A method for predicting the performance of finished automobile radiators based on machine learning

Abstract

The present invention belongs to the technical field of automobile radiator production process, and specifically relates to a method for predicting the performance of automobile radiator finished products based on machine learning. First, the radiator manufacturing data is obtained, including material composition, process parameters, environmental conditions and performance test data. Secondly, the meta-learning algorithm is used to construct the model for preliminary training, including task division, adaptive learning rate adjustment, Hessian matrix optimization and other steps. Then, physical knowledge constraints are introduced, and the heat conduction and heat convection equations are integrated into the loss function in a soft target manner. Finally, the model is evaluated by an incremental verification method, including continuous monitoring and collection of new data, batch incremental evaluation and other operations. The present invention solves the problems of poor adaptability of existing methods, lack of combination of physical knowledge, long-term performance degradation, etc., improves the prediction accuracy and model adaptability, and ensures that the model output conforms to physical laws.

Description

Technical Field

[0001] The present invention belongs to the technical field of automobile radiator production technology, and in particular relates to a method for predicting the performance of an automobile radiator finished product based on machine learning. Background Art

[0002] In the process of continuous progress of the automotive industry, the performance of automotive radiators, as key components of the vehicle cooling system, has a vital impact on the normal operation and overall performance of the vehicle. The existing evaluation of radiator performance mainly relies on a large number of experimental tests. This process not only takes a lot of time, but also requires a lot of human and material resources. With the continuous development of machine learning technology, its application in various fields is becoming more and more extensive, bringing new ideas and methods for radiator performance prediction. However, the existing machine learning-based methods have some shortcomings. On the one hand, the model has poor adaptability when facing new radiator designs, and cannot quickly and effectively predict the performance of radiators of different designs. On the other hand, these methods often do not fully combine physical knowledge, resulting in the output of the model may not conform to the laws of physics, thus affecting the accuracy of the prediction. Summary of the invention

[0003] In view of the technical problems existing in the above-mentioned background technology, the present invention proposes a method for predicting the performance of automobile radiator products based on machine learning, aiming to improve the accuracy of prediction and the adaptability of the model, while ensuring that the model output conforms to the laws of physics.

[0004] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: comprising the following steps: S1. First, obtain the manufacturing data of multiple sample radiators, including material composition, manufacturing process parameters, environmental conditions and performance test data; S2. Secondly, use the meta-learning algorithm to build a model for preliminary training, so that the model can quickly adapt to the new radiator design and optimize the parameters to minimize the loss function; S3. Then introduce physical knowledge constraints, and integrate the physical constraints into the loss function in the form of soft targets, rather than introducing complete physical residuals, to ensure that the model output conforms to the laws of physics; S4. Finally, evaluate the model and use an incremental verification method to verify the model's predictive performance to ensure the accuracy of the model.

[0005] The preliminary training using the meta-learning algorithm in step S2 includes the following steps: S21, firstly, task division, the radiator data set is divided into multiple subtasks, and the number of samples for each task is kept consistent; S22, secondly, the adaptive learning rate is adjusted, and the adaptive learning rate algorithm is used to enable the updated learning rate of each parameter to be dynamically adjusted according to the historical gradient information. The specific update formula is: ,in is the sum of the squares of the historical gradients of each parameter, is the gradient of the i-th iteration; S23, then use the Hessian matrix To optimize the update, the second-order derivative information is introduced to improve the parameter update, thereby enhancing the stability of the model update. The parameter update formula is ,in are the current model parameters, is the updated parameter, is the first-order gradient, that is, the derivative of the loss function with respect to the parameters, It is the inverse of the matrix, which is used to adjust the update step size and increase the consideration of second-order information; S24, then design the weighted loss function , where the weighting factor Adjust according to the complexity of the task, and the calculation formula is: , is the total loss function, combining the losses of all tasks, K is the total number of tasks, is the loss function of the ith task, is a small constant to prevent division by zero; S25. Finally, an integrated learning method is used to combine the outputs of different models to improve the final prediction performance.

[0006] Preferably, the manufacturing data of the radiator in step S1 includes: material composition including the type of metal material of the radiator and the proportion of each component, manufacturing process parameters including welding temperature, welding time, processing accuracy, stamping pressure, environmental conditions including temperature and humidity of the production environment, and performance data including heat dissipation efficiency and fluid flow rate.

