Double-layer adaptive collaborative optimization crank connecting rod component fault diagnosis method

Abstract

The invention discloses a double-layer self-adaptive collaborative optimization crank connecting rod component fault diagnosis method, and relates to the technical field of fault diagnos.The method adopts a double-layer collaborative structure: in a signal decomposition layer, a sparrow search algorithm is utilized, an average envelope entropy and modal stability joint objective function is introduced, the modal number and penalty factor of VMD are determined in a self-adaptive mode, and a fault diagnosis result is obtained; the sparsity and stability balance of decomposition is realized; and in the classification layer, a Bayesian optimization algorithm is adopted to optimize kernel function parameters and penalty factors of the SVM, and the target weight of the decomposition layer is dynamically corrected according to classification error feedback, so that the method can recognize target faults in a high-precision and stable manner under multiple working conditions, the reliable operation of the crank connecting rod component and the whole machine is effectively guaranteed, and the engineering application feasibility and the popularization value are relatively high.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and more specifically to a two-layer adaptive collaborative optimization method for fault diagnosis of crank-connecting rod components. Background Technology

[0002] As one of the most widely used core components in industrial equipment, the crankshaft and connecting rod assembly's operating status directly affects equipment stability and production safety. During long-term service, the load-bearing characteristics of the crankshaft and connecting rod assembly are prone to change, leading to gradual performance degradation. In severe cases, this can cause equipment downtime or damage, directly impacting industrial production efficiency and economic benefits. Therefore, conducting accurate and real-time fault diagnosis of the crankshaft and connecting rod assembly's operating status has significant engineering and economic value.

[0003] Existing fault diagnosis techniques for crankshaft and connecting rod components primarily rely on vibration signal analysis and manual feature extraction methods. These methods identify fault types by analyzing sparse pulse or envelope features in the signal. Common signal processing methods include Empirical Mode Decomposition (EMD), Wavelet Packet Decomposition (WPD), and Spectral Kujic Analysis. While these methods can achieve good results under certain operating conditions, they depend heavily on domain knowledge and manual experience, making it difficult to maintain stable diagnostic performance in complex operating conditions and nonlinear environments. Furthermore, these methods exhibit low feature extraction efficiency and insufficient generalization ability when dealing with large-scale data, limiting their further application in fault diagnosis.

[0004] With the rapid development of artificial intelligence and intelligent manufacturing technologies, data-driven intelligent diagnostic methods have gradually become a research hotspot. Deep learning-based diagnostic models can automatically extract complex signal features without human intervention, achieving efficient fault identification. In existing research, structures such as Convolutional Neural Networks (CNN), Graph Neural Networks (GNN), and Long Short-Term Memory (LSTM) networks have been used for fault diagnosis tasks, achieving high recognition accuracy in various experimental scenarios. To further enhance the model's ability to model the global dependencies of time-series signals, some studies have introduced Transformer structures to capture the relevant features of signals across different time scales. Furthermore, by fusing CNN and Transformer structures, the collaborative extraction of local and global features has been achieved, thereby improving the model's diagnostic accuracy and adaptability to different operating conditions. However, existing technologies still have the following shortcomings: (1) Vibration signals in actual industrial sites are often subject to strong noise interference, and weak early fault features are easily masked, making it difficult for the model to effectively identify key features. (2) The vibration signal generated by the crank connecting rod component during service is highly nonlinear. Existing deep learning models are not capable of modeling nonlinear dynamic characteristics and are easily affected by environmental interference. Moreover, the complex nonlinear dynamic characteristics pose challenges to the convergence stability and training efficiency of the model, further affecting the diagnostic accuracy. (3) Deep learning models have complex structures, a large number of parameters, and high requirements for computing resources, making them difficult to deploy and promote in real time in industrial settings.

[0005] In summary, existing fault diagnosis methods for crankshaft and connecting rod components have shortcomings in noise resistance, feature extraction capability, and model practicality, making it difficult to achieve a balance between diagnostic accuracy and performance. Therefore, there is an urgent need for a method that can adaptively optimize signal decomposition parameters, effectively extract key features, and achieve high-precision fault identification to improve the reliability of crankshaft and connecting rod component operating status monitoring and health assessment. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a two-layer adaptive collaborative optimization method for fault diagnosis of crank-connecting rod components that overcomes or at least partially solves the above problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide a two-layer adaptive collaborative optimization method for fault diagnosis of crank-connecting rod components, comprising: S1. Acquire vibration signals of the crankshaft and connecting rod components; S2. Based on the sparrow search optimization algorithm, a VMD optimization model with average envelope entropy and modal stability as joint objective functions is established to search for the optimal VMD parameter combination of vibration signal under each state; S3. Set the VMD parameters to the optimal parameter combination of the vibration signal in each state, decompose the vibration signal in each state, and obtain several IMFs; S4. Calculate the fuzzy entropy and several time-domain features for each IMF to form a multidimensional feature vector; S5. The Bayesian optimization algorithm is used to jointly optimize the penalty factor and kernel function parameters of SVM, search for the best parameter combination of SVM, input the multidimensional feature vector into the SVM classifier for training, obtain the SVM prediction model for vibration signal under different states, and dynamically adjust the weight coefficients in the signal decomposition layer based on the prediction error feedback of the classification layer to obtain the optimized SVM prediction model. S6. Input the multidimensional feature vector corresponding to the test signal into the optimized SVM prediction model to obtain the fault classification result.

