CN-121980152-A - Bearing life prediction method and device based on improved dynamic convolution and IGOOSE algorithm

Abstract

The application discloses a bearing life prediction method and device based on improved dynamic convolution and IGOOSE algorithm, and relates to the field of bearings. The method comprises the steps of constructing an improved dynamic convolution model, enabling the improved dynamic convolution model to introduce a multi-head attention mechanism and fusing residual errors, obtaining a bearing degradation data set, inputting the bearing degradation data set into the improved dynamic convolution model, conducting super-parameter optimization on the improved dynamic convolution model by adopting a IGOOSE algorithm, and conducting service life prediction on a target rolling bearing by utilizing the optimized improved dynamic convolution model. According to the scheme, a multi-head attention mechanism and residual fusion are introduced in the improved dynamic convolution model, so that the interactive modeling capacity of the channel and the time sequence features is enhanced, and the super-parameters of the improved dynamic convolution model are optimized by IGOOSE, so that the prediction precision and stability are improved.

Inventors

• CUI YEMEI
• DAI JIANHUA

Assignees

• 无锡商业职业技术学院

Dates

Publication Date

20260505

Application Date

20251118

Claims (10)

1. 1. A method for predicting bearing life based on improved dynamic convolution and IGOOSE algorithm, the method comprising: The method comprises the steps of constructing an improved dynamic convolution model, wherein the improved dynamic convolution model comprises an attention mechanism part and a convolution kernel fusion part, the attention mechanism part sequentially comprises an average pooling layer, an FC1 convolution layer, an activation layer, an FC2 convolution layer, a LayerNorm layer, a Softmax layer, a Dropout layer and a multi-head average fusion layer, the FC1 convolution layer expands single-head attention into multi-head attention, the LayerNorm layer independently normalizes each head, the Softmax layer adopts an exponentially decaying temperature parameter self-adaptive updating strategy, and the output of the convolution kernel fusion part adopts residual connection; Acquiring a bearing degradation data set, inputting the bearing degradation data set into the improved dynamic convolution model, and performing super-parameter optimization on the improved dynamic convolution model by adopting IGOOSE algorithm; and predicting the service life of the target rolling bearing by using the optimized improved dynamic convolution model.
2. 2. The method of claim 1, wherein the IGOOSE algorithm comprises: updating the current position of an individual in a population in a development stage or an exploration stage, obtaining a candidate solution, correcting the candidate solution exceeding the search space range, introducing a sparrow warning mechanism, comparing the fitness of the candidate solution and a current optimal solution, updating the current optimal solution according to a comparison result, and entering the next iteration updating operation based on the current position of the individual and the current optimal solution.
3. 3. The method of claim 2, wherein prior to the iterative updating operation, the method further comprises: And initializing the population by adopting a Tent chaotic mapping mechanism, wherein the chaotic coefficient takes a value of 1.1.
4. 4. A method according to claim 3, wherein after the original population is obtained by population initialization, the method further comprises: generating a reverse population for the original population by adopting a reverse learning mechanism; Combining the original population and the reverse population into a combined population, and selecting an optimal individual position from the combined population according to fitness; Performing a variation operation on each individual in the combined population to construct a variation population; performing cross operation on the combined population and the variant population to construct a cross population; if the individuals in the cross population are out of range in the corresponding dimension, the dimension value of the out-of-range individuals is replaced by the dimension value corresponding to the optimal individual position, and the individual position and the individual fitness are updated through a greedy selection strategy.
5. 5. The method of claim 2, wherein during each iteration, the method further comprises: the individual locations are updated using the Kexil distribution perturbation and the Gaussian distribution perturbation.
6. 6. The method of claim 5, wherein the updating the expression of the individual location using the cauchy distribution perturbation and the gaussian distribution perturbation is: Wherein, the The individual position of the ith individual in the jth dimension at the t-th iteration; The individual position is updated by the Kexil distribution disturbance and the Gaussian distribution disturbance; the method is used for generating a Kexil distribution disturbance; For random numbers obeying a standard normal distribution N (0, 1), for generating gaussian distribution perturbations; is the maximum number of iterations.
7. 7. A bearing life prediction apparatus based on improved dynamic convolution and IGOOSE algorithm, the apparatus comprising: The system comprises a construction module, a calculation module and a calculation module, wherein the construction module is used for constructing an improved dynamic convolution model, the improved dynamic convolution model comprises an attention mechanism part and a convolution kernel fusion part, the attention mechanism part sequentially comprises an average pooling layer, an FC1 convolution layer, an activation layer, an FC2 convolution layer, a LayerNorm layer, a Softmax layer, a Dropout layer and a multi-head average fusion layer, the FC1 convolution layer expands single-head attention into multi-head attention, the LayerNorm layer independently normalizes each head, the Softmax layer adopts an exponentially decayed temperature parameter self-adaptive update strategy, and the output of the convolution kernel fusion part adopts residual connection; The optimization module is used for acquiring a bearing degradation data set, inputting the bearing degradation data set into the improved dynamic convolution model, and performing super-parameter optimization on the improved dynamic convolution model by adopting IGOOSE algorithm; And the prediction module is used for predicting the service life of the target rolling bearing by using the optimized improved dynamic convolution model.
8. 8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 6 when the computer program is executed.
9. 9. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.
10. 10. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 6.

