Deep learning fused virtual simulation system for gas carburization and performance test

Abstract

The invention relates to a deep learning fused virtual simulation system for gas carburization and performance testing, belongs to the technical field of artificial intelligence and virtual simulation, and solves the problems that an existing virtual simulation system cannot accurately predict part heat treatment process parameters and cannot provide customized teaching assistance in the carburization experiment aspect. The system comprises a digital scene multivariate change module, a digital human customization module, a teaching instruction management and control module, an experiment teaching comprehensive evaluation and result display module, an experiment AI navigation module and a multi-task prediction and modeling module based on deep learning. Precise prediction of part heat treatment process parameters in the carburizing experiment aspect is achieved through the multi-task prediction and modeling module based on deep learning, personalized teaching guidance is provided in combination with the experiment AI navigation module enabling the knowledge graph, multiple functions are integrated, operation is convenient and free, and the teaching efficiency is improved. And an innovative and efficient virtual platform is provided for teaching and research of gas carburization and performance test.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and virtual simulation technology, and in particular to a virtual simulation system for gas carburizing and performance testing integrating deep learning. Background Art

[0002] In the education and training scenarios related to modern manufacturing, the teaching of manufacturing processes for parts such as gears is extremely important, and gas carburizing and performance testing are the core points. Traditional teaching practices encounter many obstacles here, because gas carburizing experiments require extremely high professional equipment. Equipment such as pit carburizing furnaces, sealed box furnaces, and vacuum carburizing furnaces are not only expensive, but also require a lot of money and manpower for subsequent maintenance. This makes it difficult for teaching institutions to fully equip themselves, and the problem of scarce teaching resources becomes prominent, making it difficult to meet the practical demands of many learners.

[0003] The experimental operation itself has significant safety hazards. Carburizing, quenching, tempering and other processes involve high temperature, high pressure and complex chemical environment. Any slight mistake may cause serious safety accidents and endanger the personal safety of teachers and students. In addition, it has strict requirements on site conditions. Specific ventilation, temperature control, safety protection facilities are indispensable, which further increases the difficulty and cost of conducting experiments.

[0004] With the advancement of science and technology, virtual simulation technology has brought a turning point to the teaching dilemma. It can rely on computers to simulate real experimental situations and processes, avoiding many drawbacks of traditional experiments. However, the relevant virtual simulation systems on the market currently have general shortcomings. Most of them only realize basic process simulation and have limited achievements in the expansion of deep teaching functions. In particular, in terms of using deep learning technology to achieve accurate prediction of gear heat treatment process parameters and providing customized teaching assistance based on individual differences of learners, it is almost blank. Summary of the invention

[0005] In order to solve the problems that the existing virtual simulation system cannot accurately predict the heat treatment process parameters of parts in carburizing experiments and cannot provide customized teaching assistance.

[0006] To solve the above problems, the present invention adopts the following technical solutions:

[0007] A virtual simulation system for gas carburizing and performance testing that integrates deep learning, including a digital scene multivariate change module, a digital human customization module, a teaching instruction control module, an experimental teaching comprehensive evaluation and achievement display module, an experimental AI pilot module, and a multi-task prediction and modeling module based on deep learning;

[0008] The digital scene multivariate change module is used to establish a personalized digital workshop according to preset experimental requirements. The module includes an experimental equipment autonomous selection submodule and a workshop background change submodule, wherein the experimental equipment autonomous selection submodule is used to provide virtual experimental equipment according to experimental requirements, and the virtual experimental equipment includes a carburizing furnace and a carburizing furnace hanger; the workshop background change submodule is used to provide preset scenes of different styles for the digital workshop and provide different wall, floor, table and chair styles for the sample preparation area or performance test area in the digital workshop;

