Neural network training method for high-coaxiality special vehicle axle preparation, electronic equipment and readable storage medium

Abstract

The invention discloses a neural network training method for preparing a high-coaxiality special vehicle axle, and the method employs a neural network model to obtain the pre-welding preassembling angle of an axle body and a shaft sleeve, counteracts the coaxiality deviation caused by welding deformation, and directly prepares the high-coaxiality special vehicle axle. Comprising the steps of obtaining a test data training set and a simulation data training set as a mixed training sample, establishing an initial neural network model, training the neural network model, obtaining a pre-assembly angle, guiding actual production, obtaining a high-coaxiality axle and the like. According to the method, welding deformation inevitably existing in the machining process is utilized, a neural network training method is adopted, mixed data serves as a training sample, pre-assembly angle prediction is achieved, then the posture before welding is adjusted, the high-coaxiality axle with the error being only 15-30 threads is directly obtained, the difficulty and cost of obtaining the training sample are reduced, and the accuracy of the high-coaxiality axle is improved. And moreover, accurate prediction of the pre-assembly angle can be realized, measurement of welding deformation quantity is weakened, and production can be accurately guided to obtain a high-coaxiality axle.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence manufacturing, and relates to the preparation of a special vehicle axle with high coaxiality. Specifically, it relates to a neural network training method, an electronic device, and a readable storage medium for the preparation of a special vehicle axle with high coaxiality. Background Art

[0002] An axle is an important part of the vehicle frame chassis, used to support and protect the power device, and buffer the impact force transmitted by the road surface. In the axle assembly, the connection quality between the bushing and the axle body directly affects the quality of the axle assembly. Among them, the coaxiality of the two is a key indicator for evaluating the quality, which not only affects the operation between the gears inside the axle housing, but also affects the distribution of the vehicle load on the bearing capacity and internal stress of the axle housing.

[0003] Compared with the design and optimization of automobile axles, the processing level of the connection between the rear axle body of a special vehicle and the bushing is relatively low. The coaxiality of the assembly of the bushing and the axle body and the coaxiality after welding correction can only reach about 150 μm. Due to the longer and wider external dimensions of special vehicles, greater load-bearing mass, and more severe road conditions during operation, the existing coaxiality accuracy can meet the coaxiality requirements of most light or narrow vehicles, but cannot meet the requirements of large-tonnage or wide-body special vehicles for axle coaxiality.

[0004] Currently, there are mainly two ways to improve the coaxiality of the axle. One is by the way of manual correction after welding, and the other is by the way of reprocessing through machine tool cutting. Both methods perform welding operations on the basis of ensuring a high-precision coaxiality before welding. However, the energy during the welding process is extremely high, and welding deformation will inevitably occur, and the amount of deformation is extremely difficult to accurately control and correct through existing adjustment methods. Therefore, the main body of the axle and the bushing after welding will inevitably produce deformation at the welding surface, resulting in a deviation in the coaxiality between the two again, and re-correction is required. However, on this basis, both correction methods will inevitably affect the overall performance of the axle. Among them, the method of manual correction after welding is to correct the coaxiality error by manually knocking when the weld is still in a high-temperature state. During the process, it is necessary to repeatedly move, knock, and measure the axle housing, which is not only inefficient, unable to accurately control the coaxiality error, but also may introduce stress when knocking the weld, affecting the performance of the axle. Although the method of reprocessing through machine tool cutting can relatively accurately control the coaxiality of the axle housing, defects such as work hardening, stress concentration, and non-compliance with the thickness of the effective penetration layer on the surface of the axle housing are likely to occur after surface processing, affecting the overall performance of the axle housing; and the equipment cost is directly related to the size of the axle housing. The larger the axle housing size, the more expensive the required machine tool equipment. Summary of the Invention

[0005] (I) Technical Problem The technical problem to be solved by the present invention is as follows: Most of the existing coaxiality improvements for special vehicle axles are achieved through mechanical correction after welding. It is extremely difficult to accurately control and correct the welding deformation through the existing adjustment methods. Stress is easily introduced additionally during the correction process, affecting the axle quality. It is difficult to fabricate an axle with high coaxiality, high stability, and high reliability after welding, especially an axle that meets the usage requirements of large-tonnage or wide-body special vehicles.

