Part heat treatment performance prediction method and system based on BP neural network

Abstract

The application discloses a part heat treatment performance prediction method and system based on a BP neural network, relates to the technical field of material performance prediction, and comprises the following steps: acquiring the average cooling speed of a part to be predicted at a plurality of to-be-predicted sampling points; and inputting the average cooling speed of all the to-be-predicted sampling points into a performance prediction model to obtain the performance of the part to be predicted. The performance prediction model is constructed based on a BP neural network. The performance comprises hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness and wear resistance. The application improves the prediction accuracy of the heat treatment performance of the part.

Description

Technical Field

[0001] This invention relates to the field of material property prediction technology, and in particular to a method and system for predicting the heat treatment performance of parts based on a BP neural network. Background Technology

[0002] Metal parts often require heat treatment during manufacturing to improve their microstructure and enhance their mechanical properties. Temperature differences exist at various locations within a part during heat treatment, leading to variations in mechanical properties at different points. This is particularly true for large components after quenching, where significant temperature variations during cooling result in highly uneven distribution of mechanical properties. With the continuous development of manufacturing technology, the requirements for the heat treatment performance of parts are constantly increasing. There is an urgent need to predict the distribution of heat treatment properties to design heat treatment processes and tooling, thereby improving the performance distribution of parts and extending their service life.

[0003] The influence of heat treatment processes on performance is very complex, among which temperature history is the most critical factor affecting material properties. Materials can obtain different metallographic structures, precipitates, and dislocation densities through different heating, holding, and cooling processes, which in turn affect the mechanical properties of the materials. At present, the prediction of mechanical properties is mainly based on establishing a mathematical model between temperature, microstructure, and properties. The literature "Experimental Verification of Hardness Prediction Model" (Song Dongli et al. Mechanical Engineering Materials. 2008, 32(3):29-31,34) uses the Maynier and Carsic hardness calculation model to predict the hardness distribution after end quenching. This model calculates the hardness distribution through material composition and microstructure content, so its accuracy is easily affected by the accuracy of microstructure content calculation. The results of this study show that the hardness prediction model proposed by Carsic cannot accurately predict the hardness value of all materials after cooling. The literature "Aging Precipitation Kinetics and Strengthening Model of Aluminum Alloys" (Wang Xiaona et al. Chinese Journal of Nonferrous Metals, 2013(10):2754-2768.) summarizes the mechanical property calculation model of aluminum alloys during aging. Studies have shown that current predictive models for the age hardening of aluminum alloys mainly calculate the evolution of microstructure and the morphology of precipitated phases, and then calculate mechanical properties through different strengthening mechanism models. Current quantitative calculation models for the temperature-microstructure-property relationships during heat treatment are complex, with parameters difficult to obtain, making it challenging and inaccurate for calculating part properties, thus hindering engineering applications. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting the heat treatment performance of parts based on BP neural networks, which improves the prediction accuracy of the heat treatment performance of parts.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for predicting the heat treatment performance of parts based on a BP neural network, the method comprising:

[0007] Obtain the average cooling rate of the part to be predicted at multiple sampling points;

[0008] The average cooling rate of all sampling points to be predicted is input into the performance prediction model to obtain the performance of the part to be predicted; the performance prediction model is constructed based on a BP neural network; the performance includes: hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness and wear resistance.

[0009] Optionally, obtaining the average cooling rate of the part to be predicted at multiple sampling points specifically includes:

[0010] The part to be predicted is subjected to end quenching treatment, and the temperature-time curves of the part to be predicted at multiple sampling points during the end quenching treatment process are obtained.

[0011] The average cooling rate of each sampling point to be predicted is determined based on the temperature-time curve of each sampling point to be predicted.

[0012] Optionally, the training method for the performance prediction model includes:

[0013] A final-quenched specimen is selected; the final-quenched specimen is made of the same material as the part to be predicted.

[0014] The end-quenched sample is subjected to end-quenching treatment, and the temperature-time curves and performance values ​​of the end-quenched sample at multiple training sampling points are obtained during the end-quenching treatment process.

[0015] The average cooling rate of each training sampling point is determined based on the temperature-time curve of each training sampling point.

[0016] The BP neural network is trained using the average cooling rate of all training sampling points as input and the performance values ​​of all training sampling points as output to obtain the performance prediction model.

[0017] A component heat treatment performance prediction system based on BP neural network, the system comprising:

[0018] The first average cooling rate acquisition module is used to acquire the average cooling rate of the part to be predicted at multiple sampling points to be predicted.

