Method for predicting fatigue life of defect point of hollow bolt

Abstract

The invention discloses a hollow bolt defect point fatigue life prediction method, and the method specifically comprises the steps: S1, obtaining hollow bolt structure detection data, inputting the data to a three-dimensional defect model, and outputting a defect detection value; s2, inputting the defect detection value into a defect prediction model, outputting a damage evaluation value of the hollow bolt, and comparing the damage evaluation value with a preset threshold value for screening and classification; s3, hollow bolt image data after screening and classification are acquired, and an initial data set is constructed; s4, through the initial data set, constructing a three-structure-fatigue life prediction model for the screened and classified hollow bolts; s5, using transfer learning to train the three-structure-fatigue life prediction model, and optimizing model parameters; and S6, inputting data of defect points of the hollow bolt to be predicted into the three-structure-fatigue life prediction model, and outputting a life interval and a risk level. According to the method, multi-dimensional detection of the hollow bolt is realized, and the defect state and the fatigue life are accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of hollow bolt life prediction, and specifically to a method for predicting the fatigue life of defect points of hollow bolts. Background Art

[0002] In modern industrial production, hollow bolts, as a key type of connecting component, are widely used in many fields such as aerospace, automotive manufacturing, and mechanical engineering. These fields have extremely high requirements for the safety and reliability of equipment, and the performance of hollow bolts is directly related to the stable operation of the entire system. However, during long-term service, various defects will inevitably occur in hollow bolts, such as internal cracks, wear, corrosion, etc. These defects will significantly reduce the fatigue life of hollow bolts, and further lead to serious safety accidents and huge economic losses. Therefore, accurately predicting the fatigue life of defect points of hollow bolts is of crucial significance for ensuring the safe operation of equipment, and for carrying out maintenance and replacement in advance. This is also the core reason for inventing this technology.

[0003] Traditional methods for predicting the fatigue life of defect points of hollow bolts are mainly based on empirical formulas and simple mechanical models. These methods usually establish prediction formulas for fatigue life based on the basic mechanical properties of materials, such as yield strength, tensile strength, etc., combined with the geometric dimensions of the bolts. In practical applications, some macroscopic parameters of the bolts, such as outer dimensions, surface roughness, etc., are measured and substituted into the formula to calculate the fatigue life. The advantage of this method is that the calculation is relatively simple and the requirements for computing resources are relatively low, and it has certain application value in some scenarios with low precision requirements and relatively simple working conditions. However, its disadvantages are also very obvious. Due to overly simplifying the actual situation, it does not fully consider the complex internal structural characteristics of hollow bolts, the microscopic characteristics of materials, and the diversity of the actual service environment, resulting in a large deviation between the prediction results and the actual situation. For example, it cannot accurately reflect the influence of factors such as internal residual stress and material density gradient distribution in the bolts on the fatigue life, and the prediction accuracy is even more greatly reduced when facing complex working conditions, such as high temperature, high humidity, and strong corrosion environments.

[0004] With the continuous development of technology, certain progress has been made in the prior art in predicting the fatigue life of defect points of hollow bolts. Some advanced detection technologies, such as ultrasonic detection and X-ray detection, are applied to obtain the defect information of hollow bolts, and then numerical simulation methods such as finite element analysis are combined to predict the fatigue life. Compared with traditional methods, this way can consider the structure and stress conditions of bolts in more detail and improve the prediction accuracy to a certain extent. Microdefects inside the bolts can be detected through ultrasonic detection, and the stress concentration at these defect sites can be simulated and calculated using finite element analysis to more accurately evaluate the fatigue life. However, there are still many deficiencies in the prior art. On the one hand, these detection technologies and numerical simulation methods have high requirements for equipment and operators, and the cost is also relatively expensive, which limits their wide application. On the other hand, there are still difficulties in processing multi-source data fusion and considering the comprehensive influence of various complex factors on the fatigue life, and it is difficult to comprehensively and accurately predict the fatigue life of hollow bolts under various complex working conditions. For example, the data obtained by different detection technologies have format differences and information redundancy, making it difficult to effectively fuse; when considering the coupled effects of multi-factors such as environmental factors and material aging, the accuracy and reliability of the model need to be improved.

[0005] Therefore, in view of the above problems, the present application proposes a method for predicting the fatigue life of defect points of hollow bolts, which solves the above existing technical problems. Summary of the Invention

[0006] Based on the above content, the present application proposes a method for predicting the fatigue life of defect points of hollow bolts, including the following steps:

[0007] S1. Obtain the structural detection data of the hollow bolt, input the structural detection data into a pre-constructed three-dimensional defect model, and output the defect detection value;

[0008] S2. Input the obtained defect detection value into a pre-constructed defect prediction model, output the failure evaluation value of the hollow bolt in the defect state, compare with a preset threshold, determine the failure state of the hollow bolt, and perform screening and classification;

[0009] S3. Obtain the image data of the screened and classified hollow bolts, extract the position, shape, and size information data of the defect points, and construct an initial data set;

[0010] S4. Through the initial data set, construct a three-structure - fatigue life prediction model for the screened and classified hollow bolts, and the three structures are the nut, the inner screw rod, and the outer screw rod structures;

[0011] S5. Obtain a large number of sample data of hollow bolts under different working conditions and different defect states, and use transfer learning to train the three-structure - fatigue life prediction model to optimize the model parameters;

[0012] S6. Input the data of the defect points of the hollow bolts to be predicted into the corresponding trained three-structure fatigue life prediction model after preprocessing and feature enhancement, and use the model to obtain the fatigue life prediction results of the defect points, and output the life interval and risk level.