[0006] (II) Technical Solution The present invention is realized through the following technical solutions: The present invention proposes a neural network training method for the preparation of a special vehicle axle with high coaxiality. By using a neural network model to obtain the pre-assembly angle of the main body of the special vehicle axle and the bushing before welding, it is used to offset the coaxiality deviation caused by welding deformation during actual production, and directly fabricate a special vehicle axle with high coaxiality, including the following steps: S1. Obtain a data training set for neural network model training: When obtaining actual test data, taking the welding starting position and the coaxiality before welding as the main test factors, combining the materials, dimensions of the axle body and the bushing, the shape of the welding surface, and the welding parameters, design a series of single-variable tests, and respectively collect the actual values of the coaxiality after welding of the axle as the test results. Corresponding the test factors and the test results one by one to generate a test data training set for neural network training; When obtaining simulation data, taking the same test factors as those when obtaining actual test data as the input conditions of the welding heat source model, obtain the simulated values of the coaxiality after welding under the corresponding conditions; calculate the difference between the actual value and the simulated value of the coaxiality after welding, and stop correcting the welding heat source model until the error between the actual value and the simulated value is within 5%; based on the corrected welding heat source model, obtain the simulated values of the coaxiality after welding under different simulation test conditions, and generate a simulation data training set for neural network training; S2. Establish an initial neural network model: Taking the test data training set obtained in S1 as the training samples, construct an initial neural network model, and obtain the initial weights and thresholds of the neural network; S3. Train the neural network model to obtain the predicted value of the pre-assembly angle; Taking the test data training set and the simulation data training set obtained in S1 as the training samples together, import them into the initial neural network model. Using the welding process parameters, the coaxiality after welding, and the welding starting position coordinates as the input data, and the pre-assembly angle as the output data, train and optimize the neural network model until convergence, and finally realize the prediction of the pre-assembly angle; S4. Guide actual production to obtain a special vehicle axle with high coaxiality: According to the predicted pre-assembly angle obtained in step S3, adjust the pre-assembly angle between the axle body and the bushing, and perform the welding operation to directly obtain a special vehicle axle with high coaxiality.

[0007] Based on the above technical solution, by adopting the method of neural network training, it is only necessary to collect a limited number of test data for axles of different specifications, combine the simulation model to obtain a hybrid dataset of tests and simulations as the training sample, and use the easily obtainable welding process parameters, welding starting position coordinates, and post-weld coaxiality data to achieve the prediction of the pre-assembly angle that can completely offset the welding deformation amount. Furthermore, before the actual welding process, adjust the pre-assembly angle between the axle body and the bushing to directly obtain a special vehicle axle with high coaxiality without secondary correction. This not only reduces the difficulty and cost of obtaining the training dataset but also improves the accuracy of pre-assembly angle prediction, thereby accurately eliminating the post-weld coaxiality deviation caused by welding deformation.

[0008] Furthermore, it also includes step S5, the continuous optimization of the neural network model: when performing S4, synchronously collect the real data of the pre-assembly angle, welding starting position coordinates, and post-weld coaxiality to generate a production data optimization set. Use the production data optimization set as the training sample to perform online training and optimization on the neural network model to obtain a continuously optimized neural network model. The purpose of this step is to, through the production data optimization set, provide real-time feedback on the changes in equipment states such as fine-tuning of welding parameters and wear of the welding head, and perform online training and optimization on the neural network model so that the neural network model can always accurately predict the pre-assembly angle value that fits the current processing state.

[0009] Even further, before generating the production data optimization set, the neural network model automatically checks whether the post-weld coaxiality obtained based on the predicted pre-assembly angle meets the requirements. When it meets the requirements, intelligently select some actual production data to be included in the production data optimization set; when it does not meet the requirements, all the collected actual production data is included in the production data optimization set for online optimization training of the neural network model, and perform an optimization of the neural network model once. This process can effectively reduce the amount of duplicate data in the production data optimization set on the premise of ensuring real-time feedback of state changes, thereby reducing the data storage requirements and facilitating the long-term stable operation of the algorithm.

[0010] Furthermore, when obtaining the simulated data training set, based on the modified welding heat source model, the simulated values of the post-weld coaxiality obtained by taking the same test factors as the actual test data as the input conditions again; and the simulated values of the post-weld coaxiality obtained by adjusting different variable values. These two parts together constitute the simulated data training set. This process is conducive to obtaining more and more comprehensive training data samples during the neural network training stage and eliminating the prediction result deviation existing when using pure simulated data as the training sample.

