain fault features;

[0024] A fault diagnosis module is used to input fault features into an extreme gradient boosting classifier to obtain a fault diagnosis result of the gearbox;

[0025] Among them, the extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, which combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms.

[0026] In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above gearbox fault diagnosis method when executing the computer program.

[0027] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the above-mentioned gearbox fault diagnosis method when executed by a processor.

[0028] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps in the above gearbox fault diagnosis method are implemented.

[0029] The above-mentioned gearbox fault diagnosis method and system include the following steps: obtaining the fault data of the gearbox; extracting features from the fault data using a convolutional neural network to obtain fault features; inputting the fault features into an extreme gradient boosting classifier to obtain the fault diagnosis result of the gearbox; wherein the extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, and the improved cane rat algorithm combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms. The present invention uses a convolutional neural network to mine the features of the gearbox fault data, and then uses an extreme gradient boosting classifier to perform accurate fault classification and identification based on these features. At the same time, an improved cane rat algorithm is introduced to optimize the hyperparameters of the extreme gradient boosting classifier, which improves the optimization ability, stability, and convergence speed, thereby effectively improving the overall performance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of a process flow of a gearbox fault diagnosis method in one embodiment;

[0031] Figure 2 A schematic diagram of a large cane rat shield method in one embodiment;

[0032] Figure 3 A schematic diagram of a flow chart of an improved cane rat algorithm in one embodiment;

[0033] Figure 4 It is a structural block diagram of a gearbox fault classification and identification model of GADF-CNN-IGCRA-XGBoost in one embodiment;

[0034] Figure 5 A box plot of CEC2022 benchmark function tests of six algorithms in one embodiment;

[0035] Figure 6 It is the algorithm convergence curve of F1 single peak and F2 multi-peak function;

[0036] Figure 7 It is the algorithm convergence curve of the composite and multi-modal function of F9 and F11;

[0037] Figure 8 It is a schematic diagram of the gearbox test bench structure and common gear failure types;

[0038] Fig. 9 This is a schematic diagram of the GADF-XGBOOST test results;

[0039] Fig.10 This is a schematic diagram of the GADF-GCRA-XGBOOST test results;

[0040] Fig.11 This is a schematic diagram of the GADF-IGCRA-XGBOOST test results;

[0041] Fig.12 This is a schematic diagram of the GADF-CNN-XGBOOST test results;

[0042] Fig.13 This is a schematic diagram of the GADF-CNN-GCRA-XGBOOST test results;

[0043] Fig.14 This is a schematic diagram of the GADF-CNN-IGCRA-XGBOOST test results;

[0044] Fig.15 1 is a structural block diagram of a gearbox fault diagnosis system in one embodiment. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0046] The present application embodiment provides a gearbox fault diagnosis method, such as Figure 1 As shown, the following steps are included:

[0047] Step 102, obtaining fault data of the gearbox.

[0048] The sensor is used to collect data during the operation of the gearbox. If a bearing fails, it will be reflected in the data collected by the sensor, and the bearing failure data at this time will be obtained. Since the bearing failure signal is transmitted in the form of vibration, the bearing failure data is usually a vibration signal.

[0049] Step 104: Use a convolutional neural network to extract features from the fault data to obtain fault features.

[0050] Convolutional Neural Networks (CNN) is a deep learning model that can effectively extract features from images through its unique convolutional layer structure. Traditional CNNs include convolutional layers and pooling layers, where the convolutional layer uses a set of learnable convolution kernels to slide on the input image or the feature map of the previous layer to generate a new feature map that shows the feature correspondence at different positions of the input image or the feature map of the previous layer; the pooling layer is used to reduce the spatial dimension of the feature map output by the convolutional layer while retaining the most important features. By stacking multiple convolutional layers and pooling layers, CNN can gradually extract more advanced and abstract features. After the previous convolutional layers and pooling layers extract features, these features are usually integrated using a fully connected layer to obtain the extracted features for classification or regression tasks.

[0051] Step 106, inputting the fault feature into an extreme gradient boosting classifier to obtain a fault diagnosis result of the gearbox; wherein the extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, and the improved cane rat algorithm combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms.