[0008] Preferably, S2 includes: S201 sets the VMD parameter optimization range and initializes the sparrow search optimization algorithm parameters: sets the number of decomposition modes. Punishment factor Given the number of sparrows, n, determine the ratio of explorers, followers, and watchdogs among the sparrows, with each individual corresponding to a set of VMD parameter combinations. i=1,2,...n, set the maximum number of iterations or convergence condition; S202 calculates the initial population fitness: For the i-th sparrow individual, the initial population fitness is calculated using the VMD parameters. Vibration signals Perform VMD decomposition to obtain For each IMF component, calculate the envelope entropy of each IMF component and the average envelope entropy. ; Calculate the mean center frequency of each IMF component and center frequency variance And calculate modal stability Thus, the calculation of VMD parameter combinations Lower joint objective function ; S203 SSA Iterative Optimization: Based on the joint objective function, sort the n individuals of the initial population in ascending order of fitness value, and retain the initial optimal individual's VMD parameter combination. As a reference point for subsequent iterations, the population fitness is recalculated by updating the positions of explorers, followers, and watchdogs, while retaining the current minimum fitness value. Corresponding parameter combinations ; S204: If the number of iterations reaches the maximum number of iterations or the convergence condition is met, terminate the iteration; otherwise, repeat S203. After the iteration stops, select the VMD parameter combination that minimizes the joint objective function value. This is the optimal combination of VMD parameters.

[0009] Preferably, the explorer position update formula is:

[0010] in, t This represents the current iteration number. j =1,2, , d, d Representing the individual dimension, Indicates the first In the next iteration, the first in the explorer group i The sparrow in the first j The numerical value of the dimension. Indicates the first t In the next iteration, the first in the explorer groupi The sparrow in the first j The numerical value of the dimension. iter max The maximum number of iterations, Q It is a random number. L It is a length of d A row vector with elements of 1. , These represent the current alert threshold and danger threshold for the entire group, respectively. The follower position update formula is:

[0011] in, Indicates the first The optimal position for the explorer in the next iteration. Indicates the first In the second iteration, the first in the follower group i The sparrow in the first j The numerical value of the dimension. Indicates the first t The worst position globally in the next iteration. Indicates the first t In the second iteration, the first in the follower group i The sparrow in the first j The numerical value of the dimension. represent , represent A matrix, where the elements are randomly assigned values ​​of 1 or -1; The formula for updating the vigilant's position is:

[0012] in, Indicates the first t In the next iteration, the first group of vigilants... i The sparrow in the first j The numerical value of the dimension. Indicates the first In the next iteration, the first group of vigilants... i The sparrow in the first j The numerical value of the dimension. Indicates the first t The optimal global position in the next iteration is determined by ε, which is a random number to avoid errors where the denominator is zero during the iteration process. f i Indicates the first i The fitness value of a sparrow. f g and f w These represent the current global best and worst fitness, respectively. It is a random number within the range [-1, 1], used to control the direction and step size of the optimal individual's local exploration. It is a random number that follows a normal distribution and is used to control the randomness of the step size when an individual moves toward the global optimal position.

[0013] Preferably, the VMD optimization model with average envelope entropy and modal stability as joint objective functions is as follows:

[0014] in, The average envelope entropy, Let be the modal stability function. , The adjustable weights of the VMD objective function decomposition layer. + =1, It is the joint objective function of the VMD optimization model, and the objective is to minimize To obtain the optimal decomposition parameters;

[0015] in, K The number of IMFs. Envelope entropy;

[0016] in, For the next disturbance The center frequency of the mode, Its variance reflects stability. It is the first disturbance The average value of the center frequencies of each modality.