[0011] Preferably, in step S1, when obtaining the simulation data, the welding heat source model is corrected by using a neural network model to obtain the corrected boundary conditions. The specific steps are as follows: S101: Use the test data training set generated when obtaining the actual test data as the training samples to establish a neural network model associated with the welding heat source model; S102: Train the neural network model: Use the welding start position, pre-welding coaxiality, and the materials and dimensions of the axle housing body and the bushing, the shape of the welding surface, and the welding process parameters in the training samples as the input data, and use the simulated value of the post-welding coaxiality and the model boundary conditions as the output data. Use the difference between the simulated value and the actual value of the post-welding coaxiality as the control condition to train and optimize the neural network model, and finally obtain the optimal boundary condition value of the welding heat source model; S103: Correct the welding heat source model: Correct the welding heat source model with the optimal boundary condition value obtained in S102 to obtain a welding heat source model with the smallest deviation between the simulated value and the actual value of the post-welding coaxiality; Based on the above steps, the process of correcting the welding heat source model can be simplified, the boundary conditions for model correction can be quickly obtained with less test data, and the simulation accuracy of the corrected model can be improved, with higher processing efficiency.

[0012] Preferably, in step S3, the test data training set and the simulation data training set obtained in step S1 are pre-divided into different category subsets according to the axle specifications. During training, a subset corresponding to a certain specification of the axle is pre-selected as the training sample to train and optimize the neural network model to obtain the predicted value of the pre-assembly angle of the current specification of the axle; when the number of training samples of the neural network model meets the model convergence requirements, replace the subset corresponding to another specification of the axle, and repeat the training and optimization on the existing model until it converges again. Repeat the training of multiple specifications of axles in this way until the neural network model is applicable to the prediction of the pre-assembly angle of any specification of the axle; The above process is conducive to making the neural network model global, applicable to the prediction of the pre-assembly angle of any specification of the axle, and at the same time, the local neural network model can be optimized for a certain specification of the axle to improve the prediction accuracy of the pre-assembly angle of specific objects.

[0013] Furthermore, the method for obtaining the pre-assembly angle for guiding actual production in step S4 is as follows: The intelligent system for the currently processed axle specification retrieves the corresponding subset as the data source for neural network prediction, and finally obtains the pre-assembly angle applicable to the current specification of the axle; the pre-assembly angle, start position coordinates, and post-welding coaxiality data generated during actual production are classified into the corresponding subset for the optimization of the local neural network model of this category, further improving the accuracy of the pre-assembly angle for the processing of specific specification axles, and thus ensuring the accuracy of the post-welding coaxiality.

[0014] The present invention also provides an electronic device, including a processor and a memory. The memory is used to store a computer program. When the processor executes the program stored on the memory, it implements any one of the foregoing neural network training methods for the preparation of a special vehicle axle with high coaxiality.

[0015] The present invention also provides a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are used to: execute any one of the foregoing neural network training methods for the preparation of a special vehicle axle with high coaxiality.

[0016] (III) Beneficial effects At least one technical solution in the present invention has the following advantages or beneficial effects: 1) In the present invention, by using the welding deformation that inevitably exists in the welding process of the axle housing and the bushing, through the method of neural network training, taking the test data and simulation data of a limited number of times as training samples together, predicting the pre-assembly angle that can completely offset the welding deformation amount, and guiding the pre-welding attitude adjustment of the axle housing and the bushing based on this, and then welding and processing, a vehicle axle with high coaxiality can be directly obtained. The coaxiality error of the vehicle axle prepared according to this method can reach only 15 to 30 silk after welding, without the need for post-welding correction, and has high stability and high reliability, and is not limited by the vehicle axle specifications and machine tools, especially suitable for the preparation of vehicle axles of large-tonnage or wide-body special vehicles with high-precision requirements.

[0017] 2) The neural network training method of the present invention only needs to collect test data of a limited number of times for different specifications of vehicle axles, obtain a simulation data set by combining a simulation model, use the mixed data of simulation data and test data as the training samples of the neural network model, and use welding process parameters, welding start position coordinates, post-welding coaxiality and other data that are easy to obtain in real time to achieve accurate prediction of the pre-assembly angle, reduce the difficulty and cost of obtaining training samples, eliminate the prediction result deviation when using pure simulation data as training samples, weaken the measurement requirement of the welding deformation amount, strengthen the accuracy of pre-assembly angle prediction, and is more conducive to accurately guiding production and reducing the product defect rate.