[0009] The digital human customization module is used to customize the image of the experimental subject in the workshop. The module includes a digital human body shape construction submodule, a digital human face customization submodule and a digital human image style submodule, wherein the digital human body shape construction submodule has a plurality of preset body templates built in, and the preset body templates include height, leg length, arm length, strength, neck length, and body posture parameters that can be customized; the digital human face customization submodule obtains original facial image data, and outputs the original facial image data to the facial feature deep learning modeling submodule in the multi-task prediction and modeling module based on deep learning for digital modeling, and then customizes the head and facial appearance features of the digital human according to the constructed facial three-dimensional model returned by the facial feature deep learning modeling submodule; the digital human image style submodule has a plurality of preset clothing templates built in, and the preset clothing templates include the digital human's top, pants, shoes, and accessories parameters that can be customized;

[0010] The teaching instruction control module includes an experiment preview submodule, an experiment operation submodule and an experiment assessment submodule. The experiment preview submodule and the experiment assessment submodule respectively interact with the experiment AI navigation module for data. The experiment preview submodule is used to assess the basic knowledge of users, and feed back the assessment results of each user to the experiment AI navigation module. It is also used to track the user's preview progress and visualize the preview progress. The experiment operation submodule is used to guide the user to complete the preset experiment operation process through interactive simulation after obtaining the part materials and experimental parts selected by the user, and guide the user's operation in real time during the experiment execution through one or more combinations of highlighted display areas, text prompt boxes, audio instruction broadcasts and video animation playback. At the same time, the experiment operation submodule is also connected to the part thermal treatment module based on deep learning. The heat treatment process parameter prediction submodule performs data interaction, and the part heat treatment process parameter prediction submodule sends the predicted heat treatment process parameter results to the experimental operation submodule and the experimental AI navigation module respectively to provide process parameter guidance for users, wherein the experimental operation process includes seven experimental process links, namely, hardness test before carburizing, carburizing experiment, carburized layer detection, quenching experiment, tempering experiment, hardness test after carburizing, and metallographic experiment; the experimental assessment submodule is used to assess and evaluate the user's familiarity with the experimental operation process and related knowledge according to the customized questions output by the experimental AI navigation module, generate assessment results, and judge the correctness of the user's operation through the experimental operation submodule according to the real experimental process, generate operation judgment results, and feed back the assessment results and the operation judgment results to the experimental teaching comprehensive evaluation and achievement display module;

[0011] The experimental teaching comprehensive evaluation and achievement display module is used to display the assessment results and the operation evaluation results, and output the user's knowledge weaknesses and operation errors, and output the experimental result table;

[0012] The experimental AI pilot module is used to assist the teaching instruction control module, including analyzing the user's knowledge gaps according to the collected assessment results of the experimental preview submodule and accurately customizing key review knowledge and related knowledge assessment questions, and is also used to receive question instructions output by the user, and realize intelligent response to question instructions through natural language processing technology and knowledge graph technology and display intelligent response results;

[0013] The multi-task prediction and modeling module based on deep learning includes the facial feature deep learning modeling submodule and the part heat treatment process parameter prediction submodule. The facial feature deep learning modeling submodule is used to extract key features of the original facial image data through a convolutional neural network, construct a facial three-dimensional model and output it to the digital human face customization submodule; the part heat treatment process parameter prediction submodule is used to predict the heat treatment process parameters using a convolutional neural network after obtaining the material, type and target hardness information of the part, and obtain predicted heat treatment process parameter results. The heat treatment process parameter results include carburizing parameters, quenching parameters, tempering time and temperature parameters, and are also used to receive process parameter consultation instructions input by the user, parse the process parameter consultation instructions and output recommended heat treatment process parameter values.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] The virtual simulation system of gas carburizing and performance testing that integrates deep learning proposed by the present invention realizes the accurate prediction of the heat treatment process parameters of carburizing experimental parts through the multi-task prediction and modeling module based on deep learning, and provides personalized teaching guidance in combination with the experimental AI pilot module empowered by the knowledge graph. At the same time, a highly free immersive experimental environment is constructed through the digital scene multivariate change module and the digital human customization module. The system integrates multiple functions in one, is easy to operate and free, and provides an innovative and efficient virtual platform for the teaching and research of gas carburizing and performance testing. Through artificial intelligence and virtual simulation technology, the present invention can simulate the experimental process of carburizing treatment of parts in real scenarios, thereby improving the user's familiarity and intuitive feeling of the high-risk experimental operation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is an overall framework diagram of the virtual simulation system according to an embodiment of the present invention;