[0017] Preferably, in step S4, the fuzzy entropy calculation process is as follows: (1) Given an N-dimensional IMF sequence Construct a phase space vector of length m: ; (2) Definition , Distance:

[0018] (3) Calculate fuzzy similarity:

[0019] Where m represents the embedding dimension, and r represents the similarity tolerance, controlling the width of the fuzzy function. x(N) This represents the data value at the Nth time point in the IMF sequence. That is, from the first in the IMF sequence p Starting from one data point, continuously take... m A set of data points constitutes a m A dimensional vector, where s represents the position index within the phase space vector. This represents the maximum absolute value of the difference between corresponding elements of two m-dimensional vectors; This represents the fuzzy similarity between two m-dimensional vectors. (4) For each vector Calculate the average fuzzy similarity with other vectors:

[0020] The average fuzzy similarity is taken as the mean:

[0021] (5) Repeat the above steps to obtain Calculate the fuzzy entropy:

[0022] in, Indicates the first p The average fuzzy similarity of the vectors This represents the mean fuzzy similarity when the embedding dimension is m. This represents the mean fuzzy similarity when the embedding dimension is m+1.

[0023] Preferably, a Bayesian optimization algorithm is used to jointly optimize the penalty factor and kernel function parameters of the SVM, and the joint objective function is:

[0024] Where c is the penalty factor and g is the kernel function parameter. To verify the error, This is a complexity term, reflecting the model's complexity and computational cost. , For balance coefficient, + =1;

[0025] in, It is the number of samples in the validation set. Representative sample true label, The model predicts the label. g (x) is an indicator function that takes the value 0 or 1.

[0026] Preferably, in step S5, dynamically adjusting the weight coefficients in the signal decomposition layer based on the prediction error feedback of the classification layer includes:

[0027] in, t This represents the current iteration number. For learning rate, The average envelope entropy and modal stability in the VMD objective function decomposition layer are respectively the values ​​at the 1st, 2nd, and 3rd respectively. Adjustable weights for each iteration Let be the adjustable weight for the average envelope entropy in the (t+1)th iteration. For the adjustable weights of modal stability in the (t+1)th iteration, This represents the change in classification error.

[0028] In a second aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements a two-layer adaptive collaborative optimization method for diagnosing crank-connecting rod component faults.

[0029] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements a two-layer adaptive collaborative optimization method for diagnosing crank-connecting rod component faults.

[0030] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a two-layer adaptive collaborative optimization method for fault diagnosis of crank-connecting rod components, which has the following effects: (1) This invention introduces the Sparrow Search Algorithm (SSA) in the signal decomposition layer to perform global adaptive optimization of the key parameters of variational mode decomposition (VMD), which overcomes the problems of parameter dependence on manual setting, unstable decomposition results and poor repeatability in traditional VMD methods, and realizes intelligent signal decomposition process and global optimal configuration.

[0031] (2) In the VMD parameter optimization stage, the present invention establishes a fitness function with average envelope entropy and modal stability coefficient as joint optimization objectives, so that SSA takes into account both the sparsity and stability of the decomposition results during the optimization process, thereby obtaining more physically consistent intrinsic mode components (IMF), which significantly improves the robustness of VMD decomposition and the fidelity of signal features.

[0032] (3) In the feature construction stage, the present invention calculates the fuzzy entropy features of each IMF component obtained by optimizing VMD decomposition to characterize the internal dynamic complexity of the signal. At the same time, it extracts a variety of time-domain statistical feature parameters from the original signal to construct a high-dimensional multi-scale feature vector, thereby enhancing the separability between different fault modes.

[0033] (4) In view of the problem that the parameters of the traditional SVM (Support Vector Machine) model depend on empirical settings and the classification performance is unstable, the present invention introduces the Bayesian optimization algorithm (BO) to globally optimize the kernel function parameters and penalty factors of the support vector machine, and establishes an optimization model with classification error and model complexity as dual objectives, so as to realize the adaptive adjustment of classifier parameters and complexity constraints, effectively balance the accuracy and generalization ability of the model, and make the classification performance more stable and reliable.

[0034] (5) This invention further constructs a dynamic feedback mechanism between the signal decomposition layer and the classification layer. By using the prediction error output by the classification layer as a feedback signal, the weight coefficients in the objective function of the decomposition layer are corrected in real time, thus realizing an adaptive collaborative optimization closed loop between the decomposition layer and the classification layer. This closed loop mechanism enables the signal decomposition, feature extraction and classification recognition processes to drive each other and converge together, significantly improving the fault identification accuracy and robustness of the model under complex working conditions and noisy environments.