[0018] 3) The neural network training method of the present invention uses a production data optimization set composed of the pre-assembly angle, welding start position coordinates and post-welding coaxiality detected in real time as continuously optimized training samples, and real-time feedbacks the changes in equipment states such as welding parameters and welding head wear to perform online training and optimization on the neural network model, so that the neural network model always fits the current processing state to accurately predict the required pre-assembly angle, and thus always maintains high coaxiality of the vehicle axle after welding.

[0019] 4) The present invention also obtains the corrected boundary conditions of the welding heat source model through a neural network training method, simplifies the process of correcting the welding heat source model, quickly and accurately obtains the boundary conditions with less test data, reduces the manual correction error, improves the simulation accuracy of the corrected model, and has higher processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects, and advantages of the present invention will become more apparent by reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 It is a flowchart of the prediction neural network training for the pre-assembly angle of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The present invention will be further described in detail below with reference to the embodiments, but the embodiments of the present invention are not limited thereto.

[0022] Embodiment 1 This embodiment provides a neural network training method for the preparation of a special vehicle axle with high coaxiality. By using a neural network model, the pre-assembly angle between the main body of the special vehicle axle and the bushing before welding is obtained, which is used to offset the coaxiality deviation caused by welding deformation during actual production, and directly manufacture a special vehicle axle with high coaxiality. The method includes the following steps: S1. Obtain a data training set for neural network model training: When obtaining actual test data, taking the welding starting position and the coaxiality before welding as the main test factors, combining the materials, dimensions of the axle body and the bushing, the shape of the welding surface, and the welding parameters, a series of single-variable tests are designed, and the actual values of the coaxiality after welding of the axle are respectively collected as the test results. The test factors and the test results are corresponded one by one to generate a test data training set for neural network training; It should be noted that: In the actual test design, the materials and dimensions of the axle housing body and the bushing, and the shape of the welding surface are determined according to the axle housing specifications. When the axle housing specifications are not changed, they are relatively fixed parameters; the welding parameters include the process parameters that affect the welding quality, including conventional parameters such as welding power and time. When the axle housing specifications are not changed, they are generally also relatively fixed parameters; the pre-welding coaxiality and the post-welding coaxiality mainly refer to the deflection angle and deflection direction between the axis of the axle housing body and the bushing before or after welding. Generally, they are characterized by the coaxiality error. The smaller the coaxiality error, the higher the coaxiality. The deviation between the pre-welding coaxiality and the post-welding coaxiality is the required pre-assembly angle value. The deflection direction of the pre-assembly angle is opposite to the coaxiality deflection direction before and after welding; the welding starting position mainly affects the superposition of the welding energy at the welding surface, affects the welding deformation direction, and is directly related to the deflection direction of the pre-assembly angle; therefore, the welding starting position and the pre-welding coaxiality are the main test factors, and the materials and dimensions of the axle housing body and the bushing, the shape of the welding surface, and the welding parameters are the auxiliary test factors to reduce the difficulty and cost of obtaining training samples.

[0023] When obtaining the simulation data, use the same test factors as those when obtaining the actual test data as the input conditions of the welding heat source model, and obtain the simulated value of the post-welding coaxiality under the corresponding conditions; calculate the difference between the actual value and the simulated value of the post-welding coaxiality, and stop correcting the welding heat source model until the error between the actual value and the simulated value is within 5%; based on the corrected welding heat source model, obtain the simulated values of the post-welding coaxiality under different simulated test conditions, and generate a simulated data training set for neural network training; Preferably, the simulated data training set includes two parts: One is to use the same test factors as the actual test data as the input conditions again, import the corrected welding heat source model, and re-obtain the simulated value of the post-welding coaxiality; the other is to adjust the welding starting position and the pre-welding coaxiality to values different from the test data, and use this as the input conditions of the corrected welding heat source model to obtain the simulated value of the post-welding coaxiality; this process is used to obtain more and more comprehensive training data samples during the neural network training stage and eliminate the prediction result deviation existing when using pure simulated data as the training sample.