[0017] Figure 2 It is the structural diagram of the digital workshop;

[0018] Figure 3 A flowchart for the experimental operation submodule to guide users to complete the experimental operation process;

[0019] Figure 4 This is the flow chart of the hardness test process in the experimental operation submodule. DETAILED DESCRIPTION

[0020] The technical solution of the present invention will be described clearly and completely in conjunction with the embodiments below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, this embodiment provides a virtual simulation system for gas carburizing and performance testing that integrates deep learning. The system mainly includes a digital scene multi-variable change module, a digital human customization module, a teaching instruction management module, an experimental teaching comprehensive evaluation and achievement display module, an experimental AI navigation module, and a multi-task prediction and modeling module based on deep learning.

[0022] The digital scene multivariate change module is used to establish a personalized digital workshop according to the user's preferences or preset experimental requirements. The established digital workshop includes a heat treatment area, a sample preparation area, a performance test area, etc. The digital scene multivariate change module includes an experimental equipment autonomous selection submodule and a workshop background change submodule. Among them, the experimental equipment autonomous selection submodule is used to provide virtual experimental equipment for users to choose according to different experimental requirements. The virtual experimental equipment includes but is not limited to carburizing furnaces and carburizing furnace hangers. For example, this submodule can provide 3 optional carburizing furnaces, including pit carburizing furnaces, sealed box furnaces, vacuum carburizing furnaces, and 4 carburizing furnace hangers suitable for various parts. The workshop background change submodule is used to provide preset scenes of different styles for the digital workshop. For example, 8 styles of preset scenes are provided, and users can choose 1 as the scene style. It can also be used to provide different wall, floor, table and chair styles for the sample preparation area or performance test area in the digital workshop for users to independently allocate.

[0023] The digital human customization module is used to customize the image of the experimental subject, i.e., the digital human, in the workshop according to the user's preferences. The digital human customization module includes a digital human body shape construction submodule, a digital human face customization submodule, and a digital human image style submodule.

[0024] The digital human body construction submodule has built-in multiple preset body templates, and each preset body template includes customizable height, leg length, arm length, strength, neck length, and body posture parameters, so that users can customize the height, leg length, arm length, strength, neck length, body posture, etc. of the digital human. For example, the digital human body construction submodule has built-in 5 preset body templates, and users can carry out personalized adjustments based on the preset body templates.

[0025] The digital human face generation submodule is used to customize the head and facial features of the digital human. The digital human face generation submodule first obtains the original facial image data uploaded by the user, such as 2-5 pictures of the front face and the side face, and then outputs the original facial image data to the facial feature deep learning modeling submodule in the multi-task prediction and modeling module based on deep learning for digital modeling. After modeling, the facial feature deep learning modeling submodule obtains a facial 3D model, and returns the facial 3D model to the digital human face generation submodule. The digital human face generation submodule customizes the head and facial features of the digital human according to the facial 3D model.

[0026] The digital human image style submodule is used to customize the digital human's clothing style. It has built-in multiple preset clothing templates, and each preset clothing template includes customizable parameters of the digital human's top, pants, shoes, and accessories, so that users can customize the digital human's top, pants, shoes, and accessories. For example, the digital human image style submodule has built-in 5 sets of preset clothing templates, and users can make personalized adjustments based on the preset clothing templates.

[0027] The teaching instruction management and control module is used to select and start different sub-modules according to the different needs of different learning stages. The sub-modules include the experiment preview sub-module, the experiment operation sub-module and the experiment assessment sub-module, and the experiment preview sub-module and the experiment assessment sub-module respectively interact with the experiment AI navigation module for data.

[0028] The experiment preview submodule is used to assess the basic knowledge of users, that is, to achieve the knowledge reserve and basic knowledge assessment before the experiment. This submodule will record each user's assessment results and feed back each user's assessment results to the experiment AI pilot module. The experiment AI pilot module will analyze the user's knowledge gaps and accurately customize the key review knowledge and related knowledge assessment questions according to the collected assessment results. At the same time, this submodule also has the function of tracking the user's preview progress and displaying the user's learning progress on different knowledge points through visual methods such as progress bars and charts.