[0013] Preferably, the hollow bolt structure detection data in S1 includes geometric dimension and shape deviation data of the internal hollow part of the bolt, microscopic profile of the thread profile and pitch cumulative error data, residual stress distribution data at the bolt shank and nut, and density gradient distribution data of the internal material of the hollow bolt.

[0014] Preferably, the three-dimensional defect model pre-constructed in S1 is constructed based on an improved convolutional neural network architecture. The input layer receives the hollow bolt structure detection data and converts it into a multi-dimensional feature matrix. where i d represents the data dimension, and j d represents the sample serial number; in the convolutional layer, convolution operations are performed through a custom convolutional kernel where m d , n d are the coordinate offsets of the elements of the convolutional kernel in the row direction of the input feature matrix X, and the convolution operation is where a d , b d are the size ranges of the convolutional kernel, p d , q d are the coordinates of the output feature map, is the eigenvalue at the position of the coordinate (p d , q d ) in the output feature map obtained after the convolution operation. The pooling layer adopts an adaptive pooling strategy to dynamically adjust the pooling area size according to the importance of the features. The feature vector is mapped to the defect feature space through the fully connected layer, and the model parameters are optimized by backpropagation through the loss function where is the true defect value, is the predicted defect value, s d is the number of samples, r d is the serial number of the sample, and a three-dimensional defect model is constructed.

[0015] Preferably, the defect prediction model in S2 is pre-constructed through multi-modal fusion and Bayesian regularized neural network, specifically:

[0016] Perform multi-modal partitioning on the defect detection values. The original defect detection value vector is Partition it into geometric feature modality mechanical property modality and material property modality where m u + k u + l u = n u ;

[0017] For each modality, a sub-neural network is constructed. The calculation formula from the input layer to the hidden layer of the geometric feature modality is: where is the weight matrix of the i-th hidden layer of the geometric feature modality, is the geometric feature bias vector, σ is the activation function. The calculation formula from the input layer to the hidden layer of the mechanical property modality is: where is the weight matrix of the i-th hidden layer of the mechanical property modality, is the mechanical property bias vector. The calculation formula from the input layer to the hidden layer of the material property modality is: where is the weight matrix of the i-th hidden layer of the material property modality, is the material property bias vector; The outputs of the hidden layers of each modality are fused to obtain a fused feature vector where I, J, and P are the numbers of the last hidden layers of each modality respectively; Bayesian regularization is used to optimize the fused network, and the output of the network is Y pred , and the true value is Y true , then the loss function L u is defined as: where N is the number of samples, W is the set of all weights of the network, α and β are the Bayesian regularization coefficients respectively. By iteratively updating the weights W, the loss function L is minimized to construct a defect prediction model for predicting the failure evaluation value of the hollow bolt in the defective state.

[0018] Preferably, the preset threshold in S2 obtains a basic threshold interval by obtaining the material properties of the hollow bolt. The material properties include the mechanical properties such as the yield strength and fatigue limit of the material; the actual service environment data of the hollow bolt is obtained to obtain the environmental adaptability threshold. The actual service environment data includes high temperature, high humidity, and strong corrosion environment data. The environmental adaptability threshold is associated with the basic threshold interval to dynamically correct the basic threshold interval to form a preset threshold.

[0019] Preferably, in S2, when comparing the preset threshold to determine the failure state of the hollow bolt, a fuzzy decision tree is used for screening and classification. Specifically:

[0020] The fuzzy decision tree algorithm is used to assign fuzzy membership degrees to the failure states of different hollow bolts. The nodes of the decision tree are divided according to the fuzzy membership degrees of the failure states. Each branch represents a trend of a failure state. For hollow bolts in different threshold intervals, according to the output results of the decision tree, they are divided into four categories: slight damage, moderate damage, severe damage, and failure.

[0021] Preferably, in the step S4, through the initial data set, a three-structure - fatigue life prediction model is constructed for the hollow bolts after screening and classification, specifically as follows:

[0022] The feature vectors of the nut, the inner screw rod, and the outer screw rod structures in the initial data set are obtained as where n s is the number of features. The feature vectors of each structure are weighted. The feature weight vectors of the nut, the inner screw rod, and the outer screw rod structures are respectively The weighted feature vectors are and where ⊙ represents element-wise multiplication; the weighted feature vectors are input into their respective sub-networks for feature extraction. The output of the hidden layer of the sub-network is: where W ht is the hidden layer weight matrix, b ht is the bias vector of the nut structure, f is the activation function, represents and The outputs of the three sub-networks are fused to obtain a fused feature vector, and the fatigue life prediction value L pred is output through the fully connected layer. The formula is L pred = f out (W out H + b out ), where W out is the fully connected layer weight matrix, b out is the bias vector, f out is the activation function of the output layer; by minimizing the loss function Loss between the predicted value L pred and the true fatigue life value L true , the model parameters are adjusted to construct a three-structure - fatigue life prediction model.

