Connecting rod fatigue life evaluation method and device, electronic equipment and storage medium

Abstract

The application provides a connecting rod fatigue life evaluation method and device, electronic equipment and a storage medium. Original data of connecting rod fatigue life under multi-stage load is obtained, and a structured data set is constructed. A deep neural network model is trained according to the structured data set. The life uncertainty of each load point in a preset target load range is predicted according to the trained network model, and the load point with the maximum life uncertainty is selected for fatigue test. The network model is updated according to the test result, and the optimized S-N curve parameters and life distribution parameters are obtained. Based on the real-time load spectrum data, the optimized S-N curve parameters and the life distribution parameters, the amplitude load life is calculated through the life distribution formula and the damage formula, and the real-time fatigue life and the damage contribution distribution under different preset survival rates are obtained. The application improves the accuracy of the connecting rod fatigue life evaluation, and realizes the accurate prediction and dynamic evaluation of the fatigue life under different survival rates.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of computers, and particularly relates to a connecting rod fatigue life evaluation method and device, electronic equipment and a storage medium. BACKGROUND

[0002] As a key load-bearing component in a mechanical transmission system, a connecting rod is widely used in automotive engines, engineering machinery, ship power and other equipment. Its working environment is complex and it bears periodic alternating loads. Fatigue failure is one of the main reasons for the scrapping of connecting rods and even the triggering of equipment failure. Therefore, accurately evaluating the fatigue life of connecting rods is of great significance to ensuring the safety of equipment operation, reducing maintenance costs and optimizing product design.

[0003] Traditional connecting rod fatigue life evaluation methods are mainly based on material S-N curves (stress-life curves) and Miner linear cumulative damage theory. Life data under specific loads are obtained through experiments, and then the fatigue life under different working conditions is extrapolated and predicted. However, this method has significant limitations. On the one hand, the actual working load of the connecting rod often presents multi-level and variable amplitude characteristics, and is affected by factors such as material performance dispersion, geometric size deviation and stress concentration effect. The fatigue life has significant uncertainty, and this method is difficult to accurately quantify this uncertainty. On the other hand, obtaining comprehensive fatigue test data requires a large amount of time and cost, and the test period is long, making it difficult to meet the needs of rapid evaluation in engineering practice. SUMMARY

[0004] Therefore, the present application aims to provide a connecting rod fatigue life evaluation method, device, electronic equipment and storage medium to solve at least one of the above problems.

[0005] To achieve the above-mentioned purpose, the technical solution of the present application is as follows:

[0006] In a first aspect, the present application provides a connecting rod fatigue life evaluation method, comprising:

[0007] Obtaining connecting rod fatigue life raw data under multi-level loads, preprocessing by converting multi-level load amplitudes into stress parameters and labeling, to obtain a structured data set, training a deep neural network model according to the structured data set to obtain network weight parameters, S-N curve parameters and life distribution parameters, wherein the deep neural network model is updated by the network weight parameters;

[0008] According to the trained network model, the life uncertainty of each load point in the preset target load range is predicted, and the load point with the maximum life uncertainty is selected for fatigue testing. According to the test results, the trained network model is updated to obtain the optimized S-N curve parameters and life distribution parameters.

[0009] Based on the real-time load spectrum data, the optimized S-N curve parameters and the life distribution parameters, the amplitude load life is calculated through a life distribution formula and a damage formula to obtain the real-time fatigue life and the damage contribution degree distribution under different preset survival rates.

[0010] In a second aspect, based on the same inventive concept, the application further provides a connecting rod fatigue life evaluation device, comprising:

[0011] The model training module is configured to obtain connecting rod fatigue life original data under multi-stage loads, preprocess the multi-stage load amplitudes by converting them into stress parameters and labeling them to obtain a structured data set, train a deep neural network model according to the structured data set to obtain network weight parameters, S-N curve parameters and life distribution parameters, and update the deep neural network model through the network weight parameters.

[0012] The model optimization module is configured to predict the life uncertainty of each load point in a preset target load range according to the trained network model, select the load point with the maximum life uncertainty for fatigue testing, and update the trained network model according to the test results to obtain optimized S-N curve parameters and life distribution parameters.

[0013] The life evaluation module is configured to calculate the amplitude load life based on the real-time load spectrum data, the optimized S-N curve parameters and the life distribution parameters through a life distribution formula and a damage formula to obtain the real-time fatigue life and the damage contribution degree distribution under different preset survival rates.

[0014] In a third aspect, based on the same inventive concept, the application further provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to realize the method of the first aspect.

[0015] In a fourth aspect, based on the same inventive concept, the application further provides a non-transitory computer readable storage medium, wherein the non-transitory computer readable storage medium stores computer instructions for causing the computer to execute the method of the first aspect.

[0016] Compared with the prior art, the connecting rod fatigue life evaluation method, device, electronic device and storage medium provided by the application have the following beneficial effects:

[0017] The connecting rod fatigue life evaluation method provided in the application improves the accuracy of the connecting rod fatigue life evaluation, enhances the generalization ability of the model under complex load conditions, and realizes the accurate prediction and dynamic evaluation of the fatigue life under different survival rates by fusing data driving and physical rules to construct a deep neural network and combining test optimization. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein in their entirety, and they are used to explain the application together with the description. In the drawings:

[0019] Figure 1 A connecting rod fatigue life evaluation method flow chart is provided in the embodiments of the application.

[0020] Figure 2 A connecting rod fatigue life evaluation device structure schematic diagram is provided in the embodiments of the application.

[0021] Figure 3 An electronic device hardware structure schematic diagram is provided in the embodiments of the application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below with reference to the embodiments and the accompanying drawings.

