A bolt heat treatment process optimization method and system based on deep learning

Abstract

The application discloses a bolt heat treatment process optimization method and system based on deep learning, and relates to the technical field of high-temperature alloy material processing. The bolt heat treatment process optimization method based on deep learning comprises the following steps: obtaining bolt historical heat treatment data, including heat treatment parameters, microstructure images and fatigue life data of a plurality of Inconel718 bolts; constructing a deep learning model, training the deep learning model based on the bolt historical heat treatment data, and outputting optimal heat treatment process parameters of the Inconel718 bolt after the training is completed; and performing heat treatment analysis on the Inconel718 bolt based on the optimal heat treatment process parameters, and verifying the fatigue life of the Inconel718 bolt. The application can quickly predict the fatigue life and screen optimal process parameters by fusing heat treatment parameters and microstructure image features by using the deep learning model, high-efficiency automatic optimization is realized, the development cost is reduced, and the bolt process development efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-temperature alloy material processing, in particular to a bolt heat treatment process optimization method and system based on deep learning. BACKGROUND

[0002] Inconel 718, as a typical precipitation hardening type nickel-based superalloy, is widely used in the fields of aero-engine, gas turbine, etc., and its fatigue life is directly related to the service safety and economy. The traditional heat treatment process (including solution treatment and two-stage aging treatment) usually relies on experience and trial and error in parameter selection, and it is difficult to balance the uniformity of the organization, the control of the delta phase and the precipitation behavior of the strengthening phase.

[0003] In recent years, although some studies have tried to establish the correlation between process, organization and performance, there is still a lack of high-precision and generalizable process parameter prediction model. Therefore, how to construct an efficient and intelligent heat treatment optimization mechanism based on existing process history data, microstructure characterization and fatigue life test data has become a key problem to be solved.

[0004] However, the evaluation of fatigue life in the prior art mainly relies on the traditional sample processing and fatigue test process, that is, under different heat treatment parameter combinations, a plurality of batches of samples are artificially prepared and high-cycle or low-cycle fatigue experiments are carried out, and the fatigue life distribution corresponding to each combination is obtained through statistical analysis. This method has obvious limitations. Firstly, the experimental process itself is highly dependent on the physical test platform, and the test period is long and resource-consuming, which greatly limits the coverage of the parameter space. Secondly, the experimental results of fatigue life are often affected by factors such as fluctuations in material organization details, non-uniformity of cooling conditions, etc., so that the data results have certain non-determinacy and experimental errors, and it is difficult to stably reflect the mapping relationship between the real process and performance. In addition, in the existing process, although rich data resources such as microstructure images and heat treatment process parameters have been collected, these data are often only used for archival storage or qualitative comparison, and have not been effectively structured, extracted and quantitatively modeled, resulting in that the information value has not been fully tapped and the data utilization rate is low. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a bolt heat treatment process optimization method and system based on deep learning, which solves the problem of lack of fatigue life prediction and intelligent optimization of Inconel 718 bolt heat treatment parameters based on historical data in the prior art.

[0006] To achieve the above object, the application is implemented by the following technical solutions: a bolt heat treatment process optimization method based on deep learning, comprising the following steps: obtaining bolt historical heat treatment data, including heat treatment parameters, microstructure images and fatigue life data of a plurality of Inconel718 bolts; constructing a deep learning model and training based on the bolt historical heat treatment data, and outputting the optimal heat treatment process parameters of the Inconel718 bolt after the training is completed; based on the optimal heat treatment process parameters, analyzing the heat treatment of the Inconel718 bolt and verifying its fatigue life.

[0007] Further, the heat treatment parameters include the solution treatment temperature and time of the solution treatment stage, the aging temperature and time, the cooling mode and the cooling rate of the aging treatment stage, the microstructure images include the solution treatment images and the aging treatment images, the solution treatment images include the solution pixel values of a plurality of bolt pixel points, the aging treatment images include the aging pixel values of a plurality of bolt pixel points, and the fatigue life data is specifically the fatigue life cycle number under different loads.