[0023] Preferably, in the step S5, a method combining transfer learning and reinforcement learning is used to train the three-structure - fatigue life prediction model, specifically as follows:

[0024] From the sample data of hollow bolts under a large number of different working conditions and different defect states, a basic data set corresponding to common working conditions and typical defect states is divided, and this basic data set is used to pre-train an initial three-structure fatigue life prediction model; when encountering data of new, special working conditions or complex defect states, there is no need to train the model from scratch. Some parameters of the pre-trained model are migrated as the initial parameters of the new model, and the three-structure fatigue life prediction model is learned in the data environment of different working conditions and defect states to accelerate the convergence speed of the three-structure fatigue life prediction model on new data.

[0025] Preferably, in step S6, the data of the defect points of the hollow bolt to be predicted is preprocessed and feature enhanced, specifically:

[0026] The preprocessing maps the defect point data with different value ranges to a specific interval. The defect point data includes

[0027] position coordinates, shape dimensions and stress distribution, eliminates the influence of data scale differences on the model, and removes random noise and outliers in the data; for feature enhancement, the original data is non-linearly transformed through data transformation, and the feature combination method is adopted to combine different types of original features to generate new composite features.

[0028] Preferably, in step S6, the fatigue life prediction results of the defect points are obtained by using the model, and the life interval and risk level are output. It is simulated by a large number of random samplings. The sampling data is input into the model to obtain a fatigue life prediction value. After multiple samplings, a set of prediction values is obtained. By combining prior knowledge and the prediction values obtained by sampling, the probability distribution of the fatigue life is updated to determine the life interval;

[0029] For the division of the risk level, a risk assessment system is constructed through the life interval and the characteristics of the defect points. The life interval and defect characteristics are used as input variables, and different weights are assigned according to the influence degree of the risk. According to the value range and combination of the input variables, the risks are divided into different levels. The risk levels can be set as low risk, medium-low risk, medium risk, medium-high risk and high risk in ascending order.

[0030] Compared with the prior art, the technical solution of the present application has the following technical effects:

[0031] The present invention solves the technical problem that it is difficult for traditional methods to accurately detect complex defects inside hollow bolts through a technical solution of obtaining various types of structural detection data such as geometric dimension and shape deviation data of the hollow part inside the bolt, microscopic profile of thread teeth, and pitch cumulative error data, and inputting them into a three-dimensional defect model constructed based on an improved convolutional neural network architecture. This model performs convolutional operations through custom convolutional kernels, adopts an adaptive pooling strategy and backpropagation to optimize parameters, can accurately output defect detection values, and achieves the technical effect of greatly improving the defect detection accuracy of hollow bolts, laying a solid foundation for accurately evaluating the bolt state subsequently.

[0032] The present invention constructs a defect prediction model through multimodal fusion and a Bayesian regularized neural network, and divides the defect detection values into geometric, mechanical property, and material property modalities for processing. This technical solution solves the problem that traditional prediction models cannot fully consider the influence of multiple factors on the failure of hollow bolts. By fusing the outputs of hidden layers of each modality and optimizing, it can output more accurate failure evaluation values. Comparing with a preset threshold dynamically corrected based on material properties and service environment, a fuzzy decision tree is used to screen and classify hollow bolts, achieving the technical effect of more accurately judging the failure state of hollow bolts and reasonably classifying them, which helps to take maintenance measures targeted.

[0033] The present invention solves the technical problem that previous prediction models did not fully consider the influence of different structures of hollow bolts on fatigue life through a technical solution of performing weighted processing on the feature vectors of the nut, inner screw, and outer screw structures in the initial dataset, then inputting them into their respective sub-networks to extract features and fuse them, and finally outputting fatigue life prediction values through a fully connected layer. At the same time, the model parameters are adjusted by minimizing the loss function to construct a three-structure - fatigue life prediction model. This model can comprehensively analyze different structural features and achieves the technical effect of more accurately predicting the fatigue life of defect points in different parts of hollow bolts, providing strong support for comprehensively evaluating the fatigue condition of bolts.

[0034] The present invention preprocesses the data of defect points of the hollow bolt to be predicted, maps the data to a specific interval and removes noise and outliers. At the same time, feature enhancement is performed through data transformation and feature combination, and then the data is input into the trained model. A life interval is determined by using a large number of random samplings combined with prior knowledge, and a risk assessment system is constructed based on multiple factors to divide risk levels. This technical solution solves the technical problem that the prediction result is single and cannot intuitively reflect the risk degree, achieving the technical effect of not only being able to give fatigue life prediction values, but also outputting a life interval and a risk level, providing more comprehensive and intuitive risk information for users and facilitating timely decision-making.

[0035] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, so as to be implemented in accordance with the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following will be described in detail with reference to the preferred embodiments of the present application and the accompanying drawings.