[0052] eXtreme Gradient Boosting (XGBoost) is an efficient machine learning algorithm based on the Gradient Boosting Decision Trees (GBDT) framework. XGBoost can effectively improve the classification accuracy when processing complex feature data, but in order to further optimize the performance of XGBoost, the improved Greater Cane Rat Algorithm (IGCRA) was introduced to optimize its parameters. The IGCRA algorithm combines nonlinear weights, adaptive factors, sine-cosine strategies and sparrow alert mechanisms on the basis of the original Greater Cane Rat Algorithm (GCRA), providing higher parameter optimization accuracy and global optimization capabilities, thereby improving the overall accuracy and generalization ability of fault diagnosis.

[0053] In this embodiment, the improved giant cane rat algorithm optimizes five important parameters of the XGBoost classifier, including the maximum number of trees, the maximum depth of the tree, the minimum sum of child node weights, the learning rate, and the sample sampling ratio.

[0054] In one embodiment, obtaining fault data of a gearbox includes: obtaining a one-dimensional vibration signal of the faulty gearbox in a working state; and converting the one-dimensional vibration signal into two-dimensional fault data by using a Gram angular field difference.

[0055] An acceleration sensor is installed on the gearbox to collect one-dimensional vibration signals of the gears in different operating states. Common types of gearbox faults include gear cracks, broken teeth, tooth root wear, tooth surface wear, etc. After collecting the one-dimensional vibration signal, the time series data - one-dimensional vibration signal is imaged using Gram angular field difference (GADF, Globalized Average Dynamic Frequency), which is conducive to subsequent CNN feature extraction. Therefore, combining the two-dimensional convolutional neural network to process the two-dimensional image data after GADF conversion can not only improve the accuracy of fault feature extraction, but also adaptively capture complex nonlinear relationships.

[0056] In this embodiment, the GADF conversion process includes preprocessing the original vibration signal, such as low-pass filtering and mean filtering, to improve the signal quality, and then mapping the processed signal to a two-dimensional space to highlight the change characteristics of the signal. Next, a two-dimensional convolutional neural network is used to extract deep features from the GADF image.

[0057] In one embodiment, the improved cane rat algorithm includes: generating the spatial position of the cane rat population; the spatial position of the cane rat population includes the spatial position of each cane rat; entering the iteration step, the iteration step includes: initializing the sine-cosine strategy coefficient, the nonlinear weight factor and the adaptive factor; generating a random number, and comparing the random number with a preset value; if the random number is less than the preset value, entering the exploration phase, using the adaptive factor and the first factor to update the current spatial position of the cane rat; if the random number is not less than the preset value, entering the development phase, using the adaptive factor and the second factor to update the current spatial position of the cane rat; evaluating the fitness value of the spatial position of the cane rat updated in the exploration phase or the development phase, and updating the spatial position of the cane rat based on the equation of the nonlinear weight factor and the sine-cosine strategy coefficient according to the evaluation result; comparing the fitness value of the updated spatial position of the cane rat with the current optimal fitness value, and updating the global optimal position according to the comparison result; when the number of iterations reaches the maximum number of iterations, outputting the global optimal position corresponding to the current iteration step; wherein the first factor is used to simulate the impact of abundant food sources, and the second factor is used to simulate the number of pups produced by cane rats each year.

[0058] like Figure 2 As shown in the figure, GCRA is a meta-heuristic optimization algorithm designed to solve global optimization problems by simulating the foraging behavior of cane rats in nature. GCRA is inspired by the behavioral differences between cane rats in the mating season and the non-mating season. The algorithm mainly consists of two phases: exploration phase and exploitation phase. In the exploration phase, cane rats move widely in their territory to find food. This process simulates the global search in the algorithm to identify multiple potential optimization solutions in the search space. In the exploitation phase, male cane rats leave the group and concentrate on foraging in areas rich in resources, simulating the local search process of the algorithm to further optimize the quality of the solution.

[0059] Based on GCRA, this embodiment proposes an improved Great Cane Rat Algorithm, which combines nonlinear weights, adaptive factors, sine and cosine functions, and sparrow alert strategies to enhance the global and local search capabilities of the algorithm. IGCRA optimizes the settings of weights and adaptive factors so that the algorithm can better balance the relationship between exploration and exploitation at different stages. In addition, the introduction of sine and cosine functions helps to enhance the diversity of the algorithm, while the sparrow alert strategy further improves the convergence performance of the algorithm in complex optimization problems.