[0035] (6) This invention forms a two-layer collaborative optimization structure of “SSA-VMD+BO-SVM”, and introduces adaptive optimization mechanisms in the signal decomposition stage and the fault classification stage respectively, realizing intelligentization of the entire process from feature extraction to classification and recognition. This invention can achieve adaptive collaborative optimization under different working conditions without manual intervention, providing an efficient and reliable technical solution for intelligent fault diagnosis of crank connecting rod components in complex service environments. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0037] Figure 1 This is a flowchart of a two-layer adaptive collaborative optimization crank-connecting rod component fault diagnosis method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the joint optimization of VMD parameters using the sparrow search algorithm in this embodiment of the invention. Figure 3 This is a flowchart illustrating the joint optimization process of the Bayesian optimization algorithm for SVM parameters in an embodiment of the present invention. Figure 4 This is the original waveform image of the analog signal provided in the embodiments of the present invention; Figure 5 This is a diagram of the analog signal decomposition based on adaptive optimization VMD provided in this embodiment of the invention. Figure 6The spectrum diagram of IMFs obtained after decomposing the severely worn signal provided in this embodiment of the invention; Figure 7 The fault diagnosis results of VMD-SVM based on two-layer adaptive collaborative optimization provided in the embodiments of the present invention are shown. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] This invention discloses a two-layer adaptive collaborative optimization method for fault diagnosis of crank-connecting rod components, such as... Figure 1 As shown, it includes: S1. Obtain the vibration signal x(t) of the crank-connecting rod assembly; S2. Based on the sparrow search optimization algorithm, a VMD optimization model is established with average envelope entropy and modal stability as joint objective functions. This algorithm is then used to search for the optimal VMD parameter combination of the vibration signal under various states. Where K is the number of IMF modes, It is a secondary penalty factor; S3. Set the VMD parameters to the optimal VMD parameter combination obtained in S2 for each state of vibration signal, and decompose the corresponding vibration signal x(t) for each state. S4. After each vibration signal x(t) is decomposed by VMD, several IMFs are obtained. The fuzzy entropy and several time-domain features of each IMF are calculated to form a multidimensional feature vector s. S5. The Bayesian optimization algorithm is used to jointly optimize the penalty factor c and kernel function parameter g of the SVM, searching for the optimal parameter combination of the SVM. Set the SVM parameters to The multidimensional feature vector s is input into the SVM classifier for training, thereby obtaining the SVM prediction model for vibration signals under different states. Based on the prediction error feedback of the classification layer, the weight coefficients in the signal decomposition layer are dynamically adjusted to form an adaptive collaborative closed loop between the decomposition layer and the classification layer, thus obtaining the optimized SVM prediction model. S6. Input the test signal, obtain a multi-dimensional feature vector of the test signal through S2-S4, input the multi-dimensional feature vector into the optimized SVM prediction model, and obtain the fault classification result of the test signal.

[0040] In this embodiment, K and The choice of parameters affects the decomposition effect of the VMD method, and the selection of parameters is irregular. This invention employs a sparrow search-based optimization algorithm for parameter optimization, defined as follows: In the SSA algorithm, sparrows are simulated to search for food, and their positions can be represented by the following matrix:

[0041] in, n For the number of individuals, d This represents the individual dimension. The fitness of a sparrow individual can be represented by the following matrix:

[0042] For explorers in a population, they have higher fitness values ​​and are able to prioritize finding food during exploration. Explorers provide followers with directions and foraging areas, and when predators are detected, explorers must lead their followers to a safe zone. Compared to followers, they have a larger range of activity. Therefore, the location update of explorers follows the following expression:

[0043] in, t This represents the current iteration number. j =1,2, , d, d Representing the individual dimension, Indicates the first t In the next iteration, the first in the explorer group i The sparrow in the first j The numerical value of the dimension. Indicates the first In the next iteration, the first in the explorer group i The sparrow in the first j The numerical value of the dimension. iter max The maximum number of iterations, Q It is a random number. L It is a length of d A row vector with elements of 1. , These are the current alert threshold and danger threshold for the entire group, respectively.

[0044] For most followers, they will follow the explorer in search of food. To better simulate the predation behavior of sparrows in nature, some sparrows following the explorer will compete with the explorer for food. When they succeed in the competition, they will obtain the explorer's food; otherwise, they will continue to compete for food from other explorers. The small number of sparrows that do not follow the explorer will forage in other areas to increase their probability of obtaining food. Based on the above criteria, the position iteration of the followers follows the following formula:

[0045] in, Indicates the first The optimal position for the explorer in the next iteration. Indicates the first In the second iteration, the first in the follower group i The sparrow in the first j The numerical value of the dimension. Indicates the first t The worst position globally in the next iteration. Indicates the first t In the second iteration, the first in the follower group i The sparrow in the first j The numerical value of the dimension. represent , represent A matrix, where the elements are randomly assigned values ​​of 1 or -1; i > n / 2 indicates that the predation behavior has entered the middle to late stages, and the worst-positioned individual in the follower group will appear. Therefore, these sparrows will change their foraging direction to increase their chances of obtaining food.