[0036] From the following detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings, those skilled in the art will become more clear about the above and other purposes, advantages and features of the present application. Brief Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to actual scale.

[0038] Figure 1 It is a flowchart of a method for predicting the fatigue life of defect points of a hollow bolt according to the present invention;

[0039] Figure 2 It is an implementation flowchart of a method for predicting the fatigue life of defect points of a hollow bolt according to the present invention;

[0040] Figure 3 It is a structural division diagram of a method for predicting the fatigue life of defect points of a hollow bolt according to the present invention;

[0041] Figure 4 It is a three-structure - fatigue life prediction model diagram of a method for predicting the fatigue life of defect points of a hollow bolt according to the present invention. Detailed Description of the Specific Embodiments

[0042] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. In the following description, specific details such as specific configurations and components are provided only to help a comprehensive understanding of the embodiments of the present application. Therefore, those skilled in the art should clearly understand that various changes and modifications can be made to the embodiments described here without departing from the scope and spirit of the present application. In addition, descriptions of known functions and structures are omitted for clarity and conciseness in the embodiments.

[0043] It should be understood that the "one embodiment" or "this embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, the "one embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner.

[0044] In addition, this application may repeat reference numerals and / or letters in different instances. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or arrangements discussed.

[0045] The term "and / or" in this article is merely a description of the associated relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another associated object relationship, indicating that there can be two relationships. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0046] The term "at least one" in this article is merely a description of the associated relationship of the associated objects, indicating that there can be three relationships. For example, at least one of A and B can mean: A exists alone, A and B exist simultaneously, and B exists alone.

[0047] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion.

[0048] Embodiment 1

[0049] This embodiment mainly describes a method for predicting the fatigue life of the defect points of a hollow bolt. As Figure 1 shown, it includes the following steps:

[0050] S1. Obtain the structural detection data of the hollow bolt, input the structural detection data into the pre-constructed three-dimensional defect model, and output the defect detection value;

[0051] S2. Input the obtained defect detection value into the pre-constructed defect prediction model, output the damage evaluation value of the hollow bolt in the defect state, compare it with the preset threshold, determine the damage state of the hollow bolt, and perform screening and classification;

[0052] S3. Obtain the image data of the hollow bolts after screening and classification, extract the data of the positions, shapes, and dimensions of the defect points, and construct an initial data set;

[0053] S4. Based on the initial data set, construct a three-structure fatigue life prediction model for the hollow bolts after screening and classification. The three structures are the nut, the inner screw rod, and the outer screw rod structures;

[0054] S5. Obtain a large number of sample data of hollow bolts under different working conditions and different defect states, and use transfer learning to train the three-structure fatigue life prediction model to optimize the model parameters;

[0055] S6. Input the data of the defect points of the hollow bolts to be predicted, after preprocessing and feature enhancement, into the corresponding trained three-structure fatigue life prediction model, and use the model to obtain the fatigue life prediction results of the defect points, and output the life interval and risk level.

[0056] Furthermore, the hollow bolt structure detection data in S1 includes the geometric dimension and shape deviation data of the hollow part inside the bolt, the microscopic profile of the thread profile and the pitch cumulative error data, the residual stress distribution data at the bolt rod and nut, and the density gradient distribution data of the material inside the hollow bolt.

[0057] Furthermore, the three-dimensional defect model pre-constructed in S1 is constructed based on an improved convolutional neural network architecture. The input layer receives the hollow bolt structure detection data and converts it into a multi-dimensional feature matrix where i d represents the data dimension, and j d represents the sample serial number; in the convolutional layer, convolution operations are performed through a custom convolution kernel where m d , n d are the coordinate offsets of the elements of the convolution kernel in the row direction of the input feature matrix X, and the convolution operation is where a d , b d are the size ranges of the convolution kernel, p d , q d are the coordinates of the output feature map, is the eigenvalue at the position of the coordinate (p d , q d ) in the output feature map obtained after the convolution operation. The pooling layer adopts an adaptive pooling strategy, dynamically adjusts the size of the pooling area according to the importance of the features, maps the feature vector to the defect feature space through the fully connected layer, and optimizes the model parameters through the loss function where is the true defect value, is the predicted defect value, and s dis the number of samples, r d is the serial number of the sample, and a three-dimensional defect model is constructed.

[0058] Furthermore, the defect prediction model in S2 is pre-constructed by multi-modal fusion and Bayesian regularization neural network, specifically as follows:

[0059] The defect detection values are divided into multi-modalities. The original defect detection value vector is divided into geometric feature modality mechanical property modality and material property modality where m u +k u +l u =n u ;

[0060] For each modality, a sub-neural network is constructed. The calculation formula from the input layer to the hidden layer of the geometric feature modality is: where is the weight matrix of the i-th hidden layer of the geometric feature modality, is the geometric feature bias vector, σ is the activation function. The calculation formula from the input layer to the hidden layer of the mechanical property modality is: where is the weight matrix of the i-th hidden layer of the mechanical property modality, is the mechanical property bias vector. The calculation formula from the input layer to the hidden layer of the material property modality is: where is the weight matrix of the i-th hidden layer of the material property modality, is the material property bias vector; The outputs of the hidden layers of each modality are fused to obtain the fused feature vector where I, J, and P are the serial numbers of the last hidden layer of each modality respectively; Bayesian regularization is used to optimize the fused network, and the output of the network is Y pred , the true value is Y true , then the loss function L u is defined as: where N is the number of samples, W is the set of all weights of the network, α and β are the Bayesian regularization coefficients respectively. By iteratively updating the weight W to minimize the loss function L, a defect prediction model for predicting the failure evaluation value of the hollow bolt in the defect state is constructed.