[0060] The schematic diagram of IGCRA process is as follows Figure 3 First, the optimization process formula of IGCRA is as follows:

[0061]

[0062] Where X is the space of the entire cane rat population, n and d are the population size and the dimension of the problem, respectively.

[0063] The value x of an individual cane rat at position i and dimension j i,j It is randomly generated and the calculation formula is as follows:

[0064] x i,j =rand×(ub j -lb j )+lb j (2)

[0065] In the formula, rand is a random number uniformly distributed on [0, 1], ub j and lb j are the upper and lower limits of the j-th dimension respectively.

[0066] Secondly, the preset value ρ of the IGCRA algorithm is initialized. In this embodiment, it is set to 0.5 to control the balance of the algorithm in the exploration and development process. In addition, the initial solution and the maximum number of iterations of the algorithm are initialized so that they can be adjusted during the entire optimization process. The initial solution of the algorithm is to initialize the position of the cane rat in the search space through formula (2). The fitness value of each cane rat is calculated, and the cane rat with the best fitness value is selected as the dominant male x K , whose position represents the best solution found so far. The best solution refers to the best position of the cane rat in the search space, that is, the position of the dominant male. The algorithm compares the current best fitness value with the fitness value of the global optimal solution. If the current best fitness value is better, the global optimal solution is updated to the current optimal solution. K The calculation formula for updating the position of the remaining cane rats is as follows:

[0067]

[0068] In the formula, is the new position of the giant cane rat, x i,j is the current position of the giant cane rat, x K,j is the dominant male giant cane rat in the jth dimension.

[0069] Finally, the main loop is entered. When the number of iterations is less than the maximum number of iterations, the sine and cosine strategy coefficients r1, r2, r3, the nonlinear weight factor Ω and the adaptive factor ω are initialized. The calculation formula is as follows:

[0070]

[0071] In formula (4) to formula (6), l is the current number of iterations, Max iteris the maximum number of iterations, π is the ratio of pi, rand is a random number uniformly distributed on [0,1], and e is the base of the natural logarithm.

[0072] In each iteration, the introduced sparrow vigilant mechanism is used to update the position of each cane rat to enhance the global and local search capabilities of the algorithm. Specifically, the mechanism determines the update strategy by comparing the current fitness value of the cane rat with the current best fitness value. When the fitness value of a cane rat is poor, the cane rat will search towards the global best position, and its position update formula adds Gaussian noise to increase the randomness and exploration of the search; when the fitness value of the cane rat is good, its position update is based on the deviation between the current cane rat position and the population mean to improve the accuracy and development of the search. In addition, in a more preferred embodiment, the updated position can be constrained by boundary conditions to ensure that all cane rats remain within the defined search space. Finally, the algorithm evaluates the fitness value of the updated cane rat position. If the new fitness value is better than the global best fitness value, the algorithm's global best position and global best fitness value are updated. If the new fitness value is better than the current fitness value of the cane rat individual, the individual's best position and fitness will also be updated. This mechanism ensures that the individual position changes towards the global optimal direction, thereby enhancing the algorithm's search ability. In this way, the sparrow vigilant mechanism can effectively balance the algorithm's global exploration and local development, improving the efficiency and effectiveness of optimization.

[0073] In each iteration, a random number is generated. If the random number is less than the preset value ρ, the process enters the exploration phase, otherwise it enters the development phase.

[0074] During the exploration phase, cane rats will build their nests (caves or shallow holes) scattered throughout their territories (such as swamps, riverbanks, and farmlands). They will leave different nests to forage, either following the path of previous food sources or looking for new food sources everywhere, leaving traces. Cane rats search for food everywhere in their territory. The cane rats that appear to be walking represent the different locations of the dominant bamboo rats, while the cane rats that appear to be eating represent the food source. Assuming that the dominant male retains the information of these paths, other cane rats will adjust their positions based on this data. In the search space, the new position of the remaining cane rat group is determined based on the position of the dominant male. If the adaptive factor is added, the calculation formula is as follows:

[0075]

[0076] In the formula, is the value of the cane rat in the jth dimension, ω is the adaptive factor, x i,j is the current position of the cane rat, C is a random number defined within the boundary of the problem space, simulating scattered food sources and dens, xK,j is the dominant male giant cane rat in the jth dimension, and r is the first factor, which is used to simulate the effect of abundant food sources, thereby prompting more development and utilization.