[0029] The experimental operation submodule is used to let users who are not familiar with the experimental process understand the experimental process. In the experimental operation submodule, users can select part materials and experimental parts before the experiment. For example, users can select one of four alloy materials as part materials, including 20CrMo, 20CrMnTi, 18Cr2Ni4WA, and 20MnCr5; they can select one of three gears as experimental parts, including worm gears, cylindrical gears, and bevel gears. After obtaining the part materials and experimental parts selected by the user, the experimental operation submodule guides the user to complete the preset experimental operation process through interactive simulation, and during the experimental execution, it guides the user's operation in real time through one or more combinations of highlighted display areas, text prompt boxes, audio instruction broadcasts, and video animation playback. Furthermore, the experimental operation submodule can also be combined with water flow simulation, flame effects, gas flow display, and sound prompts for joint demonstration during the experimental operation. At the same time, the experimental operation submodule also exchanges data with the part heat treatment process parameter prediction submodule in the multi-task prediction and modeling module based on deep learning. The part heat treatment process parameter prediction submodule sends the predicted heat treatment process parameter results to the experimental operation submodule and the experimental AI pilot module respectively to provide users with process parameter guidance. For example, the carburizing experiment, quenching experiment, and tempering experiment in the experimental operation process will be combined with the part heat treatment process parameter prediction submodule to propose recommended temperatures and times for the steps involving temperature and time in the experimental process based on different materials and the expected hardness of the experiment.

[0030] like Figure 3 As shown in the figure, the experimental operation process that the experimental operation submodule guides users to complete includes seven experimental process links, namely, hardness test before carburizing, carburizing experiment, carburized layer detection, quenching experiment, tempering experiment, hardness test after carburizing and metallographic experiment. Among them, the hardness test before carburizing only samples the surface of the experimental parts, while the hardness test after carburizing samples and tests different areas such as the surface and core of the experimental parts; carburized layer detection is to detect the depth of the carburized layer under different carburizing temperatures and times; metallographic experiment is to sample different areas of the carburized parts and perform metallographic microstructure analysis; the quenching experiment link includes two quenching methods, namely direct quenching and one-time quenching, and the two quenching methods will directly affect the experimental results of the subsequent hardness test and metallographic experiment link. In the quenching experiment, tempering experiment, hardness test after carburizing and metallographic experiment links, users can conduct independent exploration through the experimental operation submodule. After the experiment, the experimental operation submodule also compares and analyzes the metallographic experimental results obtained by the two quenching methods, as well as the hardness of the parts before and after carburizing, in order to explore the influence of the carburizing process on the hardness of the parts and the differential effects of different quenching methods on the changes in the microstructure and performance of the parts.

[0031] On the one hand, the experimental assessment submodule is used to assess the user's familiarity with the experimental operation process and related knowledge. For example, the experimental assessment submodule can be combined with the experimental AI pilot module to customize questions for different users. 65% of the questions cover the knowledge gaps exposed in the preview stage, and 35% of the questions are ordinary questions covering the experimental principles, general points of the operation process, and basic functions of the equipment. The user is assessed and evaluated according to the customized questions output by the experimental AI pilot module, and the corresponding assessment results are finally generated. The assessment results can be reflected in the form of scores; on the other hand, the experimental assessment submodule is used to judge the correctness of the user's operation through the experimental operation submodule according to the real experimental process. At this time, the experimental assessment content involves all links in the experimental operation submodule. The assessment process will be assessed according to the real experimental process. During the user's experimental operation, the experimental assessment submodule will judge the correctness of his operation and finally generate the operation judgment result, which can also be reflected in the form of scores. The experimental assessment submodule will feed back the assessment results and operation judgment results to the comprehensive evaluation and achievement display module of experimental teaching.