[0046] Within the entire sparrow population, the location of sparrows capable of detecting predators is random. When a threat is detected, sparrows in the flock move towards safer areas to gain a better position, while sparrows already in safe areas will move closer to nearby individuals. Based on these principles, the iterative location of the watchdog follows the formula below:

[0047] in, Indicates the first t In the next iteration, the first group of vigilants... i The sparrow in the first j The numerical value of the dimension. Indicates the first In the next iteration, the first group of vigilants... i The sparrow in the first j The numerical value of the dimension. Indicates the first tThe optimal global position in the next iteration is determined by ε, which is a random number to avoid errors where the denominator is zero during the iteration process. f i Indicates the first i The fitness value of a sparrow. f g and f w These represent the current global best and worst fitness, respectively. It is a random number within the range [-1, 1], used to control the direction and step size of the optimal individual's local exploration. It is a random number that follows a normal distribution and is used to control the randomness of the step size when an individual moves toward the global optimal position.

[0048] Further optimization is achieved in S2, where the sparrow search optimization algorithm is applied to the parameter optimization of VMD. The fitness function is selected as follows: When optimizing VMD parameters, the joint objective function composed of the average envelope entropy and modal stability of each IMF decomposed by VMD is selected as the fitness function. The specific calculation formula is as follows:

[0049] in, The average envelope entropy, Let be the modal stability function. , For adjustable weights, + =1, It is the joint objective function for the i-th individual, and the objective is to minimize In order to obtain the optimal decomposition parameters.

[0050] (1) Entropy is a measure of sparsity. The more impulse components a signal contains, the higher its sparsity and the lower its entropy value. Conversely, the fewer impulse components a signal contains, the higher its entropy value. Therefore, entropy can be used as a measure of signal sparsity. Based on information entropy, the formula for calculating envelope entropy is:

[0051] in, The number of sampling points for the signal. For the original signal at the 1st v The envelope value of each sampling point p ( v ) is the first v The standardized probability corresponding to each sampling point This is the envelope entropy value.

[0052] (2) In order to obtain better global optimization results, the objective function is selected as the average envelope entropy, and its calculation formula is as follows:

[0053] Where K is the number of IMFs.

[0054] (3) Modal stability is one of the key indicators for evaluating the robustness of VMD decomposition results under parameter perturbations. It is used to constrain the decomposition process to avoid mode aliasing and unstable reconstruction. Modal stability can be defined as:

[0055] in, Let k be the center frequency of the k-th mode under perturbation. Its variance reflects stability. It is the first disturbance The average value of the center frequencies of each modality.

[0056] The process of introducing SSA at the signal decomposition layer to perform global adaptive joint optimization of key VMD parameters is as follows: Figure 2 As shown: (1) Input the original vibration signal ; (2) Set the VMD parameter optimization range and initialize the SSA parameters: Set the number of decomposition modes. Punishment factor Given the number of sparrows, n, determine the ratio of explorers, followers, and watchdogs among the sparrows; each individual corresponds to a set of VMD parameters. Set the maximum number of iterations or convergence conditions; (3) Calculate the initial population fitness: Taking the i-th individual as an example, use the VMD parameters For the original vibration signal Perform VMD decomposition to obtain For each IMF component, calculate the envelope entropy of each IMF component and the average envelope entropy. ; Calculate the mean center frequency of each IMF component and center frequency variance And calculate modal stability Thus, the VMD parameters are calculated. Lower joint objective function ; (4) SSA Iterative Optimization: Sort the n individuals in the initial population according to their fitness values ​​from smallest to largest. The smaller the value, the better the individual; retain the initial optimal individual. This serves as a reference point for subsequent iterations; subsequent iterations recalculate the population fitness by updating the positions of explorers, followers, and watchdogs, while retaining the current minimum fitness value. Corresponding parameter combinations ; (5) If the number of iterations reaches the maximum number of iterations or the convergence condition is met, then terminate the iteration; otherwise, repeat step 4. (6) After stopping, select the option that minimizes the objective function value. This is the output as the optimal combination of VMD parameters.

[0057] In this embodiment, after VMD decomposes the signal into several IMFs, the fuzzy entropy of the IMFs and the time-domain features of x(t) will be used as the feature vectors for the next classification step. The fuzzy entropy of each IMF is specifically calculated as follows: for a given N-dimensional finite time series ( u 1, u 2,…., u N Calculating the fuzzy entropy requires determining two parameters: the embedding dimension *m* and the similarity tolerance *r*. The embedding dimension *m* is directly related to the reconstruction of the time series, and its selection is crucial for preserving fault information. When *m* is large, a large dataset *E* (*E* = 10m~30m) is required, and information loss may occur. When *m* is small, insufficient information may prevent a complete measurement of the time series. Based on these two considerations, the embedding dimension *m* in this invention is set to 2. The similarity tolerance *r* is set here as a proportion of the standard deviation (SD) of the time series, typically ranging from 0.1 × SD to 0.25 × SD. A small similarity tolerance *r* increases the sensitivity of the results to noise. Therefore, the value of *r* typically ranges from 0.1 SD to 0.25 SD (SD is the standard deviation of the dataset), hence *r* is set to 0.2 SD.