[0008] Further, the deep learning model comprises: a first input channel for receiving bolt heat treatment parameter data and performing standardization processing to generate a heat treatment parameter vector; a second input channel for respectively receiving the solution treatment images and the aging treatment images and performing image feature extraction based on a convolutional neural network to obtain corresponding solution image feature vectors and aging image feature vectors; a fusion layer for performing feature fusion processing on the heat treatment parameter vector and the image feature vectors to construct a joint feature representation; and an output regression layer for outputting a corresponding fatigue life prediction value based on the joint feature representation.

[0009] Further, the specific steps of training the deep learning model based on the bolt historical heat treatment data are as follows: taking the fatigue life cycle number in the bolt historical heat treatment data as a training label; inputting the heat treatment parameters into the first input channel and performing standardization processing; inputting the solution treatment images and the aging treatment images into the convolutional neural network to extract corresponding image feature vectors and input them into the second input channel; performing feature fusion on the inputs of the first input channel and the second input channel in the fusion layer; inputting the fusion result into the output regression layer to predict the corresponding fatigue life value; constructing a loss function based on the mean square error between the predicted value and the true value, and optimizing the network weight parameters through a back propagation algorithm; after the training is completed, the deep learning model can output the fatigue life prediction value corresponding to any input heat treatment parameter combination; repeat the above training process until the model performance meets the preset convergence index or error threshold requirement.

[0010] Further, the optimal heat treatment process parameters include an optimal solution temperature and an optimal solution time of the solution treatment stage, an optimal aging temperature, an optimal aging time, an optimal cooling mode and an optimal cooling rate of the aging treatment stage.

[0011] Further, the specific steps of outputting the optimal heat treatment process parameters of the Inconel718 bolt are as follows: a plurality of groups of candidate heat treatment process parameters are set, each group including a solution temperature and a solution time of the solution treatment stage, an aging temperature, an aging time, a cooling mode and a cooling rate of the aging treatment stage; each group of candidate heat treatment process parameters is input into the first input channel of the trained deep learning model; the standardized solution treatment image and the aging treatment image are input as the input of the second input channel; the fatigue life prediction value output by the model for each group of candidate heat treatment process parameters is obtained, and the corresponding parameter score index is analyzed; the candidate heat treatment process parameter group with the largest parameter score index is selected as the optimal heat treatment process parameter.

[0012] Further, based on the optimal heat treatment process parameters, the specific steps of heat treatment analysis of the Inconel718 bolt are as follows: according to the optimal heat treatment process parameters, the heat treatment process of the Inconel718 bolt is set; the solution treatment and the aging treatment process are sequentially performed, wherein the optimal solution temperature and the optimal solution time are used for the solution treatment, and the optimal aging temperature, the optimal aging time, the optimal cooling mode and the optimal cooling rate are used for the aging treatment; the bolt sample after heat treatment is collected.

[0013] Further, the fatigue life verification includes: performing a fatigue test on the Inconel718 bolt after heat treatment, setting a plurality of different loads; recording the fatigue life cycle number under different loads; comparing and analyzing the measured fatigue life data with the fatigue life prediction value predicted by the deep learning model.

[0014] Further, if the deviation between the verification result and the prediction result exceeds the preset threshold, the verification data is added to the historical heat treatment data as a new sample, and the deep learning model is retrained.

[0015] A bolt heat treatment process optimization system based on deep learning includes: a bolt heat treatment data acquisition unit for acquiring bolt historical heat treatment data, including heat treatment parameters, microstructure images and fatigue life data of a plurality of Inconel718 bolts; a deep learning process optimization unit for constructing a deep learning model and training based on the bolt historical heat treatment data, and outputting the optimal heat treatment process parameters of the Inconel718 bolt after training is completed.

[0016] A bolt fatigue life verification unit is used for heat treatment analysis of Inconel718 bolts based on optimal heat treatment process parameters and verification of fatigue life.