[0061] Further, the preset threshold in S2 obtains the basic threshold range by acquiring the material properties of the hollow bolt. The material properties include the mechanical properties such as the yield strength and fatigue limit of the material. The actual service environment data of the hollow bolt is obtained to get the environmental adaptability threshold. The actual service environment data includes high temperature, high humidity, and strong corrosion environment data. The environmental adaptability threshold is associated with the basic threshold range, and the basic threshold range is dynamically corrected to form the preset threshold.

[0062] Further, in S2, when determining the failure state of the hollow bolt by comparing with the preset threshold, a fuzzy decision tree is used for screening and classification. Specifically:

[0063] The fuzzy decision tree algorithm is used to assign fuzzy membership degrees to the failure states of different hollow bolts. The nodes of the decision tree are used to divide according to the fuzzy membership degrees of the failure states. Each branch represents a trend of a failure state. For the hollow bolts in different threshold ranges, according to the output results of the decision tree, they are divided into four categories: slight damage, moderate damage, severe damage, and failure.

[0064] Further, in S4, through the initial data set, a three-structure - fatigue life prediction model is constructed for the hollow bolts after screening and classification. Specifically:

[0065] The feature vectors of the nut, inner screw rod, and outer screw rod structures in the initial data set are obtained as where n s is the number of features. The feature vectors of each structure are weighted. The feature weight vectors of the nut, inner screw rod, and outer screw rod structures are respectively The weighted feature vectors are and where ⊙ represents element-wise multiplication; the weighted feature vectors are input into their respective sub-networks for feature extraction. The output of the hidden layer of the sub-network is: where W ht is the hidden layer weight matrix, b ht is the bias vector of the nut structure, f is the activation function, represents and The outputs of the three sub-networks are fused to obtain the fused feature vector, and the fatigue life prediction value L pred is output through the fully connected layer. The formula is L pred = f out (W out H + b out ), where W out is the fully connected layer weight matrix, b out is the bias vector, f outis the activation function of the output layer; by minimizing the predicted value L pred and the true fatigue life value L true the loss function Loss between them is used to adjust the model parameters, and a three-structure fatigue life prediction model is constructed.

[0066] Furthermore, in S5, the three-structure fatigue life prediction model is trained by combining transfer learning and reinforcement learning, specifically as follows:

[0067] From a large number of hollow bolt sample data with different working conditions and different defect states, a basic data set corresponding to common working conditions and typical defect states is divided, and this basic data set is used to pre-train an initial three-structure fatigue life prediction model; when encountering data with new, special working conditions or complex defect states, there is no need to train the model from scratch. Part of the parameters of the pre-trained model are transferred over as the initial parameters of the new model, and the three-structure fatigue life prediction model is learned in the data environment with different working conditions and defect states to accelerate the convergence speed of the three-structure fatigue life prediction model on new data.

[0068] Furthermore, in S6, the data of the defect points of the hollow bolt to be predicted are preprocessed and feature enhanced, specifically as follows:

[0069] The preprocessing maps the defect point data with different value ranges to a specific interval. The defect point data includes

[0070] position coordinates, shape dimensions and stress distribution, eliminates the influence of data scale differences on the model, and removes random noise and outliers in the data; for feature enhancement, the original data is non-linearly transformed through data transformation, and the feature combination method is adopted to combine different types of original features to generate new composite features.

[0071] Furthermore, in S6, the fatigue life prediction results of the defect points are obtained by using the model, and the life interval and risk level are output. Through a large number of random samplings to simulate, the sampling data is input into the model to obtain a fatigue life prediction value. After multiple samplings, a set of prediction values is obtained. By combining prior knowledge and the prediction values obtained from sampling, the probability distribution of the fatigue life is updated to determine the life interval;

[0072] For the division of the risk level, a risk assessment system is constructed through the life interval and the characteristics of the defect points. The life interval and defect characteristics are used as input variables, and different weights are assigned according to the influence degree of the risk. According to the value range and combination of the input variables, the risk is divided into different levels. The risk levels can be set as low risk, medium-low risk, medium risk, medium-high risk and high risk in ascending order.

[0073] This embodiment details the fatigue life prediction technology for hollow bolt defect points provided by this application. By obtaining multi-dimensional detection data and processing it through a three-dimensional defect model and a defect prediction model, the failure state of the bolt can be accurately judged and classified. Then, a three-structure fatigue life prediction model is constructed to output the life interval and risk level. This technology effectively solves the problem of low prediction accuracy of traditional methods, provides an accurate basis for equipment maintenance, greatly improves the safety and reliability of hollow bolts, and reduces potential safety hazards and economic losses caused by bolt failures.