[0032] The experimental teaching comprehensive evaluation and achievement display module is used to display the assessment results and operation evaluation results, and output the user's knowledge weaknesses and operation errors, and output the experimental result table. For example, the experimental teaching comprehensive evaluation and achievement display module calculates the assessment question scores in the experimental assessment submodule and the scores obtained by the experimental operation process and displays them, and also outputs and displays the user's knowledge weaknesses and operation errors, and outputs the experimental result table for user analysis.

[0033] The experimental AI pilot module is used to assist the teaching instruction control module, including analyzing the user's knowledge gaps according to the assessment results of the collected experimental preview submodules and accurately customizing key review knowledge and related knowledge assessment questions. It is also used to receive question instructions output by the user, and through natural language processing technology and knowledge graph technology, it realizes intelligent response to question instructions and displays intelligent response results, thereby providing users with personalized teaching guidance. The experimental AI pilot module can solve questions about this system and this experiment that arise during the use of this system, and assist the instruction control module in teaching. The experimental AI pilot module uses natural language processing technology and the NLTK library and framework in Python to perform lexical and syntactic analysis on user questions, extract key information about experimental equipment, materials, steps and system operations, and construct a knowledge graph covering all elements of gas carburizing and performance testing experiments and their relationships. It forms knowledge nodes and connection relationships based on data collected and organized from professional materials, and uses a classification model trained with a deep learning algorithm to identify the type of question. For questions that require numerical information, it collaborates with the multi-task prediction and modeling module based on deep learning in the system to obtain data, and uses a dialogue management system to record the dialogue process to achieve context-sensitive responses, and optimizes the answer strategy and internal model based on user satisfaction evaluation.

[0034] The multi-task prediction and modeling module based on deep learning is used to predict the time and temperature of carburizing, quenching and tempering required for gears of different materials, types and sizes to reach the predetermined hardness requirements after the test. At the same time, in the head and face modeling task, the module uses deep learning technology to extract facial features and construct a three-dimensional facial model. The multi-task prediction and modeling module based on deep learning specifically includes a submodule for predicting the process parameters of heat treatment of parts and a submodule for deep learning modeling of facial features.

[0035] The part heat treatment process parameter prediction submodule is used to accurately predict the heat treatment process parameters using a convolutional neural network after obtaining the material, type and target hardness information of the part, and obtain the predicted heat treatment process parameter results. The convolutional neural network (CNN) architecture used in the part heat treatment process parameter prediction submodule includes an input layer, three hidden layers, and an output layer. Specifically, the first hidden layer contains 64 neurons, the second hidden layer contains 128 neurons, and the third hidden layer contains 256 neurons. Between these hidden layers, the ReLU function is used as a nonlinear activation function to introduce nonlinear characteristics to help the model learn complex data patterns.

[0036] The process of training the convolutional neural network used in the part heat treatment process parameter prediction submodule includes the following steps:

[0037] Collect several sample data of parts that have been processed by different heat treatment processes to build a training data set, where the sample data of the parts includes the material, type, heat treatment process parameters and the final hardness value of the parts;

[0038] Standardize all parts sample data;

[0039] The convolutional neural network is trained using the standardized data set, the training rounds and initial values ​​of the learning rate are set, and the network is optimized using the mean square error as the loss function.

[0040] Specifically, here we take gear parts as an example. In order to train the deep learning model, 8,600 gear sample data that have been processed by different heat treatment processes (carburizing, quenching, and tempering) are collected. These sample data cover a variety of information, including gear material information, such as 20CrMo, 20CrMnTi, 18Cr2Ni4WA, 20MnCr5 and other common materials; gear types, such as worm gears, cylindrical gears, and bevel gears; and corresponding heat treatment process parameters, including carburizing temperature range of 800℃ to 1000℃, carburizing time of 5 to 20 hours, quenching temperature of 850℃ to 1200℃, quenching time of 20 to 60 minutes, tempering temperature of 500℃ to 700℃, and tempering time of 60 to 180 minutes, and also include the final hardness value. Before inputting these data into the model, they are standardized to normalize each parameter within the corresponding reasonable range, which helps to improve the training efficiency and stability of the model.