[0023] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the application should be understood as the general meaning understood by those skilled in the art to which the embodiments of the application belong. The terms "first", "second" and the like used in the embodiments of the application do not represent any order, number or importance, but are only used to distinguish different components. The terms "include" or "contain" and the like mean that the elements or objects before the terms cover the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and the like are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships may also change accordingly.

[0024] The embodiments of the application are described in detail below with reference to the accompanying drawings.

[0025] Please refer to Figure 1 As shown in the drawings, the embodiments provide a connecting rod fatigue life evaluation method, which specifically includes the following steps:

[0026] Step S1, obtaining the original data of the fatigue life of the connecting rod under multi-level load, preprocessing by converting the multi-level load amplitude into stress parameters and labeling, to obtain a structured data set, training a deep neural network model according to the structured data set, to obtain network weight parameters, S-N curve parameters and life distribution parameters.

[0027] Specifically, in the present embodiment, the connecting rod often bears different levels and frequencies of load in actual work, and the test data under multi-level load can more comprehensively reflect its fatigue characteristics under complex working conditions. Since fatigue life is directly related to stress state, stress amplitude as a key parameter to characterize the size of alternating stress can more directly reflect the mechanical nature of fatigue damage caused by stress amplitude. Moreover, material properties and geometric characteristics are important factors affecting the fatigue life of the connecting rod, and the fatigue resistance of different materials is different, and the minimum cross-sectional area is often the key part of stress concentration. Therefore, the material properties such as material tensile strength and geometric properties such as the minimum cross-sectional area of the test site are labeled, and the survival rate label is labeled to meet the life evaluation requirements under different reliability requirements in engineering.

[0028] Then, the above data is preprocessed to eliminate possible errors, noises or inconsistencies in the original data, ensure the accuracy and reliability of the data, and obtain a structured data set. The data set integrates the key factors affecting the fatigue life of the connecting rod and can provide comprehensive information for model learning.

[0029] A deep neural network is constructed, wherein the input layer is used to receive each feature parameter in the structured data set, the hidden layer learns the potential laws and complex relationships in the data through nonlinear transformation, the physical constraint layer integrates the basic physical rules in fatigue mechanics such as the characteristics of S-N curve and the influence law of average stress on fatigue life into the model, avoiding the model only relying on data fitting and possibly appearing prediction results violating the physical principles, and the output layer is used to output parameters related to fatigue life.

[0030] Illustratively, the deep neural network is trained according to the structured data set by fusing data loss and physical rules, wherein the fusion of data loss can make the prediction results of the model as close as possible to the actual test data, ensuring the accuracy of the prediction, and the fusion of physical rules can improve the generalization ability and reliability of the model, improving the accuracy of small sample prediction. Through training, network weight parameters, S-N curve parameters and life distribution parameters are obtained. The network weight parameters determine the processing method of the model to the input features, the S-N curve parameters describe the relationship between stress and life, and the life distribution parameters are used to quantify the uncertainty of fatigue life.

[0031] In some embodiments, the stress amplitude data set is obtained by converting the load amplitude values into stress amplitudes according to the ratio of the load to the minimum cross-sectional area of the test site.

[0032] The labeled data set is obtained by labeling the material properties, geometric properties, and survival rate labels, respectively, on the stress amplitude data set, and labeling the corresponding average stress, wherein the material properties include the material tensile strength, and the geometric properties include the minimum cross-sectional area of the test site.

[0033] The pre-processed data set is obtained by performing data cleaning processing on the labeled data set, and the structured data set is obtained by performing feature extraction processing on the pre-processed data set.

[0034] Specifically, in the present embodiment, the original fatigue life data of the connecting rod under multi-level load is 6 sampling fatigue life data corresponding to 5 load levels of S1=202.03kN, S2=196.59kN, S3=185.71kN, S4=180.27kN, and S5=174.83kN, for example, the life data under S1 load is 164739, 108448, 598017, 188370, 117036, and 101203. Based on the above original data, the stress amplitude data set corresponding to each load is obtained by converting the load amplitude values into stress amplitudes according to the ratio of the load to the minimum cross-sectional area of the test site of the connecting rod, for example, the minimum cross-sectional area of the connecting rod body is measured as 411.38mm², and then the stress amplitude data set corresponding to each load is obtained by calculating the ratio of the load amplitude to the minimum cross-sectional area of the rod body, for example, the stress amplitude corresponding to S1=202.03kN is 202.03×1000 / 411.38≈491.1MPa. Then, the above obtained stress amplitude data set is labeled with material properties, geometric properties, and survival rate labels, respectively, and the corresponding average stress is labeled.

[0035] Among them, the material property is labeled as the ultimate tensile strength of the material used for the connecting rod, which is 1080MPa, the geometric property is labeled as the determined minimum cross-sectional area of 411.38mm², and the survival rate label is labeled as 0.5, 0.95, and 0.99 three common survival rate levels based on the statistical requirements in the fatigue test design. The average stress is measured according to the stress state of the connecting rod in actual work, for example, the average stresses of working conditions 1 to 4 are-175.8MPa, -173.6MPa, -149.2MPa, and-160.4MPa, respectively, and then these average stresses are associated with the corresponding stress amplitudes to form a labeled data set.