[0019] The prediction module is used to input the average cooling rate of all sampling points to be predicted into the performance prediction model to obtain the performance of the part to be predicted; the performance prediction model is based on a BP neural network; the performance includes: hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness and wear resistance.

[0020] Optionally, the first average cooling rate acquisition module specifically includes:

[0021] The first end-quenching unit is used to perform end-quenching on the part to be predicted and to obtain the temperature-time curve of the part to be predicted at multiple sampling points during the end-quenching process.

[0022] The first average cooling rate determination unit is used to determine the average cooling rate of each sampling point to be predicted based on the temperature-time curve of each sampling point to be predicted.

[0023] Optionally, the prediction module includes: a performance prediction model training submodule, which includes:

[0024] An end-quenched specimen determination unit is used to determine the end-quenched specimen; the end-quenched specimen is made of the same material as the part to be predicted.

[0025] The second end-quenching unit is used to perform end-quenching on the end-quenched sample and obtain the temperature-time curve and performance values ​​of the end-quenched sample at multiple training sampling points during the end-quenching process.

[0026] The second average cooling rate determination unit is used to determine the average cooling rate of each training sampling point based on the temperature-time curve of each training sampling point.

[0027] The training unit is used to train the BP neural network with the average cooling rate of all training sampling points as input and the performance values ​​of all training sampling points as output, so as to obtain the performance prediction model.

[0028] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0029] This invention discloses a method and system for predicting the heat treatment performance of parts based on a BP neural network. The method includes: obtaining the average cooling rate of the part to be predicted at multiple sampling points; inputting the average cooling rate of all sampling points into a performance prediction model to obtain the performance of the part to be predicted; the performance prediction model is constructed based on a BP neural network; the performance includes: hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness, and wear resistance. Compared with quantitative calculation models targeting the temperature-microstructure-performance relationship during heat treatment, this invention only requires obtaining the average cooling rate of the part to be predicted at multiple sampling points as input to achieve the prediction of part performance. The average cooling rate can be obtained simply by end quenching, thus improving the prediction accuracy of the heat treatment performance of parts. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the process for predicting the heat treatment performance of parts based on a BP neural network, as provided in Embodiment 1 of the present invention.

[0032] Figure 2 A schematic diagram of a block-shaped part made of 42CrMo material;

[0033] Figure 3 Linear regression plot of the training results;

[0034] Figure 4 This is a schematic diagram illustrating the prediction results of hardness in a specific embodiment;

[0035] Figure 5 This is a schematic diagram of the structure of the part heat treatment performance prediction system based on BP neural network provided in Embodiment 2 of the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] The purpose of this invention is to provide a method and system for predicting the heat treatment performance of parts based on BP neural networks, aiming to improve the prediction accuracy of the heat treatment performance of parts.

[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] Example 1

[0040] Figure 1 This is a schematic diagram of the process for predicting the heat treatment performance of parts based on a BP neural network, as provided in Embodiment 1 of the present invention. Figure 1 As shown, the part heat treatment performance prediction method based on BP neural network in this embodiment includes:

[0041] Step 101: Obtain the average cooling rate of the part to be predicted at multiple sampling points.

[0042] Step 102: Input the average cooling rate of all sampling points to be predicted into the performance prediction model to obtain the performance of the part to be predicted; the performance prediction model is based on a BP neural network; the performance includes: hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness and wear resistance.

[0043] As an optional implementation, step 101 specifically includes:

[0044] The part to be predicted is subjected to end quenching treatment, and the temperature-time curves of the part to be predicted at multiple sampling points during the end quenching process are obtained.

[0045] The average cooling rate of each sampling point to be predicted is determined based on the temperature-time curve of each sampling point to be predicted.

[0046] As an optional implementation method, the training method for the performance prediction model includes:

[0047] Determine the end-quenched specimen; the end-quenched specimen and the part to be predicted are made of the same material.

[0048] The end-quenched samples were subjected to end-quenching treatment, and the temperature-time curves and performance values ​​of the end-quenched samples at multiple training sampling points were obtained during the end-quenching process.

[0049] The average cooling rate of each training sampling point is determined based on the temperature-time curve of each training sampling point.

[0050] The BP neural network is trained using the average cooling rate of all training sampling points as input and the performance values ​​of all training sampling points as output to obtain a performance prediction model.

[0051] In practice, the training process of a performance prediction model includes:

[0052] (1) The material that is consistent with the part to be predicted is processed into an end-quenched sample.