Description

Bearing life prediction method and device based on improved dynamic convolution and IGOOSE algorithm Technical Field The application relates to the technical field of bearings, in particular to a bearing service life prediction method and device based on improved dynamic convolution and IGOOSE algorithm. Background Along with the industrial intelligent promotion, the rolling bearing faces complex working conditions such as multiple disturbance, non-stable signals and the like in milling and mechanical systems, and the abrasion prediction becomes an important challenge for guaranteeing the processing precision and the service life of equipment. Conventional methods have difficulty capturing dynamic changes in the wear process, and life prediction techniques based on deep learning have therefore received attention. The convolutional neural network CNN is a mainstream model for time sequence signal processing and fault prediction due to the outstanding feature extraction capability. However, the conventional CNN adopts a fixed convolution kernel, which is difficult to adapt to a non-stationary timing signal, and limits further improvement of performance. Disclosure of Invention Based on the above, it is necessary to provide a method and a device for predicting the life of a bearing based on improved dynamic convolution and IGOOSE algorithm, so as to improve the prediction accuracy and stability. In a first aspect, the present application provides a method for predicting bearing life based on improved dynamic convolution and IGOOSE algorithm. The method comprises the following steps: the improved dynamic convolution model comprises an attention mechanism part and a convolution kernel fusion part, wherein the attention mechanism part sequentially comprises an average pooling layer, an FC1 convolution layer, an activation layer, an FC2 convolution layer, a LayerNorm layer, a Softmax layer, a Dropout layer and a multi-head average fusion layer, the FC1 convolution layer expands single-head attention into multi-head attention, the LayerNorm layer independently normalizes each head, the Softmax layer adopts an exponentially decayed temperature parameter self-adaptive update strategy, and the output of the convolution kernel fusion part adopts residual error connection; Acquiring a bearing degradation data set, inputting the bearing degradation data set into an improved dynamic convolution model, and performing super-parameter optimization on the improved dynamic convolution model by adopting IGOOSE algorithm; And predicting the service life of the target rolling bearing by using the optimized improved dynamic convolution model. In one embodiment, the IGOOSE algorithm includes: Updating the current position of an individual in the population in a development stage or an exploration stage, obtaining a candidate solution, correcting the candidate solution exceeding the search space range, introducing a sparrow warning mechanism, comparing the fitness of the candidate solution and the current optimal solution, updating the current optimal solution according to the comparison result, and entering the next iteration updating operation based on the current position of the individual and the current optimal solution. In one embodiment, prior to the iterative update operation, the method further comprises: And initializing the population by adopting a Tent chaotic mapping mechanism, wherein the chaotic coefficient takes a value of 1.1. In one embodiment, after obtaining the original population through population initialization, the method further comprises: generating a reverse population for the original population by adopting a reverse learning mechanism; combining the original population and the reverse population into a combined population, and selecting the optimal individual position from the combined population according to the fitness; performing a mutation operation on each individual in the combined population to construct a mutated population; performing cross operation on the combined population and the variant population to construct a cross population; If the individuals in the cross population are out of range in the corresponding dimension, the dimension value of the out-of-range individuals is replaced by the dimension value corresponding to the optimal individual position, and the individual position and the individual fitness are updated through a greedy selection strategy. In one embodiment, during each iteration, the method further comprises: the individual locations are updated using the Kexil distribution perturbation and the Gaussian distribution perturbation. In one embodiment, the expression for updating the individual location using the cauchy distribution perturbation and the gaussian distribution perturbation is: Wherein, the The individual position of the ith individual in the jth dimension at the t-th iteration; The individual position is updated by the Kexil distribution disturbance and the G