[0024] S2. Establish an initial neural network model: Use the test data training set obtained in S1 as the training sample to construct an initial neural network model, and obtain the initial weights and thresholds of the neural network; S3. Train the neural network model to obtain the predicted value of the pre-assembly angle; Using the test data training set and the simulation data training set obtained in S1 as training samples together, import them into the initial neural network model. Using the welding process parameters, the post-welding coaxiality, and the welding start position coordinates as input data, and the pre-assembly angle as output data, train and optimize the neural network model until convergence, and finally realize the prediction of the pre-assembly angle. S4. Guide actual production to obtain a special vehicle axle with high coaxiality: According to the pre-assembly angle prediction value obtained in step S3, adjust the pre-assembly angle between the axle body and the bushing, and perform the welding operation to directly obtain a special vehicle axle with high coaxiality.

[0025] Furthermore, in S5, continuous optimization of the neural network model: When performing S4, synchronously collect the real data of the pre-assembly angle, the welding start position coordinates, and the post-welding coaxiality to generate a production data optimization set. Using the production data optimization set as the training sample, perform online training and optimization on the neural network model to obtain a continuously optimized neural network model, and perform real-time online training and optimization on the neural network model according to the welding parameter fine-tuning, the wear of the welding head, and other equipment state change amounts, so that the neural network model always fits the pre-assembly angle accurately predicted by the current processing state.

[0026] Even further, before generating the production data optimization set, the neural network model automatically checks whether the post-welding coaxiality obtained based on the predicted pre-assembly angle meets the requirements. When it meets the requirements, intelligently select some actual production data to be included in the production data optimization set; when it does not meet the requirements, all the collected actual production data is included in the production data optimization set for online optimization training of the neural network model, and perform an optimization of the neural network model once. This process can effectively reduce the amount of duplicate data in the production data optimization set on the premise of ensuring real-time feedback of state change amounts, thereby reducing the data storage requirements and facilitating ensuring the long-term operation stability of the algorithm.

[0027] The advantages of this embodiment are as follows: Compared with the existing methods for correcting the coaxiality of axles, by using the welding deformation that inevitably exists in the welding process of the axle body and the bushing, and the phenomenon that the welding deformation will inevitably lead to coaxiality deviation, reversely using the pre-welding and post-welding coaxiality deviation amounts and deviation angles, the coaxiality accuracy requirement is weakened before welding, and the pre-assembly angle between the axle body and the bushing is made to be opposite to the welding deformation before welding to offset the influence of the welding deformation on the coaxiality during the welding process, thereby avoiding secondary processing or correction of the axle after welding, avoiding the introduction of additional stress, and effectively improving the coaxiality, stability, and reliability of the axle.

[0028] Based on the uncertainty of the welding deformation amount and direction, by adopting the neural network training method, it is only necessary to collect a limited number of test data for axles of different specifications, and combine the simulation model to obtain the simulated data set as the mixed data for training and the neural network model, which reduces the difficulty and cost of obtaining training samples, eliminates the prediction result deviation existing when using pure simulated data as training samples, strengthens the accuracy of pre-assembly angle prediction, and is more conducive to accurately guiding production; at the same time, by using the welding process parameters that are easy to obtain in real time, data such as the welding starting position coordinates and the post-welding coaxiality can be used to accurately predict the pre-assembly angle that can completely offset the welding deformation amount, and it is beneficial to the continuous online optimization of the neural network model, so that the neural network model always fits the current processing state, eliminates the prediction deviation caused by inevitable equipment state changes such as the wear of the welding head, further improves the accuracy of the pre-assembly angle, and thus always maintains the preparation of axles with high post-welding coaxiality, especially suitable for the preparation of special vehicle axles with large size specifications and high coaxiality accuracy requirements.

[0029] Embodiment 2 This embodiment provides a neural network training method for the preparation of high coaxiality special vehicle axles, which is different from Embodiment 1 in that: On the basis of the solution in Embodiment 1, when performing step S3, the test data training set and the simulated data training set obtained in step S1 are pre-divided into different category subsets according to the axle specifications. During training, a subset corresponding to a certain specification of the axle is pre-selected as the training sample, and the neural network model is trained and optimized to obtain the predicted value of the pre-assembly angle of the current specification of the axle; when the number of training samples of the neural network model meets the model convergence requirement, replace the subset corresponding to another specification of the axle, and repeat the training and optimization on the existing model until it converges again. Repeat the training of multiple specifications of axles in this way until the neural network model is applicable to the prediction of the pre-assembly angle of any specification of the axle.