[0036] Subsequently, in the data cleaning process of the labeled data set, the embodiment adopts the 3σ criterion for abnormal value detection, calculates the mean μ and standard deviation σ of the life data corresponding to each stress amplitude, and determines the data exceeding the range of [μ-3σ, μ+3σ] as abnormal values and removes them. For example, for the life data under S3 load, if a data exceeds the range, it is considered abnormal and removed. For a small amount of missing data, such as the blank values in some load columns in the original data, the adjacent value interpolation method is used for supplementation, that is, the average value of the adjacent two effective data in the column is selected for filling. Finally, the preprocessed data set is subjected to feature extraction processing, that is, the corresponding stress amplitude, average stress, material tensile strength, minimum cross-sectional area of the test site, and survival rate are extracted. Based on the preprocessed stress amplitude and average stress, the stress ratio is calculated to reflect the degree of asymmetry of stress cycles, or based on the material tensile strength and stress amplitude, the strength reserve coefficient is constructed to represent the reserve ability of material strength relative to working stress. The structured data set is obtained by combining the above features.

[0037] In some embodiments, the deep neural network model includes an input layer, a hidden layer, a physical constraint layer, and an output layer, wherein the input layer is used to receive the stress amplitude, the average stress, the material tensile strength, and the minimum cross-sectional area of the test site, the hidden layer processes the data features through the LeakyReLU activation function, the physical constraint layer fuses the physical rules through a preset formula, the physical rules include the Haigh correction function, and the output layer is used to output the life distribution parameters and the S-N curve parameters.

[0038] A loss function that fuses data loss, Haigh correction function constraint, and Miner criterion constraint is constructed, and the deep neural network is trained according to the structured data set through the loss function, and the training is stopped when a preset condition is met, to obtain the network weight parameters, the S-N curve parameters, and the life distribution parameters.

[0039] Specifically, in the embodiment, the structured data set is used as input to construct a deep neural network, wherein the preset formula of the physical constraint layer is:

[0040] ;

[0041] wherein, is the output feature of the physical constraint layer, is the weight matrix from the hidden layer to the physical constraint layer, is the output feature of the hidden layer, is the hyperbolic tangent activation function for feature nonlinear conversion, is the Haigh correction function, and , represents the stress amplitude, is the material tensile strength for correcting the average stress the influence on fatigue life;

[0042] The loss function is constructed by fusing the data loss, the Haigh correction constraint function and the Miner criterion constraint, wherein the data loss is calculated by the logarithmic error of the predicted life and the real life, the Haigh correction constraint is calculated by the error between the average stress correction result of the prediction and the theoretical Haigh correction result, and the Miner criterion constraint is calculated by the deviation of the damage accumulation value from 1.

[0043] Further, the deep neural network is a full connection feedforward neural network, the input layer takes the stress amplitude, the average stress, the material tensile strength and the minimum cross-sectional area of the test part as the model input features. The hidden layer adopts a 3-layer full connection structure, the number of neurons of each layer is 64, 32 and 16 in turn, the features transmitted from the input layer are processed by the LeakyReLU activation function, thereby the model can introduce nonlinearity while retaining part of the negative feature information, enhance the fitting ability of the model to complex data relationships, and generate hidden layer output features by nonlinearly transforming the input features. The physical constraint layer fuses the influence of the Haigh correction function on the fatigue life by the above preset formula. The Haigh correction function quantifies the influence of the average stress on the fatigue life through the material tensile strength, that is, when the average stress is tensile stress, the correction coefficient decreases to reflect the life attenuation effect, and when the average stress is compressive stress, the correction coefficient increases to reflect the life improvement law.

[0044] Illustratively, the output layer adopts a 2-parallel-branch structure, the first branch outputs the life distribution parameters, including the shape parameter and the scale parameter of the fatigue life obeying distribution, i.e., the Weibull distribution, which is used to describe the statistical discreteness of the life, and the second branch outputs the S-N curve parameters, i.e., the coefficient K and the index b in the power function form wherein N is the cyclic life, and σ is the stress amplitude. The output layer activation function adopts the SoftPlus function for the distribution parameters to ensure the non-negativity of the parameters, and adopts the linear activation function for the S-N curve parameters to retain the physical dimensional characteristics. In the loss function construction stage, a multi-objective constraint mechanism is adopted to realize the regularization of the model.

[0045] Illustratively, the mathematical expression of the loss function is as follows:

[0046] ;

[0047] ;

[0048] ;

[0049] ;

[0050] In the formula, is a result of the loss function, is a data loss, is a sample size, is a predicted life, is a true life, is a Haigh correction constraint, is an average stress correction factor for the sample predicted by the deep neural network in the physical constraint layer, is a theoretical correction factor for the sample calculated by the Haigh theoretical formula, is a Miner criterion constraint, is a load step, the load step corresponds to the number of groupings of different stress amplitudes in the structured data set, is the number of load cycles of the level, and are S-N curve parameters, and are constraint weights.

[0051] During training, the data of the structured data set is divided into a training set, a validation set, and a test set according to a certain ratio such as 7:2:1, the training set is used for model parameter updating, the validation set is used to monitor the model generalization ability during training, and the test set is used for final evaluation. For example: using the training set data, the gradient of each layer parameter of the network is calculated based on the total loss function through the back propagation algorithm, and the network weight parameters are gradually updated using a gradient descent optimizer such as the Adam optimizer, and the calculation logic and parameter configuration of each layer of the network are continuously adjusted. And in the training iteration process, the model performance is continuously verified on the validation set, and when the preset condition is met, the training is stopped, the preset condition is set to the loss function value on the validation set not decreasing continuously for a preset number of rounds such as 20 rounds, or the training reaches a maximum number of iterations such as 10,000 times. After training is completed, stable network weight parameters are obtained, and at the same time, S-N curve parameters and life distribution parameters output by the output layer are obtained. The above parameters are directly used for subsequent evaluation and analysis of connecting rod fatigue life, and provide model and parameter support for fatigue life prediction of connecting rods under actual working conditions.