[0017] The present application has the following beneficial effects:

[0018] (1) The bolt heat treatment process optimization method based on deep learning constructs a deep learning model, uses the historical heat treatment parameters, microstructure images and fatigue life data of Inconel718 bolts as joint training data, enables the model to learn the fatigue life law from multi-source information, unlike the traditional method of relying on experience or single parameter fitting, the model can quickly evaluate any heat treatment parameter combination after training, predict the corresponding fatigue life performance, and automatically select the optimal process parameter group accordingly. The entire optimization process has the characteristics of high automation, high efficiency and independence from human experience, and is particularly suitable for small-batch multi-variety development scenarios of high-performance bolt products, effectively shortening the heat treatment process development cycle and reducing the debugging and experimental cost.

[0019] (2) The bolt heat treatment process optimization method based on deep learning, in the heat treatment process, the microstructure changes directly affect the mechanical properties of the bolt, especially the fatigue life performance, by designing double input channels in the model structure, processing the process parameters and the microstructure image information of the solution and aging treatment stage respectively, so as to model from two dimensions of parameter layer and image layer, extract more comprehensive and deep feature expression, after the joint features are constructed in the fusion layer, the prediction result not only reflects the influence of parameter change on fatigue performance, but also considers the microevolution effect of microstructure morphology. This multi-modal information fusion method enables the model to have strong generalization ability and prediction accuracy when facing different batches of raw materials or complex heat treatment combinations.

[0020] (3) The bolt heat treatment process optimization method based on deep learning, after the heat treatment optimization is completed, a fatigue life verification and data feedback mechanism is designed, the bolt sample is heat treated by using the selected optimal process parameters, and fatigue test is carried out under multiple load conditions, the corresponding actual fatigue life data can be obtained, if there is a significant deviation between the measured result and the model prediction, the data of this batch can be returned to the model training set for incremental training, this process constitutes a complete closed loop from data training, model prediction to result verification and regression optimization, ensures that the model can continuously adapt to the changes of actual production conditions, realizes the evolution from one-time inference model to sustainable self-optimizing model.

[0021] (4), the bolt heat treatment process optimization system based on deep learning, through the construction of integrated system, is divided into heat treatment data acquisition unit, deep learning process optimization unit and fatigue life verification unit and other modules, make the whole Inconel718 bolt heat treatment optimization process has clear logical structure and function boundary, each unit has independent data input, processing and output mechanism, when system deployment can be needed for module level expansion or function iteration, improve the overall project adaptability and industrial landing ability, at the same time, the system design can be compatible with different data source format and hardware acquisition mode, realize the standardization of heat treatment data acquisition and management, reduce the deployment cost and debugging difficulty in actual engineering application, through the construction of systematic integrated scheme, make the whole process from data acquisition, model optimization to performance verification has the ability of reusability, migration, significantly improve the intelligent optimization efficiency of bolt parts in different application scenarios, provide the popularization of basic framework for intelligent manufacturing and digital process management.

[0022] Of course, it is not necessary to achieve all the advantages described above to implement any product of the application. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A flow chart of a bolt heat treatment process optimization method based on deep learning according to the present application.

[0024] Figure 2 A flow chart of the specific steps of outputting the optimal heat treatment process parameters of Inconel718 bolts in a bolt heat treatment process optimization method based on deep learning according to the present application.

[0025] Figure 3 A block diagram of a bolt heat treatment process optimization system based on deep learning according to the present application. DETAILED DESCRIPTION

[0026] Please refer to Figure 1 The embodiment of the present application provides a technical scheme: a bolt heat treatment process optimization method based on deep learning, comprising the following steps: acquiring bolt historical heat treatment data, including the heat treatment parameters, microstructure images and fatigue life data of a plurality of Inconel718 bolts; constructing a deep learning model and training based on the bolt historical heat treatment data; outputting the optimal heat treatment process parameters of Inconel718 bolts after training; based on the optimal heat treatment process parameters, analyzing the heat treatment of Inconel718 bolts and verifying the fatigue life.