[0030] Furthermore, when obtaining the pre-assembly angle for guiding actual production in step S4, the intelligent retrieval of the corresponding subset for the current axle specification to be processed as the data source for neural network prediction, and finally obtain the pre-assembly angle applicable to the current specification of the axle; the pre-assembly angle, starting position coordinates, and post-welding coaxiality data generated during actual production are classified into the corresponding subset for the local neural network model optimization of this category.

[0031] The advantage of this embodiment is that by globally and locally training and optimizing the neural network model, the neural network model has both global and local pertinence, can be applicable to the prediction of the pre-assembly angle of any specification of the axle, and can also be locally optimized for a certain specification of the axle to improve the prediction accuracy of the pre-assembly angle of a specific object, and the applicability of the neural network model is better.

[0032] Embodiment 3 This embodiment provides a neural network training method for the preparation of a special vehicle axle with high coaxiality. The difference between this embodiment and Embodiments 1 and 2 lies in: Based on the solutions of Embodiment 1 or 2, in step S1, when obtaining simulation data, the welding heat source model is corrected by using a neural network model to obtain the corrected boundary conditions. The specific steps are as follows: S101: Use the test data training set generated when obtaining actual test data as training samples to establish a neural network model that is mutually associated with the welding heat source model; S102: Train the neural network model: Use the welding start position, pre-welding coaxiality, and the materials, dimensions of the axle body and bushing, welding surface shape, and welding process parameters in the training samples as input data, use the simulated value of the post-welding coaxiality and the model boundary conditions as output data, and use the difference between the simulated value and the actual value of the post-welding coaxiality as the control condition to train and optimize the neural network model, and finally obtain the optimal boundary condition value of the welding heat source model; S103: Correct the welding heat source model: Correct the welding heat source model with the optimal boundary condition value obtained in S102 to obtain a welding heat source model with the smallest deviation between the simulated value and the actual value of the post-welding coaxiality; The advantage of this embodiment is that the corrected boundary conditions of the welding heat source model are obtained through the neural network training method, which simplifies the process of correcting the welding heat source model, quickly and accurately obtains the boundary conditions with less test data, reduces the manual correction error, improves the simulation accuracy of the corrected model, and has higher processing efficiency.

[0033] Embodiment 4 This embodiment provides an electronic device, including a processor and a memory. The memory is used to store a computer program; the processor is used to implement the neural network training method for the preparation of a special vehicle axle with high coaxiality as described in Embodiments 1 to 3 above when executing the program stored in the memory.

[0034] Embodiment 5 This embodiment provides a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are used to: execute the neural network training method for the preparation of a special vehicle axle with high coaxiality as described in Embodiments 1 to 3 above.

[0035] The computer-readable storage medium of this embodiment can be loaded on any special vehicle axle manufacturing equipment, and can also be used in association with welding models, automatic control systems, etc. to improve the intelligence of axle preparation.

[0036] The above is only a preferred embodiment of the present invention, and does not impose any limitation on the technical scope of the present invention. Therefore, any minor modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the present invention.

Claims

1. A neural network training method for preparing a high coaxiality special vehicle axle, characterized by: A neural network model is used to obtain the pre-assembly angle between the axle body and the sleeve before welding of a special vehicle, which is used to offset the coaxiality deviation caused by welding deformation in actual production and directly produce a high coaxiality special vehicle axle, including the following steps: S1, obtain the data training set for neural network model training: When obtaining actual test data, the welding starting position and the coaxiality before welding are taken as the main test factors. Combined with the materials and dimensions of the axle body and the bushing, the shape of the welding surface, and the welding parameters, a series of single variable tests are designed, and the actual values ​​of the coaxiality of the axle after welding are collected as test results. The test factors and test results are matched one by one to generate a test data training set for neural network training. When obtaining simulation data, the same test factors as those used to obtain actual test data are used as input conditions for the welding heat source model to obtain simulation values ​​of post-weld coaxiality under corresponding conditions; Calculate the difference between the actual and simulated values ​​of the coaxiality after welding, and correct the welding heat source model until the error between the actual and simulated values ​​is within 5%; Based on the modified welding heat source model, the simulated values ​​of post-weld coaxiality under different simulation test conditions are obtained, and a simulated data training set for neural network training is generated; S2, establish the initial neural network model: Using the test data training set obtained by S1 as training samples, construct an initial neural network model, and obtain the initial weights and thresholds of the neural network; S3, training the neural network model to obtain the predicted value of the pre-assembly angle; The experimental data training set and simulation data training set obtained by S1 are used as training samples and imported into the initial neural network model. The welding process parameters, post-welding coaxiality and welding starting position coordinates are used as input data, and the pre-assembly angle is used as output data. The neural network model is trained and optimized until convergence, and finally the pre-assembly angle prediction is realized. S4, guides actual production and obtains high coaxiality special vehicle axles: According to the pre-assembly angle prediction value obtained in step S3, the pre-assembly angle between the axle body and the bushing is adjusted, and the welding operation is performed to directly obtain a special vehicle axle with high coaxiality.