[0052] Step S2, according to the trained network model, the life uncertainty of each load point in the preset target load range is predicted, and the load point with the largest life uncertainty is selected for fatigue test, and the trained network model is updated according to the test result, to obtain optimized S-N curve parameters and life distribution parameters.

[0053] Specifically, in this embodiment, the network weight parameters are updated to the deep neural network to obtain the trained network model, but since the initial training data may have limitations, the reliability of the prediction result still needs to be improved.

[0054] Illustratively, subsequently, the life uncertainty of each load point in the preset target load range is predicted according to the trained network model. The life uncertainty reflects the dispersion degree and reliability degree of the prediction result of the model at the load point. The greater the uncertainty, the less the model understands the load point, and the lower the credibility of the prediction result. By selecting the load point with the maximum life uncertainty for fatigue test, the model's cognition can be maximized with the least test cost by supplementing test data in the most uncertain area of the model, thereby making up for the deficiency of the initial data. The test results obtained are updated to the trained network model, and the network parameters are adjusted by retraining, so that the model absorbs new information, thereby obtaining an optimized network model, S-N curve parameters and life distribution parameters. The prediction accuracy and reliability of the optimized model in the target load range are significantly improved, and the fatigue life characteristics of the connecting rod can be more accurately reflected.

[0055] In some embodiments, based on the trained network model and the preset target load range, multiple Monte Carlo samplings are performed at each load point to obtain multiple sets of predicted life results corresponding to the load points, and the squared difference mean of each set of predicted life results and the average predicted life of the corresponding load point is calculated to obtain the life uncertainty of each load point;

[0056] Based on the life uncertainty of each load point and the corresponding average predicted life, the ratio of the life uncertainty to the average predicted life of each load point is calculated, and the load point with the maximum ratio is selected as the test point to be verified;

[0057] Fatigue test is performed at the stress corresponding to the test point to be verified, and the cycle number of continuous loading until the connecting rod fails is recorded as the test result;

[0058] Based on the test result, the weight parameters of the trained network model are adjusted by the Adam optimization algorithm, and the uncertainty prediction, test point selection, test and model updating process are repeated until the preset termination condition is met, and finally an optimized network model is obtained;

[0059] The optimized S-N curve parameters and life distribution parameters are output based on the optimized network model.

[0060] Specifically, in the present embodiment, the target load range is set to 150-250 MPa (stress amplitude) according to the actual working conditions of the connecting rod, and is discretized into 21 load points at intervals of 5 MPa. For each load point, Monte Carlo sampling is performed based on the trained network model, that is, by introducing a slight perturbation in the stress amplitude, average stress, etc. parameters in the input layer of the network, with a perturbation amplitude of ±1% of the nominal value of the parameters, simulating measurement errors, and repeatedly performing the model inference process 100 times, 100 sets of predicted life results are obtained. Then the average predicted life of the load point is calculated, and the life uncertainty is calculated by the mean square difference formula, which quantifies the prediction dispersion degree of the model at this load point, and the larger the value, the more blurred the model's understanding of the working condition.

[0061] Illustratively, in order to balance the influence of the magnitude difference of life, the present embodiment takes the ratio of uncertainty to average predicted life as a screening index, which not only considers the absolute uncertainty, but also eliminates the interference of the difference in life value between different load points by normalizing the average life. For example, if the average life of a high stress load point is times, the uncertainty is , and the ratio is 0.1, while the average life of a low stress load point is times, the uncertainty is 2 , and the ratio is 0.2, then the latter is preferentially selected because of the higher ratio. By traversing the ratios of the 21 load points, the load point corresponding to the maximum ratio is selected as the to-be-verified test point, ensuring that the new test data can maximize the reduction of the model's cognitive blind area. Subsequently, fatigue tests are carried out under the stress conditions corresponding to the determined to-be-verified test point, that is, a sinusoidal alternating load is applied using an electro-hydraulic servo fatigue testing machine, the stress ratio R is -1, the loading frequency is 10 Hz, and three samples are tested in parallel for each test to reduce accidental errors. Moreover, during the test, the stress response of the dangerous section of the connecting rod is monitored in real time through strain gauges, and when the strain increases suddenly, such as more than 5% of the initial value, it is determined to be failed, then the corresponding cycle number is recorded as the single-sample life value, and the arithmetic mean of the three samples is taken as the final test result of the test point.

[0062] Illustratively, based on the obtained test results, the model is updated, for example, the stress parameters such as stress amplitude, average stress, etc. of the test point are combined with the measured life to form a new sample, and are added to the original structured data set to form an expanded data set. The embodiment adopts the Adam optimization algorithm to perform secondary training on the trained network model, and maintains the learning rate at 1 / 10 of the initial training stage during the optimization process to avoid parameter fluctuations. The same loss function is used, but the new sample needs to be assigned a weight of 3 times to strengthen the modification effect of the new data on the model. After completing a model update, the foregoing uncertainty prediction, test point selection and test process are repeatedly executed to form a closed-loop iteration. In addition, the iteration termination condition is set to double constraints, for example, when the maximum value of the ratio of all load points is less than a preset threshold of 0.05, it is determined that the model accuracy meets the engineering requirements, or if the number of iterations reaches a preset upper limit, the iteration is forcibly terminated. Finally, the network model obtained when the termination condition is met is the optimized network model, and the weight parameters thereof are more consistent with the measured data law than the initial training model. Moreover, through the output layer of the optimized network model, the optimized S-N curve parameters and life distribution parameters are obtained, providing higher-precision parameter support for subsequent life assessment.