[0053] Specifically, the chemical composition, initial microstructure, and grain size of the end-quenched sample material are consistent with the state of the part whose performance is to be predicted before heat treatment.

[0054] (2) End quenching treatment: The end quenching sample is heated to the austenitizing temperature using the same heating process as the heat treatment of the part whose performance is to be predicted, and held for the same amount of time. Then the sample is taken out of the furnace and transferred to the end quenching platform. The water spray device is turned on to perform end face quenching treatment. The sample is removed after it has completely cooled down.

[0055] Specifically, if the actual performance heat treatment includes tempering, the cooled end-quenched sample also needs to undergo the same tempering process.

[0056] (3) Move the end-quenched specimen along the axial direction (i.e., the direction of the axis of the end-quenched specimen, such as...) Figure 2 Samples are taken at regular intervals along the length direction of the sample to perform mechanical property testing and analysis, and the performance values ​​are obtained.

[0057] When the predicted property is hardness, a parallel plane is ground along the axis of the end-quenched sample, with a grinding depth of approximately 1.5 mm. A grinding wheel is used, and water cooling is maintained during the grinding process to prevent the sample temperature from becoming too high and altering the microstructure. The hardness of the sample is measured at a position on the ground plane, at a certain distance from the quenched end face, along the direction away from the quenched end face.

[0058] When the predicted properties are yield strength, tensile strength, elongation and reduction of area, a tensile specimen with a thickness of 1 mm needs to be prepared along the cross-section at a certain distance from the quenched end face, and the specimen is subjected to a room temperature tensile test to obtain the yield strength, tensile strength, elongation and reduction of area at different positions of the end-quenched specimen.

[0059] When the predicted properties are fatigue strength, fracture toughness, impact toughness and wear resistance, the cross-section located at a certain distance from the quenched end face should be taken as the research object, and corresponding test specimens should be designed for testing. The specimen thickness should not exceed 3 mm.

[0060] (4) Perform finite element numerical simulation on the end quenching process, calculate the temperature-time curve of the sampling position (training sampling point) during the end quenching process, and take the average cooling rate at certain temperatures from the temperature-time curve within the temperature range below material Ac3 (Ac3 represents the temperature at which all ferrite is transformed into austenite when hypoeutectoid steel is heated) to obtain the average cooling rate of each training sampling point.

[0061] Specifically, a finite element geometric model of the end-quenched specimen was established, and the model was meshed with a mesh size of less than 1 mm. The material density, thermal conductivity, and specific heat capacity of the model were set. The initial temperature of the model was set to the heat treatment heating temperature, the heat transfer coefficient of the water-quenched end was set to the heat transfer coefficient of water, and the remaining boundaries were set to air cooling. Transient temperature field calculations were performed on the above model to obtain nodes at a certain distance from the quenched end face (each node corresponds to the same distance, see...). Figure 2 The temperature history data at the sampling points (one node is one sampling point) are used to obtain the temperature-time curve.

[0062] (5) Design a BP neural network with average cooling rate data at different locations as the input layer and corresponding mechanical performance test data (performance values) as the output layer. Select a reasonable number of hidden layers, training method and training parameters for the backpropagation (BP) neural network to train the neural network and obtain the mapping relationship between cooling rate and mechanical performance.

[0063] Specifically, the BP neural network has a three-layer structure: an input layer, a hidden layer, and an output layer. The input layer is the average cooling rate extracted in step (4). The output layer is the numerical value of the corresponding performance in step (3). The number of neurons in the hidden layer satisfies the following condition:

[0064] 2 u >v(1)

[0065]

[0066] Where u is the number of hidden layer neurons; v is the number of input layer nodes; e is the number of output layer nodes; and r is an integer between 0 and 10. In this invention, the number of input layer nodes is determined by the cooling rate array, and the number of output layer nodes is a certain performance data (the cooling rate array is obtained by dividing the cooling curve into segments (e.g., every 10°C), and then taking the average cooling rate of this segment. The entire cooling process will have multiple average cooling rates, which is the cooling rate array, therefore e = 1.

[0067] The hidden layer activation function is the sigmoid function, and the training function is Bayesian regularization. The ratio of the training set, validation set, and test set in the input data is approximately 8:1:1.

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0069] This invention makes full use of the relatively accurate temperature calculation results in numerical simulation and uses the BP neural network algorithm to establish a nonlinear relationship between temperature history and heat treatment performance. This avoids the need to build a temperature-microstructure-performance model to predict the heat treatment performance distribution, thereby reducing the complexity of the model and improving the calculation accuracy.