[0058] The specific calculation process of fuzzy entropy is as follows: (1) Given an N-dimensional IMF sequence Construct a phase space vector of length m:

[0059] (2) Definition , Distance:

[0060] (3) Calculate fuzzy similarity:

[0061] (4) For each vector Calculate the average fuzzy similarity with other vectors:

[0062] The average fuzzy similarity is taken as the mean:

[0063] (5) Repeat steps 1–4, changing the embedding dimension to m+1, to obtain

[0064] (6) Calculate the fuzzy entropy:

[0065] In this embodiment, the joint objective function of the Bayesian optimization algorithm is calculated as follows:

[0066] in, To verify the error, This is a complexity term, reflecting the model's complexity and computational cost. , For balance coefficient, + =1, A smaller value indicates higher classification accuracy, which is a direct measure of SVM performance. The calculation formula is:

[0067] in, It is the number of samples in the validation set. Representative sample true label, The model predicts the label, and g(x) is the indicator function (1 for correct and 0 for incorrect).

[0068] A Bayesian optimization algorithm is used to jointly optimize the penalty factor and kernel function parameters of the SVM, searching for the optimal parameter combination for the SVM. Figure 3 As shown, the specific process is as follows: (1) Define the parameter space and objective function: Determine the range of values ​​for c and g, and adopt the objective function. Determine the balance coefficient; (2) Initialize sample points: Randomly sample z initial points For each group Train the SVM using k-fold cross-validation to obtain the validation set error. ; (3) Establishing a surrogate model: Fitting the probability distribution of the initial sample points with a Gaussian process (GP) to obtain the objective function. The posterior probability model is used to replace the true objective function (to avoid repeated training of SVM). (4) Calculate the EI function and select the next parameter point that is most likely to improve the current optimal solution. : (5) Evaluate new sampling points and update the surrogate model: For new sampling points Train the SVM using k-fold cross-validation and calculate the corresponding objective function value. Refit the posterior distribution and update the GP model; (6) Determine the stopping condition: Repeat steps 4-5 until the iteration stopping condition is met; (7) After stopping, select the point from all sample points that minimizes the objective function value. This is the optimal combination of parameters for SVM.

[0069] In this embodiment, the weight coefficients in the signal decomposition layer are dynamically adjusted based on the prediction error feedback of the classification layer. The specific process is as follows: After each classification iteration, based on the classification error Adjust the weight coefficients of the decomposition layer according to the changes. , An adaptive mechanism is implemented to enhance stability of the decomposition layer when accuracy decreases:

[0070] in, t This represents the current iteration number. For learning rate, The average envelope entropy and modal stability in the VMD objective function decomposition layer are respectively the values ​​at the 1st, 2nd, and 3rd respectively. Adjustable weights for each iteration The weights for the average envelope entropy in the (t+1)th iteration are... The weights for modal stability in the (t+1)th iteration are... This represents the change in classification error.

[0071] Specifically, when the classification error increases ( >0), insufficient feature extraction or unstable decomposition, increasing the risk. Strengthen sparsity constraints; when classification error decreases ( <0), the current decomposition is stable enough, and the value can be appropriately reduced. Or improve .

[0072] The effectiveness of the scheme described in this invention will be analyzed in detail below using simulation signals. First, the effectiveness of the adaptive VMD decomposition results is verified based on simulated signals. Two parameters affect the VMD decomposition effect: the number of modal components K and the quadratic penalty parameter α. Therefore, the influence of parameter settings on the decomposition effect is mainly verified using simulated signals. The mathematical expression of the simulated signal is as follows:

[0073] The main component of the analog signal is the oscillation signal. x 1. Low-frequency signals x 2 andx The signal consists of three components with center frequencies of 50Hz, 25Hz, and 5Hz, respectively. To make the analog signal more closely resemble a real vibration signal, a Gaussian white noise signal is added to the analog signal. noise , Figure 4 For analog signal signal Original waveform image.

[0074] from Figure 5 As can be seen, all three sub-signals in the analog signal were successfully extracted, with the modal component number parameter being K. = 3. α = 3000, corresponding to IMF1 at 50Hz, IMF2 at 25Hz, and IMF3 at 5Hz, respectively. It is evident that adaptive VMD can effectively extract the corresponding sub-signals with minimal information loss.

[0075] Secondly, the accuracy of fault diagnosis for the crank-connecting rod component based on dual-layer adaptive collaborative optimization was verified using simulation signals. Model verification was performed using vibration data output from the established crank-connecting rod component fault simulation model, which included nine sensor locations collecting vibration data. Each location contained data on normal, slight, moderate, and severe wear of the sliding bearing, with 30 sets of wear clearance data for each location, totaling 120 sets of data. Each simulation data sample was sampled at a frequency of 12.5 kHz for one cycle. Time-domain analysis of the signals yielded the following feature extraction values, as shown in Table 1.