[0027] Specifically, the heat treatment parameters include a solid solution temperature and a solid solution time of a solid solution treatment stage, an aging temperature, an aging time, a cooling mode and a cooling rate of an aging treatment stage, the microstructure images include a solid solution treatment image (after the solid solution treatment stage ends) and an aging treatment image (after the aging treatment stage ends), the solid solution treatment image includes solid solution pixel values of the plurality of bolt pixel points, the aging treatment image includes aging pixel values of the plurality of bolt pixel points, and the fatigue life data is specifically a fatigue life cycle number under different loads.

[0028] The deep learning model includes: a first input channel configured to receive bolt heat treatment parameter data and perform standardization processing on the bolt heat treatment parameter data to generate a heat treatment parameter vector; a second input channel configured to respectively receive a solid solution treatment image and an aging treatment image, and perform image feature extraction based on a convolutional neural network to obtain a corresponding solid solution image feature vector and an aging image feature vector; a fusion layer configured to perform feature fusion processing on the heat treatment parameter vector and the image feature vector to construct a joint feature representation; and an output regression layer configured to output a corresponding fatigue life prediction value based on the joint feature representation.

[0029] The specific steps of training the deep learning model based on the bolt historical heat treatment data are as follows:

[0030] The fatigue life cycle number in the bolt historical heat treatment data is used as a training label, which is specifically: extracting a plurality of fatigue life cycle number data of Inconel718 bolts under different heat treatment process parameters from historical test data, and taking the fatigue life cycle number as a real label value used in supervised learning. If there are fatigue life data under multiple load conditions, the weighted fatigue life value of each bolt sample is calculated by setting the weighting coefficients of different load intensities, and is uniformly taken as a single target variable for subsequent model training target output.

[0031] The heat treatment parameters are input into the first input channel and are subjected to standardization processing, which is specifically: inputting the heat treatment parameters (including solid solution temperature, solid solution time, aging temperature, aging time, cooling mode and cooling rate) corresponding to each bolt sample into the first input channel of the model. In order to reduce the numerical difference interference between different dimensions, all parameters are subjected to standardization processing (such as Z-score standardization or Min-Max normalization) first, so that all parameter values are compressed into a relatively unified numerical range, and the stability of gradient propagation and the convergence speed of the neural network during training are ensured;

[0032] The solid solution treatment image and the aging treatment image are respectively input into a convolutional neural network to extract corresponding image feature vectors and input into a second input channel, which is specifically: the solid solution treatment stage image and the aging treatment stage image after image preprocessing (such as size normalization, graying, etc.) are respectively input into two parallel convolutional neural network (CNN) paths. Each path includes multiple convolutional layers, pooling layers and batch normalization layers for extracting key organizational image features including grain size, phase boundary clarity, precipitated phase distribution density, and organizational uniformity. After the flatten or fully connected operation, the image features are converted into fixed-length vectors. The output vectors of the two image paths are spliced or fused as the final image feature input of the second input channel;

[0033] In the fusion layer, the inputs of the first input channel and the second input channel are fused, which is specifically: the normalized heat treatment parameter vector from the first input channel and the image feature vector of the second input channel are input into the fusion layer. The fusion layer can use vector splicing, weighted summation or attention mechanism for multi-modal feature fusion, thereby constructing a joint feature representation containing both process parameters and organizational morphology. The joint feature can comprehensively reflect the nonlinear relationship between heat treatment process and microstructure;

[0034] The fusion result is input into the output regression layer to predict the corresponding fatigue life value, which is specifically: the fused joint feature vector is input into the fully connected regression layer to obtain the fatigue life prediction value corresponding to the input sample through nonlinear mapping. This value represents the fatigue life cycle number that the Inconel718 bolt can reach under the current process parameters and organizational conditions;