[0063] Step S3, based on the real-time load spectrum data, the optimized S-N curve parameters and the life distribution parameters, the variable amplitude load life calculation is performed through the life distribution formula and the damage formula to obtain the real-time fatigue life and the damage contribution degree distribution under different preset survival rates.

[0064] Specifically, in the embodiment, the load spectrum data reflects the load change in the actual work of the connecting rod, and the optimized S-N curve parameters and the life distribution parameters ensure the accuracy and reliability of the calculation. Based on the fatigue cumulative damage theory, the damage caused by each load cycle to the connecting rod is quantified to obtain a single-cycle damage, and the size of the single-cycle damage is related to the stress level of the cycle and the fatigue characteristics of the material. By accumulating all single-cycle damages, the total damage is calculated, and the size of the total damage directly reflects the fatigue damage degree of the connecting rod under the entire load spectrum. Moreover, by combining the life distribution formula, the total damage can be associated with the fatigue life under different survival rates, thereby performing variable amplitude load life calculation. Finally, the real-time fatigue life and the damage contribution degree distribution under different preset survival rates are obtained.

[0065] Among them, the real-time fatigue life provides a key basis for the remaining life assessment and maintenance decision of the connecting rod, and the damage contribution degree distribution can clearly determine the contribution proportion of different load cycles to the total damage, which helps to identify the load component that has the greatest impact on the fatigue life of the connecting rod, and provides guidance for optimizing the load working condition and improving the service life of the connecting rod.

[0066] In some embodiments, the load spectrum data of the connecting rod is acquired in real time, and the load spectrum data includes stress amplitudes, mean stresses, and corresponding cycle numbers of a plurality of stress cycles;

[0067] With the load spectrum data, the optimized S-N curve parameters, and the optimized life distribution parameters as inputs, and in combination with a preset mean stress correction function based on a Gerber equation, single-cycle damages of each stress cycle are calculated;

[0068] According to the single-cycle damages of each stress cycle and the corresponding cycle numbers, a total damage is accumulated by the Miner criterion, and the total damage is a sum of products of the single-cycle damages of each stress cycle and the cycle numbers;

[0069] Based on the total damage and the optimized life distribution parameters, real-time fatigue lives under different preset survival rates are predicted according to a life distribution formula.

[0070] Specifically, in the present embodiment, the load spectrum data of the connecting rod is acquired in real time by a strain gauge array and a data acquisition system, and the data includes stress amplitudes, mean stresses, and corresponding cycle numbers of n stress cycles. The stress amplitude is calculated by stress peaks and valleys of adjacent cycles, and the mean stress is the average of the maximum and minimum stresses of the corresponding cycle. Subsequently, with the load spectrum data, the optimized S-N curve parameters, and the tensile strength of the material as inputs, and in combination with a mean stress correction function based on the Gerber equation, the single-cycle damages of each stress cycle are calculated.

[0071] Illustratively, based on the above-mentioned mean stress correction function, when the mean stress approaches 0, the function value approaches 1, and the damage calculation degenerates into a pure stress amplitude loading condition; when the mean stress increases, i.e., a tensile stress, the function value decreases to reflect the mechanical properties of tensile stress accelerating fatigue damage. The single-cycle damage calculation process ensures that the damage quantification conforms to the stress state-dependent characteristics of fatigue mechanics by embedding the influence of the mean stress into the damage calculation explicitly through the Gerber equation.

[0072] Further, in the variable-amplitude load life calculation, the total damage is calculated by the Miner criterion accumulation, and the formula is:

[0073] ;

[0074] wherein the single-cycle damage of each stress cycle is calculated by the following formula:

[0075] ;

[0076] In the formula, n is the number of stress cycles, is the number of actions of the nth stress cycle, is the number of actions of the nth stress cycle, is the number of actions of the nth stress cycle, single-cycle damage of the i-th stress cycle, the i-th stress cycle, the stress amplitude of the i-th stress cycle, and the optimized S-N curve parameters, the average stress correction function based on the Gerber equation, expressed as: , the average stress of the i-th stress cycle, the tensile strength of the material.

[0077] total damage represents the cumulative fatigue damage of the connecting rod under the current load spectrum. When , the connecting rod theoretically fails due to fatigue (corresponding to the median life of 50% survival rate). Find a constant equivalent stress amplitude Seq on the S-N curve of the median life versus stress relationship, such that after cycling times (total number of actions of each stress cycle) at Seq, the cumulative damage .

[0078] Take this calculated Seq and its corresponding equivalent average stress as input and substitute it into the optimized network model again. The network model outputs the corresponding life distribution parameters μ and σ based on Seq and the equivalent average stress.

[0079] Based on the output life distribution parameters and , the life at different survival rates is calculated using the following life distribution formula:

[0080] ;

[0081] wherein, is the fatigue life value when the preset survival rate is , is the mean value in the real-time life distribution parameters, is the standard deviation in the real-time life distribution parameters, is the quantile function of the standard normal distribution.

[0082] In some embodiments, the proportion of each stress cycle to the total damage is calculated to obtain a damage contribution distribution.

[0083] ​Specifically, in the present embodiment, the Miner linear damage accumulation criterion can be understood as that the fatigue damage has additivity, when the total damage reaches 1, the connecting rod fails due to fatigue, and through the corresponding relationship between the total damage and the failure critical damage (i.e. the total damage is 1), the life prediction value under different survival rates is deduced. Moreover, by statistically analyzing the contribution proportion of the total damage of each stress cycle to the overall total damage, the damage contribution degree is obtained, and by traversing all stress cycles, the damage contribution degree distribution is obtained.