[0077] At this stage of the cane rat movement simulation, if the fitness value of another cane rat exceeds the fitness value of the strongest cane rat, the strongest cane rat is updated and the positions of the other cane rats are adjusted according to the newly determined strongest cane rat. If it does not exceed, they will deviate from the position of the strongest bamboo rat. The calculation formula for the movement strategy of this cane rat is as follows:

[0078]

[0079] Where, X i is the next or new state of the ith cane rat, x i,j is the position of the ith cane rat in the jth dimension, x m, j is the position of the mth cane rat in the jth dimension, x K,j is the dominant male in the jth dimension, and C is a random number defined within the boundary of the problem space, simulating scattered food sources and dens.

[0080] In formula (7), the calculation formula of r is as follows:

[0081]

[0082] In the formula, is the fitness value of the dominant male cane rat, is the current fitness value of the cane rat, C iter Indicates the current number of iterations, Max iter Indicates the maximum number of iterations.

[0083] In formula (8), α is a coefficient that simulates the reduction of food sources and forces the search for new food sources or nests. Its calculation formula is as follows:

[0084] α=2×r×rand-r (10)

[0085] In formula (8), β is the coefficient that encourages the greater cane rat to move to other available abundant food sources in the breeding area, and its calculation formula is as follows:

[0086] β=2×r×μ-r (11)

[0087] Where μ is the second factor, which takes a random value from 1 to 4 and is used to simulate the number of pups produced by each female cane rat each year.

[0088] The last step of the exploration phase shows that the rat will move to the newly calculated position only if the fitness value of the new position is improved; otherwise, the rat will remain at the previous position.

[0089] In the development phase, the breeding season varies depending on the habitat and usually occurs during the rainy season. During the breeding season, males usually separate from the group. It is assumed that once the group separates, foraging activities will be concentrated in areas with abundant food, and different small groups will only forage in promising areas. The simulation of this stage begins by randomly selecting a female m, and m ≠ k (dominant male). Since breeding occurs in areas with abundant food, reinforcement is carried out around the selected female. The modeling calculation formula for this process is as follows:

[0090]

[0091] In the formula, x m,j represents the position of a randomly selected female on the jth dimension.

[0092] If the newly calculated position of the cane rat improves the fitness value, it will replace the previous position. As the number of iterations increases, the algorithm continuously updates the fitness value of the cane rat and continuously adjusts the global optimal solution based on the new solution. After each iteration, a new dominant male x is selected. K , to further improve the quality of the solution. Finally, when the predetermined maximum number of iterations is reached, the algorithm outputs the global optimal solution, which is the optimal solution to the problem.

[0093] During each iteration, the parameters C, r, μ, α, and β are responsible for achieving better exploration and exploitation throughout the iteration process.

[0094] In one embodiment, a gearbox fault classification and identification model based on GADF-CNN-IGCRA-XGBoost is proposed, such as Figure 4 As shown, it aims to improve the fault classification accuracy in complex mechanical systems and address the limitations of traditional methods in dealing with multi-dimensional features and nonlinear optimization.

[0095] First, in order to fully explore the characteristic information of the gearbox under different fault conditions, the one-dimensional vibration signal is converted into a two-dimensional image through GADF. The GADF conversion process includes preprocessing the original vibration signal, such as low-pass filtering and mean filtering, to improve the signal quality, and then mapping the processed signal to a two-dimensional space to highlight the changing characteristics of the signal. Next, a two-dimensional CNN is used to extract deep features from the GADF image. The CNN network structure contains multiple convolutional layers and pooling layers, where the first convolutional layer uses 32 3x3 convolutional kernels, the second convolutional layer contains 64 3x3 convolutional kernels, and the third convolutional layer contains 128 3x3 convolutional kernels. Each convolutional layer is followed by a maximum pooling layer (2x2). In order to deal with the degradation problem in the deep network, the residual network (ResNet) structure is introduced into the network. Each residual block includes a convolutional layer and a skip connection to ensure the effective transmission of information in the network. In addition, in order to improve the performance of the CNN model, batch normalization and Dropout techniques are combined. Batch normalization is used to accelerate training and improve the generalization ability of the model, while Dropout technology helps reduce the risk of overfitting. The output of the fully connected layer of CNN is used as the input of the XGBoost classifier. XGBoost can effectively improve the classification accuracy when processing complex feature data, but in order to further optimize the performance of XGBoost, the study introduced the IGCRA algorithm, which combines nonlinear weights, adaptive factors, sine and cosine strategies, and sparrow warning mechanisms to provide higher parameter optimization accuracy and global optimization capabilities.