[0074] Based on Embodiment 1, this embodiment describes the specific process of the fatigue life prediction method for hollow bolt defect points, as Figure 2 shown, specifically:

[0075] Collect and preliminarily process the structural detection data of the hollow bolt to obtain multi-dimensional information such as the geometric size and shape deviation data of the hollow part inside the bolt, the microscopic profile of the thread profile and the pitch cumulative error data, the residual stress distribution data at the bolt shank and nut, and the density gradient distribution data of the material inside the hollow bolt;

[0076] After the data is collected, it is input into a pre-constructed three-dimensional defect model based on an improved convolutional neural network architecture. At the input layer of the model, the data is transformed into a multi-dimensional feature matrix. Subsequently, in the convolutional layer, convolution operations are performed using a custom convolution kernel, and feature extraction is performed on the data according to the coordinate offset and size range of the convolution kernel. The pooling layer adopts an adaptive pooling strategy, dynamically adjusting the size of the pooling area according to the importance of the features, effectively reducing the data dimension while retaining key information. After passing through the fully connected layer, the feature vector is mapped to the defect feature space, and then the model parameters are optimized through backpropagation of the loss function, and finally an accurate defect detection value is output, realizing high-precision detection of internal defects of hollow bolts;

[0077] Input the obtained defect detection value into a pre-constructed defect prediction model. Through multi-modal fusion and Bayesian regularized neural network construction, first divide the defect detection value into geometric feature mode, mechanical property mode, and material property mode, and construct sub-neural networks respectively for feature extraction. Each modal sub-network transforms the input data into the output of the hidden layer through specific calculation formulas, and then fuses the output of each modal hidden layer into a feature vector. Use Bayesian regularization to optimize the fused network, and minimize the loss function by iteratively updating the weights, so as to output the failure evaluation value of the hollow bolt in the defect state.

[0078] Compare the damage evaluation value with a preset threshold, which is dynamically corrected by combining the material properties of the hollow bolt and the actual service environment data. Using the fuzzy decision tree algorithm, divide according to the fuzzy membership degrees of different damage states of the hollow bolt, and classify the hollow bolt into four categories: slight damage, moderate damage, severe damage, and failure. For the screened and classified hollow bolts, obtain their image data, extract the position, shape, and size information of the defect points to construct an initial data set for constructing a three-structure - fatigue life prediction model. This model performs weighted processing on the feature vectors of the nut, inner screw, and outer screw structures respectively, fuses them after extracting features through their respective sub-networks, outputs the fatigue life prediction value through a fully connected layer, and adjusts the model parameters by minimizing the loss function. Finally, preprocess and enhance the features of the data of the defect points of the hollow bolt to be predicted, input them into the trained model, obtain the fatigue life prediction results of the defect points, output the life interval and risk level, and complete the entire detection process.

[0079] This embodiment details a set of technical solutions that, by obtaining multi-type structure detection data of hollow bolts, processing them through a three-dimensional defect model and a defect prediction model, screening and classifying them in combination with a preset threshold, constructing and training and optimizing a three-structure - fatigue life prediction model, and finally inputting the data to be predicted to output the life interval and risk level. This set of technical solutions solves the problem of difficult to accurately predict the fatigue life of the defect points of hollow bolts, realizes high-precision defect detection and fatigue life prediction, provides a strong basis for equipment maintenance decision-making, and effectively improves the operation safety and reliability of equipment.

[0080] Based on Embodiment 1, this embodiment describes the three-structure - fatigue life prediction model constructed for hollow bolts, specifically as follows:

[0081] As Figure 3 shown, separate the hollow bolt into a nut, an inner screw, and an outer screw according to its structure, because these three parts have their own characteristics in the mechanical properties and fatigue life influence mechanism of the hollow bolt; the nut mainly bears axial force and torque, and its structural characteristics, such as the integrity of the thread and the thickness of the nut, have a significant impact on the connection stability and fatigue life; the inner screw is in direct contact with the medium or environment and is easily affected by corrosion, wear, etc., and its material properties, surface quality, and defect conditions will significantly change the fatigue performance; the outer screw is the key part that bears external loads, and its geometric dimensions, residual stress distribution, and other factors determine the load-bearing capacity under complex working conditions; separately modeling can conduct in-depth analysis according to the characteristics of each part, extract exclusive feature vectors, assign different weights, accurately reveal the contribution of each part to the fatigue life, improve the accuracy and reliability of fatigue life prediction, and provide a more scientific basis for the maintenance and replacement of hollow bolts;

[0082] As Figure 4As shown, for the three-structure fatigue life prediction model constructed for the hollow bolt, the feature vectors of the nut, the inner screw rod, and the outer screw rod structures in the initial dataset are respectively where n s is the number of features. The feature vectors of each structure are weighted. The feature weight vectors of the nut, the inner screw rod, and the outer screw rod structures are respectively The weighted feature vectors are and where ⊙ represents element-wise multiplication;

[0083] The weighted feature vector of the nut is input into the sub-network of the nut for feature extraction. The output of the hidden layer of the nut sub-network is: where is the weight matrix of the hidden layer of the nut sub-network, is the structure bias vector of the nut sub-network, and f is the activation function;