[0084] The distribution can intuitively present the stress cycles in the load spectrum that play a leading role in fatigue damage, and provide key basis for load spectrum editing and life strengthening design. For example, if the total damage of a certain type of stress cycle, such as high stress amplitude or specific average stress combination, is significantly high, then the damage contribution can be reduced by optimizing the working condition or improving the structure.

[0085] In some embodiments, further comprising:

[0086] Based on the optimized network model, the grid parameters of the preset stress amplitude and average stress are traversed, the corresponding life values under different preset survival rates are calculated, and visual processing is performed to generate load-life three-dimensional mapping data, equal life curves and visual graphics.

[0087] Specifically, in the present embodiment, the stress amplitude and the average stress are two key stress parameters affecting the fatigue life, based on the optimized network model, the grid parameters of the preset stress amplitude and average stress are traversed, the life values under different combinations are systematically investigated, and thus the fatigue performance of the connecting rod under various stress states can be comprehensively mastered. Moreover, by calculating the corresponding life values under different preset survival rates, the diversified needs of reliability in different engineering scenarios are met, for example, in the case of extremely high safety requirements, the life value under high survival rate is concerned.

[0088] In addition, the present embodiment also performs visual processing based on the corresponding life values under different preset survival rates to generate load-life three-dimensional mapping data, equal life curves and visual graphics, which facilitates the understanding and application of engineering and technical personnel. Among them, the equal life curve can intuitively reflect the different combinations of stress amplitude and average stress under the same life, and provide clear reference basis for the design, use and maintenance of the connecting rod.

[0089] Illustratively, first, according to the stress analysis of the connecting rod in the actual power transmission scene, the stress amplitude range is accurately set to 200MPa to 300MPa, and the average stress range is set to -200MPa to 0MPa. And by using the equidistant scatter method, the stress amplitude is divided into 21 discrete points with a step of 5MPa, and the average stress is divided into 21 discrete points with a step of 10MPa, thereby forming 441 grid points, each grid point corresponding to a unique combination of stress amplitude and average stress, so as to completely cover the alternating stress working condition interval that the connecting rod can bear.

[0090] Subsequently, for each grid point, the corresponding stress amplitude, average stress, combined with the inherent properties of the connecting rod such as material tensile strength, minimum cross-sectional area, are input into the optimized network model. The model receives parameters through the input layer, processes features through the hidden layer LeakyReLU activation function, fuses the Haar modification function through the physical constraint layer to perform physical rule constraints, and finally outputs the real-time life distribution parameters of the grid point, i.e. mean and standard deviation, which quantitatively describe the statistical distribution characteristics of the fatigue life of the connecting rod under the corresponding stress working condition. Moreover, based on the above life distribution formula, the fatigue life values under different preset survival rates are calculated. The preset survival rates are 0.5, 0.9, 0.99, and then the fatigue life value set is obtained by summarizing the fatigue life values corresponding to different preset survival rates.

[0091] Further, the stress amplitude and average stress of each grid point are taken as two coordinates of the load dimension, and the fatigue life values under different preset survival rates are taken as life dimension data, which are stored in layers according to the survival rate. For example, three data layers with survival rates of 0.5, 0.9 and 0.99 are created, and each layer of data is indexed by the grid point coordinates and associated with the corresponding fatigue life value to form a mapping relationship data set of load and life in a three-dimensional space, providing a basic data structure for subsequent visualization. Moreover, in the load-life three-dimensional mapping data, for each preset survival rate layer, the isosurface point set of the preset life value is extracted. For example, the preset life value is set to The second loop iterates through the survival rate layer data and filters out the grid points with fatigue life values close to to form a discrete isosurface point set.

[0092] Since the grid points are evenly distributed, the isosurface point set is not a continuous curve, so the cubic spline interpolation method is used to fit it, that is, by constructing an interpolation function, the function is a cubic polynomial between adjacent isosurface points, and the first and second derivatives are continuous at the nodes. By solving the interpolation function coefficients, the discrete point set is converted into a smooth and continuous isosurface curve, which intuitively presents the same fatigue life working condition corresponding to different stress amplitude and average stress combinations under the corresponding survival rate.

[0093] Finally, according to the load-life three-dimensional mapping data, a three-dimensional rectangular coordinate system is constructed with the stress amplitude as the X-axis, the average stress as the Y-axis, and the fatigue life as the Z-axis. In this coordinate system, the data of different preset survival rate layers are loaded respectively. For each survival rate layer, according to its equal life curve and discrete grid point life value, color mapping and surface fitting techniques are used for visual rendering, that is, different colors are used to distinguish different survival rate layers, and discrete grid point life values are fitted into a continuous surface through surface fitting, and the equal life curve is used as the contour line of the surface. At the same time, by adding coordinate axis labels, color legends, life scales and other visualization elements, the final generated visualization graphics can intuitively and multi-dimensionally present the fatigue life distribution law of the connecting rod under different stress load combinations, which is convenient for engineers to quickly identify dangerous working conditions such as areas with significantly reduced life under high stress amplitude and specific average stress, and provides intuitive decision basis for the design optimization and life evaluation of the connecting rod.