[0070] This invention can be applied to predicting the heat treatment properties of most metals, including steel, aluminum alloys, titanium alloys, and zirconium alloys. It is also applicable to more complex heat treatment processes, such as multi-pass quenching or multi-pass tempering.

[0071] The experimental and simulation techniques of this invention are relatively simple, the parameter settings are easy, the research and development cycle is short, the cost is controllable, and it has good engineering application value.

[0072] Specific embodiment: A block-shaped part (100mm×100mm×50mm) made of 42CrMo material is quenched. The quenching process is to hold at 850℃ for 1 hour and then quench in water. The hardness distribution after quenching needs to be predicted and illustrated with examples.

[0073] Step 1: Process the material that matches the part into an end-quenched sample.

[0074] In this embodiment, the material is 42CrMo. The same material was used to process end-quenched samples, and the sample dimensions are as follows: Figure 2 As shown, the chemical composition, initial microstructure, and grain size of the material are consistent with the state of the block part before quenching.

[0075] Step 2: Heat the end-quenched sample to the austenitizing temperature using the same heating process as the heat treatment of the part, and hold it at that temperature for the same amount of time. Then, transfer the sample from the furnace to the end-quenching platform, turn on the water spray device, and perform end-face quenching treatment. After the sample has completely cooled down, remove it from the furnace.

[0076] In this embodiment, the end-quenched sample is heated to 850°C and held for 1 hour. Then, the sample is taken out of the furnace and transferred to the end-quenching platform. The water spray device is turned on to perform end-face quenching treatment. The sample is removed after it has completely cooled down.

[0077] Step 3: Take samples at regular intervals along the axial direction of the end-quenched specimen for mechanical property testing and analysis.

[0078] In this embodiment, a parallel plane is ground along the axial direction of the end-quenched sample, with a grinding depth of approximately 1.5 mm. A grinding wheel is used, and water cooling is maintained during the grinding process to prevent excessive sample temperature from altering the microstructure. On the grinding plane, along the direction away from the quenched end face, Rockwell hardness measurements are performed at positions 1.5 mm, 5 mm, 10 mm, 15 mm, 20 mm, 25 mm, 30 mm, 35 mm, 40 mm, 45 mm, 50 mm, 55 mm, 60 mm, 65 mm, and 70 mm away from the quenched end face. Figure 2 (as shown), and record the hardness data.

[0079] Step 4: Perform finite element numerical simulation of the end quenching process, calculate the temperature curve of the sampling location during the end quenching process, and take the average cooling rate at certain temperature intervals within the temperature range below Ac3 of the material.

[0080] In this embodiment, the specific details are as follows:

[0081] A two-dimensional axisymmetric model was used to establish the end-quenching geometric model, and the model was divided into quadrilateral meshes with a mesh size of 0.5 mm.

[0082] The model material parameters are set to the density, thermal conductivity, and specific heat capacity of 42CrMo. The initial model temperature is set to 850℃, and the heat transfer coefficient at the water-quenched end is set to 10000 W / (m²). 2 ·℃), with the remaining boundaries set as air-cooled heat exchange, and a heat transfer coefficient of 50W / (m²). 2 ·℃).

[0083] Transient temperature field calculations were performed on the above model to obtain temperature history data at nodes 1.5mm, 5mm, 10mm, 15mm, 20mm, 25mm, 30mm, 35mm, 40mm, 45mm, 50mm, 55mm, 60mm, 65mm and 70mm away from the quenched end face.

[0084] The Ac3 temperature of 42CrMo material is 800℃. Therefore, data from 800℃ to 100℃ were extracted from the temperature data, and the average cooling rate within this temperature range was calculated every 20℃ to form the cooling rate array for that node. The cooling rate array for each node was then correlated with the corresponding location performance data obtained in step three. Table 1 shows some of the data.

[0085] Table 1 Neural Network Training Data

[0086]

[0087]

[0088] Step 5: Design a BP neural network, using the average cooling rate data at different locations as the input layer and the corresponding mechanical performance test data as the output layer. Select an appropriate number of hidden layers, training method, and training parameters for the BP neural network training to obtain the mapping relationship between cooling rate and mechanical performance.