[0084] The weighted feature vector of the inner screw rod is input into the sub-network of the inner screw rod for feature extraction. The output of the hidden layer of the inner screw rod sub-network is: where is the weight matrix of the hidden layer of the inner screw rod sub-network, is the structure bias vector of the inner screw rod sub-network, and f is the activation function;

[0085] The weighted feature vector of the outer screw rod is input into the sub-network of the outer screw rod for feature extraction. The output of the hidden layer of the outer screw rod sub-network is: where is the weight matrix of the hidden layer of the outer screw rod sub-network, is the structure bias vector of the outer screw rod sub-network, and f is the activation function;

[0086] The outputs of the three sub-networks are fused to obtain a fused feature vector, and the fatigue life prediction value L pred is output through the fully connected layer. The formula is L pred = f out (W out H + b out ), where W out is the weight matrix of the fully connected layer, b out is the bias vector, and f out is the activation function of the output layer; By minimizing the loss function Loss between the predicted value L pred and the true fatigue life value L true , the model parameters are adjusted to construct a three-structure fatigue life prediction model.

[0087] This embodiment details the extraction of the eigenvectors of the nut, inner screw rod, and outer screw rod structures. After weighted processing, they are input into the sub-network to extract features and fuse them, and the fatigue life prediction value is output through the fully connected layer. This solves the problem that the traditional prediction model cannot accurately consider the influence of different structures of the hollow bolt on the fatigue life, can analyze the characteristics of each structure more carefully, improves the accuracy and reliability of the prediction, provides strong data support for the maintenance decision of the hollow bolt, and helps to prevent safety accidents caused by bolt fatigue failure in advance.

[0088] The above are only the preferred embodiments of the present invention, and it does not limit the protection scope of the present invention. For those skilled in the art, the present invention can have various changes and modifications; within the spirit and principle of the present invention, any changes, modifications, substitutions, integrations, and parameter changes to these embodiments through conventional substitutions or the ability to achieve the same functions without departing from the principle and spirit of the present invention fall within the protection scope of the present invention.

Claims

1. A method for predicting the fatigue life of defect points of a hollow bolt, characterized in that, Specifically as follows: S1. Obtain the detection data of the hollow bolt structure, input the structure detection data into the pre-constructed three-dimensional defect model, and output the defect detection value; S2. Input the obtained defect detection value into the pre-constructed defect prediction model, output the failure evaluation value of the hollow bolt in the defect state, compare with the preset threshold, determine the failure state of the hollow bolt, and conduct screening and classification; S3. Obtain the image data of the hollow bolts after screening and classification, extract the position, shape and size information data of the defect points, and construct the initial data set; S4. Through the initial data set, construct a three-structure - fatigue life prediction model for the hollow bolts after screening and classification, and the three structures are respectively the nut, the inner screw rod and the outer screw rod structures; S5. Obtain a large number of sample data of hollow bolts under different working conditions and different defect states, and use transfer learning to train the three-structure - fatigue life prediction model to optimize the model parameters; S6. Input the data of the defect points of the hollow bolt to be predicted, after preprocessing and feature enhancement, into the corresponding trained three-structure - fatigue life prediction model, use the model to obtain the fatigue life prediction result of the defect points, and output the life interval and risk level.

2. The fatigue life prediction method for defect points of a hollow bolt according to claim 1, characterized in that The detection data of the hollow bolt structure in S1 includes the geometric dimension and shape deviation data of the hollow part inside the bolt, the microscopic profile of the thread profile and the pitch cumulative error data, the residual stress distribution data at the bolt rod body and nut, and the density gradient distribution data of the material inside the hollow bolt.

3. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1, characterized in that, The three-dimensional defect model pre-constructed in S1 is constructed based on an improved convolutional neural network architecture. The input layer receives the hollow bolt structure detection data and converts it into a multi-dimensional feature matrix. where i d represents the data dimension, and j d represents the sample serial number; in the convolutional layer, convolution operations are performed through a custom convolutional kernel where m d , n d are the coordinate offsets of the elements of the convolutional kernel in the row direction of the input feature matrix X, and the convolution operation is where a d , b d are the size ranges of the convolutional kernel, p d , q d are the coordinates of the output feature map, is the eigenvalue at the position with coordinates (p d , q d )) in the output feature map obtained after the convolution operation. The pooling layer adopts an adaptive pooling strategy, dynamically adjusts the size of the pooling area according to the importance of the features, maps the feature vectors to the defect feature space through the fully connected layer, and optimizes the model parameters through the loss function where is the true defect value, is the predicted defect value, s d is the number of samples, r d is the serial number of the sample, and a three-dimensional defect model is constructed.

4. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1, characterized in that, The defect prediction model in S2 is pre-constructed by multi-modal fusion and Bayesian regularized neural network. Specifically: Perform multimodal partitioning on the defect detection values. The original defect detection value vector is Partition it into geometric feature modalities Mechanical property modalities and material property modalities where m u + k u + l u = n u ; For each modal construction sub-neural network, the calculation formula from the input layer to the hidden layer of the geometric feature modality is as follows: Where is the weight matrix of the i-th hidden layer of the geometric feature modality, is the geometric feature bias vector, σ is the activation function, and the calculation formula from the input layer to the hidden layer of the mechanical property modality is as follows: Where is the weight matrix of the i-th hidden layer of the mechanical property modality, is the mechanical property bias vector, and the calculation formula from the input layer to the hidden layer of the material property modality is as follows: Where is the weight matrix of the i-th hidden layer of the material property modality, is the material property bias vector; the outputs of the hidden layers of each modality are fused to obtain the fused feature vector where I, J, and P are the numbers of the last hidden layers of each modality respectively; the fused network is optimized using Bayesian regularization, and the output of the network is Y pred , and the true value is Y true , then the loss function L u is defined as: Where N is the number of samples, W is the set of all weights of the network, α and β are the Bayesian regularization coefficients respectively, and the weights W are iteratively updated to minimize the loss function L, thus constructing a defect prediction model for predicting the failure evaluation value of the hollow bolt in the defective state.

5. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1, characterized in that, The preset threshold in S2 obtains the basic threshold interval by obtaining the material properties of the hollow bolt. The material properties include the mechanical properties such as the yield strength and fatigue limit of the material; obtain the actual service environment data of the hollow bolt to obtain the environmental adaptability threshold. The actual service environment data includes high temperature, high humidity, strong corrosion environment data, and correlate the environmental adaptability threshold with the basic threshold interval to dynamically correct the basic threshold interval to form the preset threshold.

6. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1 or 5, characterized in that, In S2, when comparing the preset threshold to determine the failure state of the hollow bolt, a fuzzy decision tree is used for screening and classification. Specifically: Use the fuzzy decision tree algorithm to assign fuzzy membership degrees to the failure states of different hollow bolts, and divide according to the fuzzy membership degrees of the failure states through the nodes of the decision tree. Each branch represents a trend of a failure state. For hollow bolts in different threshold intervals, according to the output result of the decision tree, divide them into four categories: slight damage, moderate damage, severe damage and failure.

7. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1, characterized in that, In S4, through the initial data set, construct a three-structure - fatigue life prediction model for the hollow bolts after screening and classification. Specifically: The feature vectors of the nut, inner screw and outer screw structures in the initial dataset are respectively where n s is the number of features. The feature vectors of each structure are weighted. The feature weight vectors of the nut, inner screw and outer screw structures are respectively The weighted feature vectors are and where ⊙ represents element-wise multiplication; The weighted feature vectors are input into their respective sub-networks for feature extraction. The output of the hidden layer of the sub-network is: where W ht is the hidden layer weight matrix, b ht is the bias vector of the nut structure, f is the activation function, represents and The outputs of the three sub-networks are fused to obtain a fused feature vector, and the fatigue life prediction value L pred is output through the fully connected layer. The formula is L pred = f out (W out H + b out ), where W out is the fully connected layer weight matrix, b out is the bias vector, f out is the activation function of the output layer; By minimizing the loss function Loss between the predicted value L pred and the true fatigue life value L true , the model parameters are adjusted to construct a three-structure - fatigue life prediction model.

8. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 7, characterized in that, In S5, the three-structure - fatigue life prediction model is trained by combining transfer learning and reinforcement learning. Specifically: From the sample data of hollow bolts under a large number of different working conditions and different defect states, a basic data set corresponding to common working conditions and typical defect states is divided, and this basic data set is used to pre-train an initial three-structure fatigue life prediction model; when encountering data of new, special working conditions or complex defect states, there is no need to train the model from scratch. Some parameters of the pre-trained model are migrated as the initial parameters of the new model, and the three-structure fatigue life prediction model is learned in the data environment of different working conditions and defect states to accelerate the convergence speed of the three-structure fatigue life prediction model on new data.

9. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1, characterized in that, In S6, the data of the defect points of the hollow bolts to be predicted are preprocessed and feature enhanced, specifically as follows: The preprocessing maps the defect point data with different value ranges to a specific interval. The defect point data includes position coordinates, shape dimensions, and stress distribution, eliminates the influence of data scale differences on the model, and removes random noise and outliers in the data; For feature enhancement, the original data is non-linearly transformed through data transformation, and the feature combination method is adopted to combine different types of original features to generate new composite features.

10. A method for predicting the fatigue life of defect points of a hollow bolt according to claim 1, characterized in that In S6, the fatigue life prediction results of the defect points are obtained by using the model, and the life interval and risk level are output. Through a large number of random samplings for simulation, the sampling data is input into the model to obtain a fatigue life prediction value. After multiple samplings, a set of prediction values is obtained. By combining prior knowledge and the prediction values obtained from sampling, the probability distribution of the fatigue life is updated to determine the life interval; For the division of risk levels, a risk assessment system is constructed through the life interval and the characteristics of the defect points. The life interval and defect characteristics are used as input variables, and different weights are assigned according to the degree of influence of the risk. According to the value ranges and combination situations of the input variables, the risks are divided into different levels. The risk levels can be set from low to high as low risk, medium-low risk, medium risk, medium-high risk, and high risk in turn.

Images