[0094] The connecting rod fatigue life evaluation method described in the embodiment can systematically integrate key factors affecting fatigue life by converting connecting rod fatigue life original data under multi-level load into stress amplitude and labeling material, geometric properties and survival rate labels to construct a structured data set, providing a comprehensive data basis for deep neural network training. Secondly, the deep neural network containing the physical constraint layer is constructed and trained by fusing data loss and physical rules, which can make the model follow the basic principles of fatigue mechanics while learning data rules, avoid the prediction results from violating physical common sense, and improve the reliability of life evaluation. And based on the maximum prediction life uncertainty of the trained model, the model parameters are optimized based on the test results, further enhancing the generalization ability and prediction accuracy of the model in the full load range. Finally, based on the optimized model, a visual load-life mapping relationship is generated, and the life and damage contribution under variable amplitude load are calculated based on the real-time load spectrum, which can intuitively present the fatigue life law and realize dynamic evaluation, accurately quantify the fatigue performance under different survival rates, and provide a scientific basis for connecting rod design optimization, fault warning and maintenance decision-making.

[0095] It should be noted that some embodiments of the application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than those described above and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0096] Based on the same inventive concept, the embodiments of the application also provide a connecting rod fatigue life evaluation device corresponding to the method of any of the above embodiments.

[0097] AsFigure 2 The connecting rod fatigue life evaluation device shown comprises:

[0098] The model training module 11 is configured to obtain connecting rod fatigue life original data under multi-level loads, pre-process the multi-level load amplitudes by converting them into stress parameters and labeling them, to obtain a structured data set, train a deep neural network model according to the structured data set, to obtain network weight parameters, S-N curve parameters and life distribution parameters, and update the deep neural network model through the network weight parameters;

[0099] The model optimization module 12 is configured to predict the life uncertainty of each load point in a preset target load range according to the trained network model, select the load point with the maximum life uncertainty for fatigue testing, and update the trained network model according to the test results, to obtain optimized S-N curve parameters and life distribution parameters;

[0100] The life evaluation module 13 is configured to perform variable-amplitude load life calculation through a life distribution formula and a damage formula based on real-time load spectrum data, optimized S-N curve parameters and life distribution parameters, to obtain real-time fatigue life and damage contribution distribution under different preset survival rates.

[0101] For the convenience of description, the above device is described in various modules according to functions. Of course, the functions of the modules can be implemented in one or more software and / or hardware when implementing the embodiments of the present application.

[0102] The device of the above embodiments is used to implement the corresponding method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which are not described here.

[0103] Based on the same inventive concept, corresponding to any of the above method embodiments, the embodiments of the present application also provide an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method according to any of the above embodiments when executing the program.

[0104] Figure 3 A more specific hardware structure of an electronic device according to the present embodiment is shown in the schematic diagram, which can include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040 and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030 and the communication interface 1040 are connected to each other through the bus 1050 for communication within the device.

[0105] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, etc., for executing relevant programs to implement the technical solutions provided by the embodiments of the present specification.

[0106] The memory 1020 can be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 1020 can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present specification are implemented by software or firmware, the relevant program codes are saved in the memory 1020 and called and executed by the processor 1010.

[0107] The input / output interface 1030 is configured to connect input / output modules to realize information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. The input devices can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices can include a display, a speaker, a vibrator, an indicator light, etc.

[0108] The communication interface 1040 is configured to connect a communication module (not shown in the figure) to realize the communication interaction between the device and other devices. The communication module can realize communication through a wired manner (such as USB, network cable, etc.) or through a wireless manner (such as mobile network, WIFI, Bluetooth, etc.).

[0109] The bus 1050 includes a channel for transmitting information between various components (such as the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040) of the device.

[0110] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in the specific implementation process, the device can also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device can also only include the components necessary to implement the solutions of the embodiments of the present specification, and does not have to include all the components shown in the figure.

[0111] The electronic device of the above embodiments is used to implement the corresponding method in any of the preceding embodiments, and has the beneficial effects of the corresponding method embodiments, which are not described here again.

[0112] Based on the same inventive concept, the present application also provides a non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any of the above embodiments.

[0113] The computer readable medium of the embodiments can include permanent and non-permanent, removable and non-removable media, which can be implemented by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage device, or any other non-transmission medium that can be used to store information accessible by a computing device.

[0114] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to perform the method of any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which are not described here.

[0115] Those skilled in the art should understand that the discussion of any of the above embodiments is only exemplary, and is not intended to imply that the scope (including claims) of the present application is limited to these examples; the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of the different aspects of the embodiments of the present application as described above. In order to be brief, they are not provided in detail.

[0116] Although the present application has been described in conjunction with specific embodiments thereof, many alternatives, modifications and variations will be apparent to those skilled in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) can use the embodiments discussed.

[0117] The embodiments of the present application are intended to cover all such alternatives, modifications and variations as falling within the broad scope of the appended claims. Accordingly, any omission, modification, equivalent replacement, improvement, etc. made in the spirit and principle of the embodiments of the present application shall be included in the protection scope of the present application.

Claims

1. A method of evaluating a fatigue life of a connecting rod, characterized by, The method comprises the following steps: Obtain the original data of the fatigue life of the connecting rod under multi-level loads, preprocess the multi-level load amplitudes by converting them into stress parameters and labeling them, to obtain a structured data set, train a deep neural network model based on the structured data set, to obtain network weight parameters, S-N curve parameters and life distribution parameters, and update the deep neural network model through the network weight parameters; Predict the life uncertainty of each load point in a predetermined target load range based on the trained network model, and select the load point with the maximum life uncertainty for fatigue testing, and update the trained network model based on the test results to obtain optimized S-N curve parameters and life distribution parameters; Based on the real-time load spectrum data, the optimized S-N curve parameters and the life distribution parameters, calculate the variable amplitude load life through the life distribution formula and the damage formula to obtain the real-time fatigue life and damage contribution distribution under different predetermined survival rates, comprising: Real-time acquisition of load spectrum data of the connecting rod, the load spectrum data containing stress amplitudes, average stresses and corresponding cycle times of multiple stress cycles; Taking the load spectrum data, the optimized S-N curve parameters and the optimized life distribution parameters as inputs, and combining a predetermined average stress correction function based on the Gerber equation, the single-cycle damage of each stress cycle is calculated; According to the single-cycle damage of each stress cycle and the corresponding cycle times, the total damage is accumulated through the Miner criterion, which is the sum of the product of the single-cycle damage of each stress cycle and the cycle times; Based on the total damage and the optimized life distribution parameters, the real-time fatigue life under different predetermined survival rates is predicted according to the life distribution formula.