[0035] A loss function is constructed based on the mean square error between the predicted value and the true value, and the network weight parameters are optimized through a back propagation algorithm, which is specifically: the loss function is set as the weighted mean square error (MSE) between the predicted fatigue life value and the true fatigue life value, and a sample weight factor is added to reflect the importance or data confidence of the sample. Samples with larger or more difficult to fit residuals can be given higher weights to enhance the model's learning ability for abnormal samples. Gradient update is performed using a back propagation algorithm (such as Adam or SGD) to gradually optimize the parameters of each layer in the network and improve the model's accuracy and robustness;

[0036] After training is completed, the deep learning model can output the fatigue life prediction value corresponding to any input heat treatment parameter combination, which is specifically: the trained model has generalization ability and can infer and predict the input of any new heat treatment parameter combination (even if it is not in the training set) to output the corresponding fatigue life prediction value. This function can be used to explore the parameter space and assist in selecting the optimal process parameter combination;

[0037] The above training process is repeated until the model performance meets the preset convergence indicators or error threshold requirements, which are specifically: after each round of training is completed, the model is evaluated using the validation set or cross-validation set. If the evaluation indicators (such as validation loss, prediction accuracy or error distribution) have reached the set convergence criteria (such as validation error less than a certain threshold, or no significant improvement for a certain number of consecutive rounds), the model training is considered complete, and the current optimal model weight is retained.

[0038] wherein the loss function is specifically as follows: ; wherein, is the loss function, , , are the true fatigue life value (which is the weighted result of the fatigue life cycle times under different loads), the model predicted fatigue life value, and the sample weight of the i-th sample, respectively, i = 1, 2, 3, …, , , is the sample number.

[0039] And the sample weight is obtained by the residual algorithm: .

[0040] In the embodiment, by introducing a multi-modal deep learning modeling method, the heat treatment parameters and the organizational image information are fused and modeled, and a multi-load fatigue life label and a sample weighting strategy are introduced, which can effectively improve the prediction accuracy and stability of the fatigue life of Inconel718 bolts. Specifically, by constructing a neural network model with dual input channels, on the one hand, the numerical regularity in the heat treatment parameters is fully retained, and on the other hand, the microstructure features (such as grain size, phase boundary density, and precipitate phase morphology) in the solution treatment image and the aging treatment image are deeply extracted. Then, through the feature fusion layer, the two types of information are integrated to realize accurate regression prediction of the fatigue life. Compared with the traditional modeling method based on only heat treatment parameters or single image information, this method effectively bridges the nonlinear coupling relationship between process and microstructure. At the same time, by introducing the load weighting label and the residual driven sample weight mechanism, the model pays more attention to samples that are difficult to predict, improving the overall robustness and generalization ability.

[0041] Specifically, the optimal heat treatment process parameters include the optimal solution temperature and the optimal solution time in the solution treatment stage, the optimal aging temperature, the optimal aging time, the optimal cooling mode, and the optimal cooling rate in the aging treatment stage.

[0042] As Figure 2As shown, the specific steps of outputting the optimal heat treatment process parameters of the Inconel718 bolt are as follows: setting multiple groups of candidate heat treatment process parameter groups of the Inconel718 bolt, each group including the solution temperature and solution time of the solution treatment stage, the aging temperature, aging time, cooling mode and cooling rate of the aging treatment stage; inputting each group of candidate heat treatment process parameter groups of the Inconel718 bolt into the first input channel of the deep learning model trained; inputting the normalized solution treatment image and aging treatment image as the input of the second input channel to maintain the consistency of image input; obtaining the fatigue life prediction value output by the model for each group of candidate heat treatment process parameters, and analyzing the corresponding parameter score index; screening the candidate heat treatment process parameter group with the largest parameter score index as the optimal heat treatment process parameter of the Inconel718 bolt.

[0043] The specific steps of setting multiple groups of candidate heat treatment process parameter groups are as follows:

[0044] Firstly, the definition dimension and value range of the candidate parameters are set. The heat treatment parameters include the solution temperature and solution time of the solution treatment stage, the aging temperature, aging time, cooling mode and cooling rate of the aging treatment stage. The value range of each parameter can be set according to historical heat treatment process data, existing process specifications, material process standards and equipment capacity. For example, the solution temperature can be set to 980-1080℃ with a step of 20℃; the solution time can be set to 20-80min with a step of 10min; the cooling mode can include water cooling, oil cooling and air cooling; the cooling rate is set to 5-50℃ / min.