[0076] Table 1. Extracted Time-Domain Features of Vibration Signals

[0077] When using SSA to optimize VMD-related hyperparameters, the relevant settings are shown in Table 2.

[0078] Table 2 SSA Related Parameters

[0079] After adaptive VMD decomposition, the vibration signals under four different wear conditions were obtained as optimized parameters (10,6000), (10,5932), (6,4188) and (7,4921), respectively. Figure 6 The spectrum of IMFs obtained after decomposing a severely worn signal is shown. The decomposition results of the IMF components show that there is no mode mixing phenomenon in the individual sub-signals obtained after decomposition, and each mode is successfully separated from the original signal.

[0080] The above IMFs are subjected to fuzzy entropy calculation. Since the SVM model requires the input feature vectors to have the same dimension, otherwise sample classification cannot be achieved, the maximum dimension of the fuzzy entropy feature vector of the IMFs input to the SVM is 6. To improve the accuracy of fault diagnosis and reduce the impact of scale differences between different features on the SVM classifier, the above time-domain feature values ​​and the fuzzy entropy of the decomposed IMFs components are normalized. The normalized data of the time-domain features are shown in Table 3, and the normalized data of the fuzzy entropy of the IMFs components are shown in Table 4.

[0081] Table 3. Data table of time-domain feature mean normalization.

[0082] Table 4. Normalized data table of frequency domain feature mean

[0083] 120 datasets were used, divided into training and testing sets at a ratio of 2:1. Each dataset was assigned a label: 1 for normal condition, 2 for slight wear, 3 for moderate wear, and 4 for severe wear. All labeled data were used as input features to an optimized Support Vector Machine (SVM) classifier for fault diagnosis. To ensure the statistical reliability of the classification results, 10 independent fault diagnosis experiments were conducted. In each experiment, the training and testing sets were randomly re-split, and the classifier parameters were adaptively adjusted using a collaborative optimization mechanism. To reduce errors caused by random data allocation, the final results were statistically analyzed using the average accuracy of the 10 classification experiments, as shown in Table 5. The specific classification results are presented in... Figure 7 The value includes the classification accuracy for each test set.

[0084] Table 5. SVM prediction results for 10 runs

[0085] In summary, examples demonstrate that the fault diagnosis method presented in this paper can effectively improve the accuracy of fault diagnosis for crankshaft and connecting rod components.

[0086] Based on the same inventive concept, this embodiment provides a computer device, including: a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements a two-layer adaptive collaborative optimization method for fault diagnosis of crank connecting rod components.

[0087] This embodiment provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements a two-layer adaptive collaborative optimization method for diagnosing crank-connecting rod components.

[0088] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0089] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0090] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A double-layer self-adaptive collaborative optimization crank connecting rod component fault diagnosis method, characterized in that, include: S1. Acquire vibration signals of the crankshaft and connecting rod components; S2. Based on the sparrow search optimization algorithm, a VMD optimization model with average envelope entropy and modal stability as joint objective functions is established to search for the optimal VMD parameter combination of vibration signal under each state; S3. Set the VMD parameters to the optimal parameter combination of the vibration signal in each state, decompose the vibration signal in each state, and obtain several IMFs; S4. Calculate the fuzzy entropy and several time-domain features for each IMF to form a multidimensional feature vector; S5. The Bayesian optimization algorithm is used to jointly optimize the penalty factor and kernel function parameters of SVM, search for the best parameter combination of SVM, input the multidimensional feature vector into the SVM classifier for training, obtain the SVM prediction model for vibration signal under different states, and dynamically adjust the weight coefficients in the signal decomposition layer based on the prediction error feedback of the classification layer to obtain the optimized SVM prediction model. S6. Input the multidimensional feature vector corresponding to the test signal into the optimized SVM prediction model to obtain the fault classification result.

2. The method of claim 1, wherein, S2 includes: S201 Set the VMD parameter optimization range, initialize the sparrow search optimization algorithm parameters: set the number of decomposition modes , the penalty factor , the number of sparrows n, and determine the number proportion of explorers, followers and sentinels in the sparrows, each individual corresponds to a set of VMD parameters , i=1,2,...n, set the maximum number of iterations or convergence condition; S202 Calculate initial population fitness: for the i-th sparrow individual, calculate the fitness value of the VMD parameter combination by the fitness function VMD decomposition is performed on the vibration signal to obtain IMF components, calculate the envelope entropy of each IMF component, and calculate the average envelope entropy ; calculate the mean of the center frequency of each IMF component and the center frequency variance , and calculate the modal stability , so as to calculate the joint objective function under the VMD parameter combination ;​ S203 SSA iterative optimization: according to the joint objective function, sort the n individuals of the initial population in ascending order of fitness value, and keep the initial optimal individual VMD parameter combination As a reference point for subsequent iterations, the subsequent iterations recalculate the population fitness by updating the explorer, follower and sentinel positions, and keep the current minimum fitness value The corresponding VMD parameter combination ; S204: If the number of iterations reaches the maximum number of iterations or the convergence condition is met, terminate the iteration; otherwise, repeat S203. After the iterations stop, the VMD parameter combination that minimizes the joint objective function value is selected as the optimal VMD parameter combination .