[0089] In this embodiment, there are 35 average cooling rate data points for each location, and one hardness data point. The number of nodes in the input layer of the BP neural network is set to 35, and the number of nodes in the output layer is set to 1. The number of hidden layer neurons should be in the range of 6-16 according to formula (2). Here, 10 hidden layer neurons are selected. The final designed BP neural network structure is a 35-10-1 three-layer structure. The activation function of the hidden layer is set to the sigmoid function, and the training function is Bayesian regularization. There are a total of 15 sets of input data. According to the approximate ratio of training set, validation set, and test set data of 8:1:1, 11 sets of training set, 2 sets of validation set, and 2 sets of test set are designed. Figure 3 The image shows a linear regression plot of the training results. The accuracy of the trained model is 99.8%.

[0090] Step Six: Establish a finite element numerical model of the heat treatment of the part to be predicted, obtain cooling curves at different locations, and take the average cooling rate at certain temperature intervals below Ac3. Using the neural network algorithm established in the above steps, with the average cooling rate data as input, calculate the performance data at different locations.

[0091] In this embodiment, the specific details are as follows:

[0092] A 3D geometric model of the part is created. Since the part is a cube and has symmetry, the model can be set to 1 / 8 scale for analysis. A hexahedral mesh with a size of 1 mm is used for mesh generation.

[0093] The model material parameters are set to the density, thermal conductivity, and specific heat capacity of 42CrMo. The initial model temperature is set to 850℃, and the heat transfer coefficient at the water-quenched end is set to 10000 W / (m²). 2 ·℃), and set the remaining boundaries as air-cooled heat exchange, with a heat transfer coefficient of 50W / (m2·℃).

[0094] Transient temperature field calculations were performed on the above model to obtain temperature history data at all nodes of the model.

[0095] Extract the 800-100℃ temperature data from the temperature data. For the temperature history data at different nodes, calculate the average cooling rate within that temperature range every 20℃, and use it as the cooling rate array for that node.

[0096] Using the cooling rate arrays of all nodes as input layers, and employing the BP neural network model constructed in step five, performance data at different nodes can be obtained. Importing the data into the model yields hardness values ​​at different locations, such as... Figure 4 As shown.

[0097] The predicted surface hardness of the part was 62.8 HRC, and the core hardness was 57.2 HRC. The actual measured surface hardness was 62.3 HRC, and the core hardness was 56.7 HRC. The simulation and measurement results are basically consistent.

[0098] Example 2

[0099] Figure 5 This is a schematic diagram of the component heat treatment performance prediction system based on a BP neural network provided in Embodiment 2 of the present invention. Figure 5 As shown, the part heat treatment performance prediction system based on BP neural network in this embodiment includes:

[0100] The first average cooling rate acquisition module 201 is used to acquire the average cooling rate of the part to be predicted at multiple sampling points to be predicted.

[0101] The prediction module 202 is used to input the average cooling rate of all sampling points to be predicted into the performance prediction model to obtain the performance of the part to be predicted. The performance prediction model is based on a BP neural network. The performance includes: hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness and wear resistance.

[0102] As an optional implementation, the first average cooling rate acquisition module 201 specifically includes:

[0103] The first end-quenching unit is used to perform end-quenching on the part to be predicted and to obtain the temperature-time curves of the part to be predicted at multiple sampling points during the end-quenching process.

[0104] The first average cooling rate determination unit is used to determine the average cooling rate of each sampling point to be predicted based on the temperature-time curve of each sampling point to be predicted.

[0105] As an optional implementation, the prediction module 202 includes: a performance prediction model training submodule, which includes:

[0106] The end-quenched specimen determination unit is used to determine the end-quenched specimen; the end-quenched specimen and the part to be predicted are made of the same material.

[0107] The second end-quenching unit is used to perform end-quenching on the end-quenched sample and obtain the temperature-time curves and performance values ​​of the end-quenched sample at multiple training sampling points during the end-quenching process.

[0108] The second average cooling rate determination unit is used to determine the average cooling rate of each training sampling point based on the temperature-time curve of each training sampling point.

[0109] The training unit is used to train the BP neural network with the average cooling rate of all training sampling points as input and the performance values ​​of all training sampling points as output, so as to obtain the performance prediction model.