[0007] Preferably, the physical knowledge constraint introduced in step S3 is introduced by introducing simplified heat conduction equation and heat convection equation as constraints, the formula is: ,in is the data-driven total loss function, is a weight coefficient used to balance the impact of data loss and physical constraint loss, Represents the rate of change of temperature over time, that is, the temperature of the i-th sample The rate of change with time t, is the Laplace term in the heat conduction equation, is the divergence of the fluid velocity.

[0008] Preferably, the incremental verification in step S4 is implemented by first continuously monitoring and collecting newly generated radiator performance data, performing batch incremental evaluation, dividing the new data into small batches, evaluating the impact of each batch on the model prediction performance in turn, and setting up real-time performance monitoring to track the prediction accuracy and physical consistency of the model during the incremental verification process. When the performance degrades, an alarm is automatically triggered and the indicators are adjusted.

[0009] Compared with the prior art, the advantages and positive effects of the present invention are that the model can quickly adapt to new designs by using a meta-learning algorithm, task division, adaptive learning rate algorithm, and Hessian matrix second-order optimization. Heat conduction and convection equations are introduced as soft targets into the loss function to avoid the results from deviating from physical common sense, improve accuracy, and ensure consistency with the actual environment. Incremental verification is adopted, new data is continuously monitored and collected, batch incremental evaluation is performed, real-time monitoring is set, and automatic alarms and indicators are adjusted when performance degrades to ensure high prediction accuracy after new data is added. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 This is a performance comparison chart of the heat dissipation efficiency experimental results and the predicted results;

[0012] Figure 2 This is a performance comparison chart of the fluid flow rate experimental results and the predicted results. DETAILED DESCRIPTION

[0013] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0014] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0015] Embodiment, in the automobile industry, the radiator is a core component of the vehicle cooling system, and its performance directly affects the efficiency and safety of the engine. The existing radiator performance evaluation relies on a large number of experimental tests, which is time-consuming, labor-intensive, and costly. Based on this background, the present invention proposes a new method for predicting the performance of automobile radiator finished products. First, in the data collection part, the present invention takes into account the need to collect data including material composition, manufacturing process parameters, environmental conditions and performance test data. The material composition includes the type of metal material of the radiator and the proportion of each component, the manufacturing process parameters include welding temperature, welding time, processing accuracy, stamping pressure, the environmental conditions include the temperature and humidity of the production environment, and the performance data include heat dissipation efficiency and fluid flow rate.

[0016] Secondly, in order to improve the adaptability and generalization ability of the model in dealing with different types of radiators, a task division scheme is adopted to divide the data set into multiple subtasks, and an adaptive learning rate algorithm and Hessian matrix are used for second-order optimization, which significantly enhances the stability and prediction performance of the model. First, task division is performed. The radiator data set is divided into multiple subtasks, and the number of samples for each task remains consistent; secondly, the adaptive learning rate is adjusted. The adaptive learning rate algorithm is used to enable the update learning rate of each parameter to be dynamically adjusted according to the historical gradient information. The specific update formula is: ,in is the sum of the squares of the historical gradients of each parameter, is the gradient of the i-th iteration; then use the Hessian matrix To optimize the update, the second-order derivative information is introduced to improve the parameter update, thereby enhancing the stability of the model update. The parameter update formula is ,in are the current model parameters, is the updated parameter, is the first-order gradient, that is, the derivative of the loss function with respect to the parameters, It is the inverse of the matrix, which is used to adjust the update step size and increase the consideration of second-order information; then design the weighted loss function , where the weighting factor Adjust according to the complexity of the task, and the calculation formula is: , is the total loss function, combining the losses of all tasks, K is the total number of tasks, is the loss function of the ith task, is a small constant to prevent division by zero; finally, an integrated learning method is used to combine the outputs of different models.

[0017] In order to ensure that the prediction results of the model conform to the laws of physics, a physical knowledge constraint scheme is adopted to integrate the heat conduction equation and the heat convection equation into the loss function as soft targets. This constraint avoids the problem that the model relies solely on data-driven prediction results that deviate from physical common sense, improves the rationality and accuracy of the prediction, and ensures that the radiator performance prediction is consistent with the actual working environment. The introduction of physical knowledge constraints is to introduce simplified heat conduction equations and heat convection equations as constraints, and the formula is ,in is the data-driven total loss function, is a weight coefficient used to balance the impact of data loss and physical constraint loss, Represents the rate of change of temperature over time, that is, the temperature of the i-th sample The rate of change with time t, is the Laplace term in the heat conduction equation, is the divergence of the fluid velocity.