[0041] During the training process, considering that the submodule needs to adapt to the parameter prediction requirements of gears of different materials (covering 20CrMo, 20CrMnTi, 18Cr2Ni4WA, 20MnCr5), types (such as worm gears, cylindrical gears, bevel gears) and sizes under heat treatment processes such as carburizing, quenching, and tempering, the mean square error (MSE) is selected as the evaluation indicator. The mean square error is a commonly used loss function. By minimizing this loss function, the model performance can be optimized to minimize the error between the model's predicted value and the true value. In terms of optimizing model parameters, the Adam optimization algorithm is used. This algorithm has the advantage of adaptive learning rate and can automatically adjust the learning rate according to the first-order moment estimation and second-order moment estimation of the gradient during training, thereby accelerating the convergence speed. The training rounds are set to 3000 times, the initial value of the learning rate is 0.001, and according to the training process, the learning rate is exponentially decayed every 500 rounds to ensure that the model parameters can be adjusted more finely in the later stage of training.

[0042] The core function of the part heat treatment process parameter prediction submodule is to predict the required heat treatment process parameters (time and temperature of carburizing, quenching, and tempering) for parts of different materials and types, given the expected hardness target value by the user. When the user inputs information such as the material, type, and target hardness of the part, this information will be passed as input to the trained deep learning model, and then the corresponding heat treatment process parameter results will be obtained.

[0043] The part heat treatment process parameter prediction submodule also has the ability of self-update and reinforcement learning. When new gear samples and heat treatment data are input, they will be integrated into the existing training data set, the model will be retrained, and the model parameters will be updated to enhance the generalization prediction ability of part heat treatment process parameters under different working conditions. This ensures that the model can continuously learn and improve with the addition of new data, thereby improving its accuracy and reliability in different application scenarios.

[0044] The part heat treatment process parameter prediction submodule interacts with the experimental operation submodule and the experimental AI pilot module respectively. Specifically, the part heat treatment process parameter prediction submodule receives information such as part material, type, size and target hardness from the experimental operation submodule, and then feeds back the predicted heat treatment process parameter results to the experimental operation submodule and the experimental AI pilot module to provide accurate process parameter guidance for the experimental operation. For example, the user can input the basic information and hardness target of the gear through the experimental operation submodule, and the part heat treatment process parameter prediction submodule will predict the appropriate heat treatment process parameters. The user can perform actual operations in the experimental operation submodule based on these prediction results. At the same time, the part heat treatment process parameter prediction submodule can also assist in solving users' questions about part heat treatment process parameters, receive process parameter consultation instructions input by users, and output recommended heat treatment process parameter values ​​after parsing the process parameter consultation instructions. For example, when a user is uncertain about certain process parameters, he can consult the part heat treatment process parameter prediction submodule, which will give corresponding suggestions based on the knowledge and experience gained from its training.

[0045] The facial feature deep learning modeling submodule is used to extract key features of the original facial image data obtained by the digital human face customization submodule through a convolutional neural network, construct a facial three-dimensional model and output it to the digital human face customization submodule. The convolutional neural network (CNN) architecture used by the facial feature deep learning modeling submodule includes an input layer, three hidden layers, and an output layer. Specifically, the first hidden layer contains 128 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 512 neurons. Between these hidden layers, the ReLU function is used as a nonlinear activation function. The ReLU function enables the network to better learn nonlinear relationships, thereby improving the expressiveness of the model.

[0046] The process of training the convolutional neural network used in the facial feature deep learning modeling submodule includes the following steps:

[0047] Collect a number of facial images covering different genders, ages, races, expressions, facial postures, and multiple angles including front, side, upward, and downward views to form a facial image dataset;

[0048] Perform preprocessing operations on each facial image, including size cropping, normalization, and grayscale processing;

[0049] The convolutional neural network is trained using the preprocessed data set, the training rounds and initial learning rate are set, and the network is optimized using the mean square error as the loss function.