2. The method of claim 1, wherein: According to the ratio relationship between the load and the minimum cross-sectional area of the connecting rod test site, the load amplitude of each level is converted into a stress amplitude to obtain a stress amplitude data set; By labeling the material properties, geometric properties and survival rate labels of the stress amplitude data set respectively, and labeling the corresponding average stress, a labeled data set is obtained, wherein the material properties include material tensile strength, and the geometric properties include the minimum cross-sectional area of the test site; The labeled data set is subjected to data cleaning processing to obtain a preprocessed data set, and the preprocessed data set is subjected to feature extraction processing to obtain the structured data set.

3. The method of claim 2, wherein: The deep neural network model comprises an input layer, a hidden layer, a physical constraint layer and an output layer, wherein the input layer is used to receive the stress amplitude, the average stress, the material tensile strength and the minimum cross-sectional area of the test site, the hidden layer processes data features through a LeakyReLU activation function, the physical constraint layer fuses physical rules through a predetermined formula, the physical rules include a Haigh correction function, and the output layer is used to output the life distribution parameters and the S-N curve parameters; wherein the predetermined formula is: ； wherein, is the output feature of the physical constraint layer, is the weight matrix of the hidden layer to the physical constraint layer, is the output feature of the hidden layer, is the hyperbolic tangent activation function for feature non-linear transformation, is the Haar modification function, and , denotes the stress amplitude, is the tensile strength of the material for modifying the mean stress effect on fatigue life; A loss function is constructed by fusing data loss, Haigh correction function constraint and Miner criterion constraint, the deep neural network is trained according to the structured data set through the loss function, and the training is stopped when a preset condition is met, so as to obtain the network weight parameter, the S-N curve parameter and the life distribution parameter.

4. The method of claim 1, wherein: Based on the trained network model and the preset target load range, multiple Monte Carlo samplings are performed at each load point to obtain corresponding multiple groups of predicted life results, and the square difference mean of each group of predicted life results and the average predicted life of the corresponding load point is calculated to obtain the life uncertainty of each load point; Based on the life uncertainty of each load point and the corresponding average predicted life, the ratio of the life uncertainty to the average predicted life of each load point is calculated, and the load point with the maximum ratio is selected as the test point to be verified; Fatigue test is performed at the stress corresponding to the test point to be verified, and the cycle number of continuous loading to the failure of the connecting rod is recorded as the test result; Based on the test result, the weight parameter of the trained network model is adjusted through the Adam optimization algorithm, and the uncertainty prediction, test point selection, test and model updating process are repeated until a preset termination condition is met, and finally an optimized network model is obtained; Based on the optimized network model, the optimized S-N curve parameter and the life distribution parameter are output.

5. The method of claim 1, wherein, Also includes: Based on the optimized network model, the grid parameters of the preset stress amplitude and average stress are traversed, the corresponding life values under different preset survival rates are calculated and visualized, and the load-life three-dimensional mapping data, the equal life curve and the visualized graphics are generated.

6. The method of claim 1, wherein: The contribution degree distribution of each stress cycle to the total damage is obtained by calculating the contribution proportion of each stress cycle to the total damage.

7. A connecting rod fatigue life assessment device, characterized by, Includes: The model training module is configured to obtain connecting rod fatigue life original data under multiple levels of load, preprocess the data by converting the multiple level load amplitudes into stress parameters and labeling, obtain a structured data set, train a deep neural network model according to the structured data set, and obtain network weight parameters, S-N curve parameters and life distribution parameters, wherein the deep neural network model is updated through the network weight parameters; The model optimization module is configured to predict the life uncertainty of each load point in a preset target load range according to the trained network model, select the load point with the maximum life uncertainty for fatigue test, and update the trained network model according to the test result to obtain optimized S-N curve parameters and life distribution parameters; The life evaluation module is configured to calculate the variable amplitude load life based on real-time load spectrum data, optimized S-N curve parameters and life distribution parameters through a life distribution formula and a damage formula, to obtain real-time fatigue life and damage contribution degree distribution under different preset survival rates, including: The load spectrum data of the connecting rod is acquired in real time, and the load spectrum data includes stress amplitudes, mean stresses and corresponding cycle numbers of a plurality of stress cycles; The single-cycle damage of each stress cycle is calculated by taking the load spectrum data, the optimized S-N curve parameters and the optimized life distribution parameters as inputs and combining a preset mean stress correction function based on a Gerber equation; According to the single-cycle damage of each stress cycle and the corresponding cycle number, a total damage is accumulated by a Miner criterion, and the total damage is a sum of products of the single-cycle damage of each stress cycle and the cycle number; Based on the total damage and the optimized life distribution parameters, the real-time fatigue life under different preset survival rates is predicted according to the life distribution formula. 8.An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1-6 when executing the program.

9. A non-transitory computer-readable storage medium, comprising: In the formula, The non-transitory computer readable storage medium stores computer instructions for causing a computer to execute the method of any one of claims 1-6.

Images