3. The method of claim 2, wherein, The explorer location update formula is: wherein, t is the current iteration number, j = 1, 2, , d, d represents the individual dimension, represents the value of the t th sparrow in the i th dimension in the explorer population at the j th iteration, represents the value of the th sparrow in the i th dimension in the explorer population at the j th iteration, iter max is the maximum iteration number, Q is a random number, L is a row vector with length d and elements of 1, , are the current alert threshold and danger threshold of the whole population, respectively; The follower position update formula is: in, Indicates the first The optimal position for the explorer in the next iteration. Indicates the first In the second iteration, the first in the follower group i The sparrow in the first j The numerical value of the dimension. Indicates the first t The worst position globally in the next iteration. Indicates the first t In the second iteration, the first in the follower group i The sparrow in the first j The numerical value of the dimension. represent , represent A matrix, where the elements are randomly assigned values ​​of 1 or -1; The formula for updating the vigilant's position is: in, Indicates the first t In the next iteration, the first group of vigilants... i The sparrow in the first j The numerical value of the dimension. Indicates the first In the next iteration, the first group of vigilants... i The sparrow in the first j The numerical value of the dimension. Indicates the first t The optimal global position in the next iteration is determined by ε, which is a random number to avoid errors where the denominator is zero during the iteration process. f i Indicates the first i The fitness value of a sparrow. f g and f w These represent the current global best and worst fitness, respectively. It is a random number within the range [-1, 1], used to control the direction and step size of the optimal individual's local exploration. It is a random number that follows a normal distribution and is used to control the randomness of the step size when an individual moves toward the global optimal position.

4. The method as described in claim 2, characterized in that, The VMD optimization model with average envelope entropy and modal stability as joint objective functions is as follows: in, The average envelope entropy, Let be the modal stability function. , The adjustable weights of the VMD objective function decomposition layer. + =1, It is the joint objective function of the VMD optimization model, and the objective is to minimize To obtain the optimal decomposition parameters; in, K The number of IMFs. Envelope entropy; in, For the next disturbance The center frequency of the mode, Its variance reflects stability. It is the first disturbance The average value of the center frequencies of each modality.

5. The method as described in claim 1, characterized in that, In S4, the fuzzy entropy calculation process is as follows: (1) Given an N-dimensional IMF sequence Construct a phase space vector of length m: ； (2) Definition , Distance: (3) Calculate fuzzy similarity: Where m represents the embedding dimension, and r represents the similarity tolerance, controlling the width of the fuzzy function. x(N) This represents the data value at the Nth time point in the IMF sequence. That is, from the first in the IMF sequence p Starting from one data point, continuously take... m A set of data points constitutes a m A dimensional vector, where s represents the position index within the phase space vector. This represents the maximum absolute value of the difference between corresponding elements of two m-dimensional vectors; This represents the fuzzy similarity between two m-dimensional vectors. (4) For each vector Calculate the average fuzzy similarity with other vectors: The average fuzzy similarity is taken as the mean: (5) Repeat the above steps to obtain Calculate the fuzzy entropy: in, Indicates the first p The average fuzzy similarity of the vectors This represents the mean fuzzy similarity when the embedding dimension is m. This represents the mean fuzzy similarity when the embedding dimension is m+1.

6. The method as described in claim 1, characterized in that, The Bayesian optimization algorithm is used to jointly optimize the penalty factor and kernel function parameters of the SVM. The joint objective function is: Where c is the penalty factor and g is the kernel function parameter. To verify the error, This is a complexity term, reflecting the model's complexity and computational cost. , For balance coefficient, + =1; in, It is the number of samples in the validation set. Representative sample true label, The model predicts the label. g (x) is an indicator function that takes the value 0 or 1.

7. The method as described in claim 1, characterized in that, In step S5, based on the prediction error feedback of the classification layer, the weight coefficients in the signal decomposition layer are dynamically adjusted, including: in, t This represents the current iteration number. For learning rate, The average envelope entropy and modal stability in the VMD objective function decomposition layer are respectively the values ​​at the 1st, 2nd, and 3rd respectively. Adjustable weights for each iteration Let be the adjustable weight for the average envelope entropy in the (t+1)th iteration. For the adjustable weights of modal stability in the (t+1)th iteration, This represents the change in classification error.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

Images