[0050] Specifically, in terms of training data preparation, a dataset of 10,000 facial images was used. This dataset is rich in diversity, covering facial images of different genders, ages, races, expressions, and facial postures. These images include facial feature information from multiple angles, such as front, side, upward, and downward, and in order to ensure the clarity and detail of the image, the resolution is no less than 1024 pixels × 1024 pixels.

[0051] A series of preprocessing operations are performed on the input facial images. The first is image cropping, which crops the image to a uniform size (256×256 pixels). This is done to ensure that the size of all input images is consistent, which is convenient for subsequent processing and training; the second is normalization, which normalizes the pixel value range to the [0,1] interval, so that the pixel values ​​of different images are in the same numerical range, avoiding the adverse effects of differences in brightness, contrast, etc. on model training; the last is grayscale processing, which converts color images into grayscale images, which helps to reduce the amount of calculation and extract more representative features, because grayscale images only contain brightness information, which can highlight the structural features of the face in some cases and eliminate the interference of color factors.

[0052] During the training process, in order to evaluate the accuracy of the model in facial feature extraction and three-dimensional model construction, the mean absolute error (MAE) is used as an evaluation indicator. MAE can intuitively reflect the average difference between the predicted value and the true value. By minimizing this loss function, the model performance can be optimized and the model's prediction results can be closer to the actual situation. When optimizing the model parameters, the RMSProp optimization algorithm is used. This algorithm is based on the gradient change characteristics of facial image data during the training process and can dynamically adjust the parameter update step size to meet the training needs of different stages. The training rounds are set to 5000 times, the initial learning rate is set to 0.001, and the learning rate is decayed according to the exponential law of 1000 rounds per cycle. This setting allows the model to learn faster in the early stage of training, and the parameters can be adjusted more finely in the later stage to achieve better convergence effect.

[0053] The core function of the facial feature deep learning modeling submodule is to extract key facial features from the input facial image. These key features include the position, shape, texture, and facial contour of the facial features. After the features are extracted, they are stored in a feature vector, which is a data structure that is convenient for storing and processing feature information. Subsequently, based on these stored features, the facial feature deep learning modeling submodule will construct a three-dimensional model of the face to provide richer facial information for subsequent applications.

[0054] There is data interaction between the facial feature deep learning modeling submodule and the digital human face customization submodule. The facial feature deep learning modeling submodule receives the original facial image data from the digital human face customization submodule, which is the starting point of the entire data flow. The original facial image data will go through the above-mentioned preprocessing and model processing process. After the processing is completed, the constructed facial 3D model is transmitted to the digital human face customization submodule, thereby providing a facial 3D model foundation for the creation of virtual characters, providing key support for the personalized customization of digital human faces, and making the created virtual characters have more realistic facial features.

[0055] Furthermore, if Figure 2 As shown in the figure, the digital workshop includes a heat treatment area, a sample preparation area, and a performance test area. The heat treatment area includes five virtual experimental devices: a box furnace, two oil tanks, two water tanks, a carburizing furnace, and a carburizing furnace hanger. The sample preparation area includes two virtual experimental devices: a metallographic sample polishing machine and a sample and reagent storage cabinet. The performance test area includes two virtual experimental devices: an electron microscope and a Rockwell hardness tester.

[0056] When the user operates the digital human to move near any of the virtual experimental equipment, including the box furnace, metallographic specimen polishing machine, carburizing furnace, Rockwell hardness tester and electron microscope, a pop-up window will be triggered on the user interface, such as a pop-up window in the upper right corner of the screen. The pop-up window is used to display equipment-related information, operation methods, precautions, etc.

[0057] The following only takes the hardness test link in the experimental operation process as an example, and uses gears as experimental parts to explain the process of how the experimental operation submodule guides users to complete the preset experimental operation process through interactive simulation. Figure 4 As shown in the figure, the hardness test process uses the Rockwell hardness tester in the performance test area of ​​the digital workshop, which specifically includes the following steps:

[0058] S1: Obtain the gear-clicking command input by the user, and place the gear into the test bench of the Rockwell hardness tester;

[0059] S2: Obtain the user's input of a click height adjustment bolt instruction to raise the test bench for preloading;

[0060] S3: Obtain the click loading handle command input by the user and apply a preset load to the gear;

[0061] S4: Obtaining the click unloading handle command input by the user, unloading the load applied to the gear, and displaying the hardness value on the dial of the Rockwell hardness tester;

[0062] S5: Obtain the test bench click command input by the user, so that the test bench descends to the original position, the gear is removed, and the hardness test experiment is exited.