[0096] In practical applications, the trained GADF-CNN-IGCRA-XGBoost model was deployed to an industrial control computer (ICC). The system is able to receive vibration signals from the gearbox in real time and perform fault identification and diagnosis through an embedded model. The data is uploaded to the IoT cloud platform via 4G communication for remote monitoring and further analysis. The system has been tested for reliability to ensure stability and accuracy in actual industrial environments. Ultimately, the study demonstrated the practicality and promotion value of the model in industrial gearbox fault monitoring and diagnosis.

[0097] To illustrate the superiority of the IGCRA algorithm proposed in this application, it is tested on 12 benchmark functions provided by CEC2022 together with GCRA, Coati Optimization Algorithm (COA), Subtraction Average Based Optimizer (SABO), Sine cosine algorithm (SCA) and Osprey Optimization Algorithm (OOA). These benchmark functions include unimodal, multimodal, mixed and composite functions, covering different types of optimization problems. The local optimization performance of the algorithm is evaluated using the unimodal function F1, the global optimization performance of the algorithm is evaluated using the multimodal functions F2-F5, the global and local optimization performance of the algorithm is evaluated using the mixed functions F6-F8, and the comprehensive performance of the algorithm is evaluated using the composite and multimodal functions F9-F12. For each algorithm, the population size, maximum number of iterations and problem dimension are set to 30, 500 and 10 respectively. A comprehensive evaluation is performed by executing each test function 30 times. The optimal value and median value of the fitness value are used as evaluation indicators to examine the performance of the algorithm in terms of optimization accuracy and convergence speed. The test results are shown in Table 1 and the box plots are shown in Table 1. Figure 5 shown.

[0098] Table 1 Test results of six algorithms on CEC2022 benchmark function

[0099]

[0100]

[0101] The results show that the optimal fitness value of the IGCRA algorithm on the multi-peak function F2 is 400.2709269, which is closest to the theoretical minimum value of the function 400, while the optimal values ​​of the GCRA, COA, SABO, SCA, and OOA algorithms are 1212.27102, 734.3918511, 404.3971027, 437.9567762, and 497.895328, respectively, and the median value of the IGCRA algorithm is 408.9161019, which is closest to 400 among the medians of these six algorithms. On the composite and multimodal function F9, the optimal fitness value of the IGCRA algorithm is 2529.284383, which is closest to the theoretical minimum value of 2300, while the optimal values ​​of other algorithms are 2660.824467, 2642.340889, 2557.545797, 2542.686565, and 2677.950235, respectively. The median value of the IGCRA algorithm is 2529.284383, which is closest to 2300 among the medians of these six algorithms. The optimization accuracy results of the six algorithms in the 12 benchmark function tests of CEC2022 show that the optimal fitness value of the IGCRA algorithm is the best among the six algorithms, and the median fitness value is slightly inferior to the SCA algorithm in the F10 function test, while the others are the best among the six algorithms. Therefore, the IGCRA algorithm shows relatively excellent optimization accuracy and robustness. Figure 6 and Figure 7 The convergence curves of the six algorithms on representative functions are shown, with a maximum number of iterations of 500. The horizontal axis is the number of iterations, and the vertical axis is the fitness value. It can be further seen that the IGCRA algorithm also has a relatively good convergence speed. IGCRA has achieved a good balance between global exploration and local development, effectively improving the algorithm's search ability and convergence speed, and providing a new solution to complex global optimization problems.

[0102] In order to further illustrate the superiority of the GADF-CNN-IGCRA-XGBoost gearbox fault classification and identification model proposed in this application over traditional identification methods, a model test and comparative analysis of gearbox fault diagnosis is carried out based on a public gearbox dataset. The gearbox test bench consists of a motor, a motor controller, a planetary gearbox, a reduction gearbox, a brake, and a brake controller. The dataset contains gears in five different operating states, including the following types of faults: defect, which refers to cracks on the gears during operation; broken teeth, which refers to the breakage or missing of some teeth on the gears; root wear, which refers to cracks or damage on the root of the teeth; tooth surface wear, which refers to obvious wear marks on the surface of the gear. In addition, the dataset also includes a normal operating state, which means that the gears are operating normally without any damage. By analyzing these data, different types of gear faults can be effectively identified and classified, which provides an important theoretical basis for the condition monitoring and maintenance of gear systems, such as Figure 8 shown.