[0045] Secondly, multiple groups of candidate parameters are generated based on the above parameter range. Grid search method can be used, uniform sampling strategy is used for each continuous parameter, and discrete parameters such as cooling mode are combined to form multiple legal parameter combination sets; Latin hypercube sampling, Monte Carlo sampling or genetic algorithm method can also be used to reduce redundancy and improve combination efficiency while ensuring parameter space coverage.

[0046] Then, each group of candidate parameter groups generated is subjected to structure unification processing, and each group of parameter combinations is constructed into a six-dimensional structure vector in the format of [solution temperature, solution time, aging temperature, aging time, cooling mode, cooling rate]. The cooling mode is processed in the form of coding, for example, air cooling = 0, oil cooling = 1, water cooling = 2, to ensure that the model can receive and recognize such input.

[0047] Then, the candidate parameter group vectors are standardized. For continuous variables, z-score standardization or min-max normalization is used to convert different scale physical parameters to a uniform scale, improving the compatibility and prediction stability of the model for different dimensional features.

[0048] Subsequently, all candidate parameter groups are checked for physical reasonableness. If there is an obviously conflicting combination (such as the aging temperature being higher than the solution temperature, the time value exceeding the upper limit of the equipment, etc.), the parameter group is removed to ensure that the input model data has engineering feasibility.

[0049] Finally, all candidate heat treatment parameter groups that pass the screening and processing form the parameter set to be evaluated, which is used as the input data for the model to predict fatigue life.

[0050] The specific steps for analyzing the parameter score index of each candidate heat treatment process parameter are as follows: based on the preset scoring rules, a cooling energy consumption score is assigned to each cooling method in each candidate heat treatment process parameter; based on the fatigue life prediction value, the solution temperature, the aging temperature, and the cooling energy consumption score, a comprehensive analysis is performed to obtain the parameter score index of each candidate heat treatment process parameter.

[0051] The specific formula for calculating the parameter score index of a certain candidate heat treatment process parameter is as follows: ; wherein, , , , , The parameter score index, the fatigue life prediction value, the solution temperature, the aging temperature, and the cooling energy consumption score of a certain candidate heat treatment process parameter are in turn, , , The life prediction weight, the temperature difference weight, and the cooling weight stored in the database are in turn 0.8, 0.1, and 0.3 in this embodiment.

[0052] In this embodiment, by introducing multiple sets of candidate heat treatment process parameters and combining a deep learning model for scoring and screening, the scientificity and practicality of heat treatment process optimization are significantly improved. Compared with the traditional trial-and-error method relying on experience, this method can generate a large number of parameter combinations based on reasonable parameter ranges, and use the trained neural network model to predict the fatigue life corresponding to each parameter combination, greatly improving the efficiency and accuracy of parameter optimization. At the same time, the comprehensive scoring function considers the fatigue life prediction value, as well as key factors such as solution temperature, aging temperature, and cooling energy consumption, introduces adjustable weight coefficients, and realizes flexible process selection under different optimization objectives (such as performance priority or energy consumption priority). In addition, through standardization processing and physical rationality verification, the uniformity and engineering feasibility of the input parameters are ensured, so that the final selected optimal heat treatment process parameters have strong practicality and reliability.

[0053] Specifically, based on the optimal heat treatment process parameters, the specific steps of heat treatment analysis of Inconel718 bolts are as follows: according to the optimal heat treatment process parameters, the heat treatment process of Inconel718 bolts is set; the solution treatment and aging treatment processes are executed in turn, wherein the optimal solution temperature and the optimal solution time are used for solution treatment, and the optimal aging temperature, the optimal aging time, the optimal cooling mode and the optimal cooling rate are used for aging treatment; the bolt samples after heat treatment are collected for subsequent fatigue life verification.