[0110] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0111] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting the heat treatment performance of parts based on a BP neural network, characterized in that... The method includes: Obtain the average cooling rate of the part to be predicted at multiple sampling points; The average cooling rate of all sampling points to be predicted is input into the performance prediction model to obtain the results. The performance of the part to be predicted is described; the performance prediction model is constructed based on a BP neural network; the performance includes: hardness, yield strength, tensile strength, elongation, reduction of area, fatigue strength, fracture toughness, impact toughness, and wear resistance; Obtain the average cooling rate of the part to be predicted at multiple sampling points, specifically including: The part to be predicted is subjected to end quenching treatment to obtain the part during the end quenching process. Predict the temperature-time curve of the part at multiple of the predicted sampling points; Each sampling point to be predicted is determined based on its temperature-time curve. The average cooling rate; Training methods for performance prediction models include: A final-quenched specimen is selected; the final-quenched specimen is made of the same material as the part to be predicted. The end-quenched sample is subjected to end-quenching treatment to obtain the end-quenched sample during the end-quenching process. Temperature-time curves and performance values ​​of the sample at multiple training sampling points; specifically including: The end-quenched sample was heated to the austenitizing temperature using the same heating process as the heat treatment of the part, and held at that temperature for the same time. Then the sample was removed from the furnace and transferred to the end-quenching platform. The water spray device was turned on to perform end-face quenching treatment. The sample was removed after it had completely cooled down. Samples are taken at regular intervals along the axial direction of the end-quenched specimen for mechanical property testing and analysis, specifically including: A parallel plane is ground along the axis of the end-quenched sample, with a grinding depth of [missing information]. 1.5mm; The grinding method is adopted, and water cooling is maintained during the grinding process to prevent the sample temperature from being too high and causing changes in the microstructure; The flatness of each training sampling point is determined based on the temperature-time curve of each training sampling point. Uniform cooling rate; Using the average cooling rate of all training sampling points as input, and the performance of all training sampling points as input... The value is used as the output to train the BP neural network and obtain the performance prediction model; During the training process of the performance prediction model, a finite element geometric model of the end-quenched specimen is established. The finite element geometric model is meshed using a mesh size of less than 1 mm, and the following settings are applied: Material density, thermal conductivity, and specific heat capacity data; initial temperature set to the heat treatment heating temperature. Set the heat transfer coefficient of the water-quenched end to the heat transfer coefficient of water, and set the heat transfer of the remaining boundaries to air cooling. Transient temperature field calculations were performed on the finite element geometric model to obtain the temperature field at a certain distance from the quenched end face. The temperature history data at a given point is used to obtain a temperature-time curve.

2. A part heat treatment performance prediction system based on a BP neural network, applied to the part heat treatment performance prediction method based on a BP neural network in claim 1, characterized in that, The system includes: The first average cooling rate acquisition module is used to acquire the part to be predicted from multiple samples to be predicted. Average cooling rate at the point; The prediction module is used to input the average cooling rate of all sampling points to be predicted into the performance prediction. The model yields the performance of the part to be predicted; the performance prediction model is based on BP. The neural network is used to construct the properties, which include: hardness, yield strength, tensile strength, and elongation. Factors such as reduction of area, fatigue strength, fracture toughness, impact toughness, and wear resistance; The first average cooling rate acquisition module specifically includes: The first end-quenching unit is used to perform end-quenching treatment on the part to be predicted, and obtain... Temperature of the part to be predicted at multiple sampling points during the end-quenching process Time curve; The first average cooling rate determination unit is used to determine the temperature and time of each sampling point to be predicted. The curve determines the average cooling rate for each sampling point to be predicted. The prediction module includes: a performance prediction model training submodule, wherein the performance prediction model... The training submodule includes: An end-quenched specimen determination unit is used to determine the end-quenched specimen; the end-quenched specimen and the specimen to be determined The parts are predicted to be made of the same material; The second end-quenching unit is used to perform end-quenching treatment on the end-quenched sample to obtain end... Temperature-time curves and performance values ​​of the end-quenched specimen at multiple training sampling points during the quenching process; The second average cooling rate determination unit is used to determine the temperature-time curve for each training sampling point. The line determines the average cooling rate for each training sampling point; Training units are used to take the average cooling rate of all training sampling points as input and generate training data from all training samples. The performance values ​​of the training sampling points are used as the output to train the BP neural network and obtain the performance prediction model. During the training process of the performance prediction model, a finite element geometric model of the end-quenched specimen is established. The finite element geometric model is meshed with a mesh size of less than 1 mm, and the following settings are applied. Material density, thermal conductivity, and specific heat capacity data; initial temperature set to the heat treatment heating temperature. Set the heat transfer coefficient of the water-quenched end to the heat transfer coefficient of water, and set the heat transfer of the remaining boundaries to air cooling. Transient temperature field calculations were performed on the finite element geometric model to obtain the temperature field at a certain distance from the quenched end face. The temperature history data at a given point is used to obtain a temperature-time curve.

Images