[0103] The compared models include GADF-XGBOOST, GADF-GCRA-XGBOOST, GADF-IGCRA-XGBOOST, GADF-CNN-XGBOOST, GADF-CNN-GCRA-XGBOOST and GADF-CNN-IGCRA-XGBOOST. All models use the GADF method to convert the one-dimensional vibration signal into a two-dimensional image to improve the effect of feature extraction. During the experiment, each model conducted 10 independent fault diagnosis experiments, and finally took the average value of these experiments for comparative analysis. Figures 9 to 14 shown.

[0104] The experimental results show that there are significant differences in the classification accuracy of different models. First, when the GADF-XGBOOST model directly uses the features extracted by GADF for classification, the classification accuracy of the test set is 61.75%, which is the lowest among all models. By optimizing the key parameters of XGBOOST through the GCRA algorithm, the classification accuracy of the GADF-GCRA-XGBOOST model is improved to 69%. After further optimization, the GADF-IGCRA-XGBOOST model achieves a classification accuracy of 74.25% through the IGCRA algorithm. Compared with the aforementioned models, the classification accuracy of the GADF-CNN-XGBOOST model that introduces CNN for deep feature extraction is significantly improved, reaching 95.25%. The classification accuracy of the GADF-CNN-GCRA-XGBOOST model further combined with the GCRA algorithm is improved to 97%. Among all the compared models, the GADF-CNN-IGCRA-XGBOOST model performs best, with a test set classification accuracy of 98.75%, demonstrating excellent fault diagnosis capabilities.

[0105] In summary, the classification accuracy from low to high is: GADF-XGBOOST, GADF-GCRA-XGBOOST, GADF-IGCRA-XGBOOST, GADF-CNN-XGBOOST, GADF-CNN-GCRA-XGBOOST and GADF-CNN-IGCRA-XGBOOST. The experimental results show that the combination of GADF image features, CNN deep feature extraction and IGCRA optimization strategy can significantly improve the fault diagnosis accuracy and reliability based on the Southeast University gearbox dataset.

[0106] The present invention proposes a gearbox fault diagnosis model based on GADF-CNN-IGCRA-XGBoost, which comprehensively utilizes technologies such as GADF, CNN and XGBoost, and optimizes model parameters through the IGCRA algorithm to improve the accuracy of fault diagnosis. The specific diagnosis method is as follows: First, the one-dimensional vibration signal is converted into a two-dimensional image through GADF, and the model can effectively extract the spatial features of the gearbox fault data. CNN further explores these features, thereby improving the quality of feature representation. Finally, the XGBoost classifier achieves accurate fault classification based on these high-quality features. Secondly, the IGCRA algorithm is introduced to optimize the hyperparameters of XGBoost. IGCRA combines nonlinear weights, adaptive factors, sine and cosine and sparrow warning mechanism strategies, and performs well in optimization ability, stability and convergence speed, effectively improving the overall performance of the model. Finally, after training and testing using a public gearbox dataset, the model shows high fault diagnosis accuracy and stability. Experimental results show that the GADF-CNN-IGCRA-XGBoost model achieves a classification accuracy of 98.75% in gearbox fault diagnosis, providing reliable technical support for intelligent monitoring and fault prevention of industrial equipment. Through the above research, the proposed model demonstrates its superiority in fault diagnosis of complex mechanical systems and can provide an effective solution for intelligent monitoring in practical industrial applications.

[0107] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0108] Based on the same inventive concept, the embodiment of the present application also provides a gearbox fault diagnosis system for implementing the gearbox fault diagnosis method involved above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more gearbox fault diagnosis system embodiments provided below can refer to the limitations of the gearbox fault diagnosis method above, and will not be repeated here.

[0109] In one embodiment, Fig.15 As shown, a gearbox fault diagnosis system is provided, comprising:

[0110] The data acquisition module 1502 is used to obtain the fault data of the gearbox;

[0111] The feature extraction module 1504 is used to extract features from the fault data using a convolutional neural network to obtain fault features;

[0112] A fault diagnosis module 1506 is used to input the fault feature into the extreme gradient boosting classifier to obtain a fault diagnosis result of the gearbox;

[0113] Among them, the extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, which combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms.