[0054] In this embodiment, by executing the actual heat treatment process of Inconel718 bolts based on the optimal heat treatment process parameters, the engineering feasibility and operability of the model recommendation results are ensured. By specifying the optimal temperature, time, cooling mode and cooling rate required for solution treatment and aging treatment, the production process can be standardized, and the test consistency and repeatability can be improved. At the same time, the acquisition of bolt samples after heat treatment provides key support for the actual measurement and verification of fatigue life, which helps to evaluate the accuracy of the model prediction results and the effectiveness of the optimization scheme, and further improves the closed-loop feedback capability and practical application value of the overall process optimization system.

[0055] Specifically, fatigue life verification includes: performing fatigue test on the Inconel718 bolts after heat treatment, setting multiple different loads; recording the fatigue life cycle number under different loads; comparing and analyzing the measured fatigue life data (the weighted result of the fatigue life cycle number under different loads) with the fatigue life prediction value predicted by the deep learning model to verify the reliability and effectiveness of the optimal heat treatment process parameters.

[0056] If the deviation between the verification result and the prediction result exceeds the preset threshold, the verification data is added to the historical heat treatment data as a new sample, and the deep learning model is retrained to realize iterative optimization of the heat treatment process parameters.

[0057] In the embodiment, the comparison between the measured fatigue life and the model prediction result verifies the mechanism, which provides a reliable closed-loop evaluation method for the prediction performance of the deep learning model. On the one hand, fatigue tests are carried out under different loads, and the corresponding cycle times are recorded, which can comprehensively reflect the durability performance of Inconel718 bolts under actual service conditions. On the other hand, the weighted results of the measured fatigue life are accurately compared with the model prediction values, which not only helps to confirm the practical effectiveness of the optimal heat treatment process parameters, but also identifies potential error sources of the model. When the deviation exceeds the threshold value, the system supports feeding back new data for model retraining, realizing iterative optimization under data-driven, thereby continuously improving the model robustness and prediction accuracy, and constructing a dynamic and evolving heat treatment process optimization system.

[0058] Please refer to Figure 3 The embodiment of the present application provides a technical scheme: a bolt heat treatment process optimization system based on deep learning, comprising: a bolt heat treatment data acquisition unit, configured to acquire bolt historical heat treatment data, including heat treatment parameters, microstructure images and fatigue life data of a plurality of Inconel718 bolts; a deep learning process optimization unit, configured to construct a deep learning model and train based on the bolt historical heat treatment data, and output optimal heat treatment process parameters of the Inconel718 bolt after training; and a bolt fatigue life verification unit, configured to analyze the heat treatment of the Inconel718 bolt based on the optimal heat treatment process parameters, and verify the fatigue life thereof.

[0059] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to these embodiments once they know the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including all changes and modifications falling within the scope of the present application.

[0060] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for optimizing bolt heat treatment process based on deep learning, characterized in that, Includes the following steps: Acquire historical heat treatment data for bolts, including heat treatment parameters, microstructure images, and fatigue life data for several Inconel 718 bolts. Heat treatment parameters include solution temperature and solution time in the solution treatment stage, aging temperature, aging time, cooling method, and cooling rate in the aging treatment stage. Microstructure images include solution treatment images and aging treatment images. Solution treatment images include solution pixel values ​​for several bolt pixels, and aging treatment images include aging pixel values ​​for several bolt pixels. Fatigue life data specifically refers to the number of fatigue life cycles under different loads. A deep learning model is constructed and trained based on historical heat treatment data of bolts. After training, the optimal heat treatment process parameters for Inconel718 bolts are output. Based on the optimal heat treatment process parameters, the heat treatment of Inconel718 bolts was analyzed, and its fatigue life was verified. Deep learning models include: The first input channel is used to receive bolt heat treatment parameter data and standardize it to generate a heat treatment parameter vector. The second input channel is used to receive the solid solution processed image and the time-effect processed image respectively, and to extract image features based on the convolutional neural network to obtain the corresponding solid solution image feature vector and time-effect image feature vector. The fusion layer is used to perform feature fusion processing on the heat treatment parameter vector and the image feature vector to construct a joint feature representation; The output regression layer is used to output the corresponding fatigue life prediction value based on the joint feature representation. The specific steps for training a deep learning model based on historical heat treatment data of bolts are as follows: The fatigue life cycle count in the historical heat treatment data of bolts was used as the training label. The heat treatment parameters are input into the first input channel and then standardized. The solid solution processed image and the time-effect processed image are respectively input into a convolutional neural network to extract the corresponding image feature vectors, which are then input into the second input channel; In the fusion layer, features are fused between the inputs of the first input channel and the second input channel; The fusion results are input into the output regression layer to predict the corresponding fatigue life value; A loss function is constructed based on the mean squared error between the predicted and the true values, and the network weight parameters are optimized through the backpropagation algorithm. After training, the deep learning model can output the fatigue life prediction value corresponding to any combination of input heat treatment parameters. Repeat the above training process until the model performance meets the preset convergence metric or error threshold requirements.