2. A neural network training method for preparing a high coaxiality special vehicle axle according to claim 1, characterized in that: It also includes step S5, continuous optimization of the neural network model: When executing S4, the real data of the pre-assembly angle, welding starting position coordinates and post-weld coaxiality are collected synchronously to generate an optimized production data set. The optimized production data set is used as a training sample to perform online training and optimization on the neural network model to obtain a continuously optimized neural network model.

3. A neural network training method for preparing a high coaxiality special vehicle axle according to claim 2, characterized in that: Before generating the production data optimization set, the neural network model automatically checks whether the post-weld coaxiality obtained based on the predicted pre-assembly angle meets the requirements. When the requirements are met, part of the actual production data is intelligently selected and included in the production data optimization set; when the requirements are not met, all the actual production data collected are included in the production data optimization set for online optimization training of the neural network model, and the optimization of the neural network model is performed once.

4. A neural network training method for preparing a high coaxiality special vehicle axle according to claim 1, characterized in that: When obtaining the simulation data training set, on the basis of the modified welding heat source model, the simulation value of the coaxiality after welding is obtained again with the same test factors as the actual test data as the input conditions; and the simulation value of the coaxiality after welding is obtained by adjusting different variable values. The two parts together constitute the simulation data training set.

5. A neural network training method for preparing a high coaxiality special vehicle axle according to claim 1, characterized in that: In step S1, when obtaining simulation data, the welding heat source model is corrected by using a neural network model to obtain correction boundary conditions. The specific steps are as follows: S101, using the test data training set generated when obtaining actual test data as a training sample, and establishing a neural network model that is interrelated with the welding heat source model; S102, training a neural network model: using the welding starting position, coaxiality before welding, and the materials and dimensions of the axle body and the bushing, the shape of the welding surface, and the welding process parameters in the training sample as input data, using the simulated value of the coaxiality after welding and the model boundary conditions as output data, using the difference between the simulated value and the actual value of the coaxiality after welding as a control condition, training and optimizing the neural network model, and finally obtaining the optimal boundary condition value of the welding heat source model; S103, correcting the welding heat source model: correcting the welding heat source model with the optimal boundary condition value obtained in S102, and obtaining a welding heat source model with the smallest deviation between the simulated value and the actual value of the coaxiality after welding.

6. A neural network training method for preparing a high coaxiality special vehicle axle according to claim 1, characterized in that: In step S3, the test data training set and simulation data training set obtained in step S1 are pre-divided into different category subsets according to the axle specifications. During training, a subset corresponding to a certain specification of axle is pre-selected as a training sample, and the neural network model is trained and optimized to obtain the predicted value of the pre-assembly angle of the current specification of the axle; when the number of training samples of the neural network model meets the model convergence requirements, the subset corresponding to another specification of the axle is replaced, and the training and optimization are repeated on the existing model until convergence again, and the training of multiple specifications of axles is repeated until the neural network model is suitable for the prediction of the pre-assembly angle of axles of any specification.

7. A neural network training method for preparing a high coaxiality special vehicle axle according to claim 6, characterized in that: The method for obtaining the pre-assembly angle to guide actual production in step S4 is as follows: the corresponding subset of the current axle specification to be processed is intelligently retrieved as the data source for neural network prediction, and finally the pre-assembly angle suitable for the current specification axle is obtained; the welding starting pre-assembly angle, starting position coordinates and post-weld coaxiality data generated by actual production are classified into the corresponding subset for optimization of the local neural network model of this category.

8. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory is used to store computer programs; the processor is used to implement the neural network training method for preparing axles of special vehicles with high coaxiality as described in any one of claims 1 to 7 when executing the programs stored in the memory.

9. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to: execute the neural network training method for preparing a high coaxiality special vehicle axle as described in any one of claims 1-7.

Images