[0114] Each module in the above gearbox fault diagnosis system can be implemented in whole or in part by software, hardware and a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0115] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in all the above method embodiments when executing the computer program.

[0116] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in all the above method embodiments are implemented.

[0117] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in all the above method embodiments when executed by a processor.

[0118] It should be noted that the user information (including but not limited to user device information, user 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 user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0119] 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 (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive 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.

[0120] 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.

[0121] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A gearbox fault diagnosis method, characterized in that: The method comprises: Obtain gearbox fault data; Using a convolutional neural network to extract features from the fault data to obtain fault features; Inputting the fault feature into an extreme gradient boosting classifier to obtain a fault diagnosis result of the gearbox; The extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, and the improved cane rat algorithm combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms.

2. The method according to claim 1, characterized in that The obtaining of the fault data of the gearbox comprises: Acquire a one-dimensional vibration signal of the faulty gearbox in a working state; The one-dimensional vibration signal is converted into two-dimensional fault data using Gram angular field difference.

3. The method according to claim 1, characterized in that The improved cane rat algorithm comprises: Generate the spatial position of a cane rat population; the spatial position of the cane rat population includes the spatial positions of each cane rat; determine a dominant male cane rat according to the fitness value of each cane rat, and update the spatial position of the cane rat population based on the position of the dominant male; Entering the iteration step, the iteration step includes: initializing the sine-cosine strategy coefficient, the nonlinear weight factor and the adaptive factor; generating a random number, and comparing the random number with a preset value; if the random number is less than the preset value, entering the exploration phase, using the adaptive factor and the first factor to update the spatial position of the great cane rat population; if the random number is not less than the preset value, entering the development phase, using the adaptive factor, the first factor and the second factor to update the spatial position of the great cane rat population; sequentially evaluating whether the fitness value of each great cane rat in the great cane rat population updated in the exploration phase is better than the fitness value of the dominant male, if so, updating the current optimal fitness value and the dominant male position, if not, updating the current great cane rat position according to the nonlinear weight factor, the sine strategy coefficient and the first factor, and at the current great cane rat position After the position of the large cane rat is updated, check whether the current position of the large cane rat meets the boundary conditions, and recalculate the fitness value of the current large cane rat. If it is better than the current optimal fitness value, update the current optimal fitness value and the dominant male position again; sequentially evaluate whether the fitness values ​​of the large cane rats in the large cane rat population updated in the development stage are better than the fitness value of the dominant male. If so, update the current optimal fitness value and the dominant male position; if not, update the current position of the large cane rat according to the nonlinear weight factor, the cosine strategy coefficient, the first factor and the second factor, and after the current position of the large cane rat is updated, check whether the current position of the large cane rat meets the boundary conditions, and recalculate the fitness value of the current large cane rat. If it is better than the current optimal fitness value, update the current optimal fitness value and the dominant male position again; When the number of iterations reaches the maximum number of iterations, outputting the current best fitness value and the dominant male position corresponding to the current iteration step; The first factor is used to simulate the impact of abundant food sources, and the second factor is used to simulate the number of pups produced by cane rats each year.

4. The method according to claim 3, characterized in that The sine-cosine strategy coefficient is related to the current number of iterations, the maximum number of iterations and the random number, and the nonlinear weight factor and the adaptive factor are both related to the current number of iterations and the maximum number of iterations.

5. The method according to claim 3, characterized in that: After the updating of the current spatial position of the great cane rat, the method further comprises: Boundary conditions are used to constrain all of the cane rats to remain within the search space.

6. The method according to claim 1, characterized in that The convolutional neural network introduces residual connections.

7. A gearbox fault diagnosis system, characterized in that: The system comprises: A data acquisition module, used to obtain fault data of the gearbox; A feature extraction module, used to extract features from the fault data using a convolutional neural network to obtain fault features; A fault diagnosis module, used for inputting the fault feature into an extreme gradient boosting classifier to obtain a fault diagnosis result of the gearbox; The extreme gradient boosting classifier uses an improved cane rat algorithm for parameter optimization, and the improved cane rat algorithm combines nonlinear weights, adaptive factors, sine and cosine, and sparrow warning mechanisms.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are 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 steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images