2. The bolt heat treatment process optimization method based on deep learning according to claim 1, characterized in that, The optimal heat treatment process parameters include the optimal solution temperature and optimal solution time in the solution treatment stage, the optimal aging temperature, optimal aging time, optimal cooling method, and optimal cooling rate in the aging treatment stage.

3. The bolt heat treatment process optimization method based on deep learning according to claim 2, characterized in that, The specific steps for outputting the optimal heat treatment process parameters for Inconel 718 bolts are as follows: Multiple sets of candidate heat treatment process parameters are set up. Each set includes the solution temperature and solution time in the solution treatment stage, and the aging temperature, aging time, cooling method and cooling rate in the aging treatment stage. Each set of candidate heat treatment process parameters is input into the first input channel of the trained deep learning model; The standardized solid solution processed image and the time-treated image are used as inputs to the second input channel; Obtain the fatigue life prediction values ​​output by the model for each group of candidate heat treatment process parameters, and analyze the corresponding parameter scoring index; The candidate heat treatment process parameter group with the highest parameter score index is selected as the optimal heat treatment process parameter.

4. The bolt heat treatment process optimization method based on deep learning according to claim 2, characterized in that, The specific steps for analyzing the heat treatment of Inconel 718 bolts based on optimal heat treatment process parameters are as follows: Based on the optimal heat treatment process parameters, the heat treatment process for Inconel 718 bolts was set. The solution treatment and aging treatment processes are performed sequentially. The solution treatment uses the optimal solution temperature and optimal solution time, while the aging treatment uses the optimal aging temperature, optimal aging time, optimal cooling method, and optimal cooling rate. Bolt samples were collected after heat treatment.

5. The bolt heat treatment process optimization method based on deep learning according to claim 4, characterized in that, Fatigue life verification includes: Fatigue tests were conducted on heat-treated Inconel 718 bolts under multiple different loads. Record the number of fatigue life cycles under different loads; The measured fatigue life data were compared and analyzed with the fatigue life prediction values ​​predicted by the deep learning model.

6. The bolt heat treatment process optimization method based on deep learning according to claim 5, characterized in that, If the deviation between the verification result and the prediction result exceeds a preset threshold, the verification data will be added as a new sample to the historical heat-processed data to retrain the deep learning model.

7. A deep learning-based bolt heat treatment process optimization system, employing the deep learning-based bolt heat treatment process optimization method according to any one of claims 1-6, characterized in that, include: The bolt heat treatment data acquisition unit is used to acquire historical heat treatment data of bolts, including heat treatment parameters, microstructure images, and fatigue life data of several Inconel 718 bolts. The deep learning process optimization unit is used to build a deep learning model and train it based on historical heat treatment data of bolts. After training, it outputs the optimal heat treatment process parameters for Inconel718 bolts. The bolt fatigue life verification unit is used to perform heat treatment analysis on Inconel718 bolts based on optimal heat treatment process parameters and to verify their fatigue life.

Images