[0018] After training is completed, in order to ensure that the model can maintain high prediction accuracy after new data is added, an incremental verification scheme is adopted to continuously monitor and collect new radiator performance data to ensure that the model can adapt to data changes at any time and avoid the problem of performance degradation of the model during long-term use. First, continuously monitor and collect newly generated radiator performance data, perform batch incremental evaluation, divide the new data into small batches, evaluate the impact of each batch on the model's prediction performance in turn, and set up real-time performance monitoring to track the model's prediction accuracy and physical consistency during the incremental verification process. When performance degrades, alarms are automatically triggered and indicators are adjusted.

[0019] Next, the present invention provides several sets of performance comparisons between experimental results and prediction results to verify the feasibility of the present invention, as follows Figure 1 Figure 2 As shown in the two figures, it can be seen that the predicted results are close to the true values, and the errors are within the acceptable range for both heat dissipation efficiency and fluid flow rate indicators. By comparing these experimental results with the predicted results, the good effects achieved by the present invention in improving prediction accuracy, enhancing model adaptability, and ensuring that the model output complies with physical laws are fully verified, and further proves that the method proposed by the present invention has high practical value and application prospects in the field of performance prediction of finished automobile radiators.

[0020] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting the performance of finished automobile radiators based on machine learning, characterized in that: The following steps are involved: S1. First, obtain the manufacturing data of multiple sample heat sinks, including material composition, manufacturing process parameters, environmental conditions and performance test data; S2. Secondly, a meta-learning algorithm is used to build a model for preliminary training, so that the model can quickly adapt to the new heat sink design and optimize the parameters to minimize the loss function; S3, then introduce physical knowledge constraints, and integrate the physical constraints into the loss function in the form of soft targets, rather than introducing complete physical residuals, to ensure that the model output conforms to physical laws. The physical knowledge constraints introduced in step S3 are introduced by introducing simplified heat conduction equations and heat convection equations as constraints. The formula is: where Γ total1 (θ) is the data-driven total loss function, λ is the weight coefficient used to balance the impact of data loss and physical constraint loss, Represents the rate of change of temperature over time, that is, the temperature T of the i-th sample i The rate of change over time t, α▽ 2 T i is the Laplace term in the heat conduction equation, ▽·u i is the divergence of the fluid velocity; S4. Finally, the model is evaluated and the prediction performance of the model is verified by incremental verification to ensure the accuracy of the model. The preliminary training using the meta-learning algorithm in step S2 includes the following steps: S21. First, the task is divided. The radiator dataset is divided into multiple subtasks, and the number of samples in each task remains the same; S22, then the adaptive learning rate is adjusted. The adaptive learning rate algorithm is used to dynamically adjust the update learning rate of each parameter according to the historical gradient information. The specific update formula is: in is the sum of the squares of the historical gradients of each parameter, g i is the gradient of the i-th iteration; S23, then use the Hessian matrix To optimize the update, the second-order derivative information is introduced to improve the parameter update, thereby enhancing the stability of the model update. The parameter update formula is Where θ is the current model parameter, θ′ is the updated parameter, ▽ θ Γ(θ) is the first-order gradient, that is, the derivative of the loss function with respect to the parameter, H -1 It is the inverse of the matrix, which is used to adjust the update step size and increase the consideration of second-order information; S24, then design the weighted loss function The weighting factor ω i Adjust according to the complexity of the task, and the calculation formula is: Γ total1 (θ) is the total loss function, combining the losses of all tasks, K is the total number of tasks, and Γ i (θ) is the loss function for the ith task, and ε is a small constant to prevent division by zero; S25. Finally, an integrated learning method is used to combine the outputs of different models to improve the final prediction performance.

2. The method for predicting the performance of a finished automobile radiator based on machine learning according to claim 1, characterized in that: The manufacturing data of the radiator in step S1 includes: material composition including the type of metal material of the radiator and the proportion of each component, manufacturing process parameters including welding temperature, welding time, processing accuracy, stamping pressure, environmental conditions including temperature and humidity of the production environment, and performance data including heat dissipation efficiency and fluid flow rate.

3. The method for predicting the performance of a finished automobile radiator based on machine learning according to claim 1, characterized in that: The incremental verification in step S4 is implemented by first continuously monitoring and collecting the newly generated radiator performance data, performing batch incremental evaluation, dividing the new data into small batches, evaluating the impact of each batch on the model prediction performance in turn, and setting up real-time performance monitoring to track the prediction accuracy and physical consistency of the model during the incremental verification process. When the performance degrades, an alarm is automatically triggered and the indicators are adjusted.

Images