[0063] The virtual simulation system of gas carburizing and performance testing that integrates deep learning proposed by the present invention realizes the accurate prediction of the heat treatment process parameters of carburizing experimental parts through the multi-task prediction and modeling module based on deep learning, and provides personalized teaching guidance in combination with the experimental AI pilot module empowered by the knowledge graph. At the same time, a highly free immersive experimental environment is constructed through the digital scene multivariate change module and the digital human customization module. The system integrates multiple functions in one, is easy to operate and free, and provides an innovative and efficient virtual platform for the teaching and research of gas carburizing and performance testing. Through artificial intelligence and virtual simulation technology, the present invention can simulate the experimental process of carburizing treatment of parts in real scenarios, thereby improving the user's familiarity and intuitive feeling of the high-risk experimental operation process.

[0064] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A virtual simulation system for gas carburizing and performance testing integrating deep learning, characterized in that: It includes a digital scene multi-variable change module, a digital human customization module, a teaching instruction control module, a comprehensive experimental teaching evaluation and achievement display module, an experimental AI pilot module, and a multi-task prediction and modeling module based on deep learning; The digital scene multivariate change module is used to establish a personalized digital workshop according to preset experimental requirements. The module includes an experimental equipment autonomous selection submodule and a workshop background change submodule, wherein the experimental equipment autonomous selection submodule is used to provide virtual experimental equipment according to experimental requirements, and the virtual experimental equipment includes a carburizing furnace and a carburizing furnace hanger; the workshop background change submodule is used to provide preset scenes of different styles for the digital workshop and provide different wall, floor, table and chair styles for the sample preparation area or performance test area in the digital workshop; The digital human customization module is used to customize the image of the experimental subject in the workshop, and the module includes a digital human body shape construction submodule, a digital human face customization submodule and a digital human image style submodule, wherein the digital human body shape construction submodule has multiple preset body templates built in, and the preset body templates include height, leg length, arm length, strength, neck length, and body posture parameters that can be customized; the digital human face customization submodule obtains original facial image data, and outputs the original facial image data to the facial feature deep learning modeling submodule in the multi-task prediction and modeling module based on deep learning for digital modeling, and then customizes the head and facial appearance features of the digital human according to the constructed facial three-dimensional model returned by the facial feature deep learning modeling submodule; The digital human image style submodule has multiple preset clothing templates built in, and the preset clothing templates include customizable digital human tops, pants, shoes, and accessories parameters; The teaching instruction control module includes an experiment preview submodule, an experiment operation submodule and an experiment assessment submodule, wherein the experiment preview submodule and the experiment assessment submodule respectively interact with the experiment AI pilot module for data, wherein the experiment preview submodule is used to conduct basic knowledge assessment on users, and feed back each assessment result of each user to the experiment AI pilot module, and is also used to track the user's preview progress and visualize the preview progress; The experimental operation submodule is used to guide the user to complete the preset experimental operation process through interactive simulation after obtaining the part material and experimental parts selected by the user, and to guide the user's operation in real time through one or more combinations of highlighted display areas, text prompt boxes, audio instruction broadcasts and video animation playback during the experimental execution. At the same time, the experimental operation submodule also interacts with the part heat treatment process parameter prediction submodule in the multi-task prediction and modeling module based on deep learning for data exchange. The part heat treatment process parameter prediction submodule sends the predicted heat treatment process parameter results to the experimental operation submodule and the experimental AI navigation module respectively, providing users with Provide process parameter guidance, wherein the experimental operation process includes seven experimental process links, namely, hardness test before carburizing, carburizing experiment, carburized layer detection, quenching experiment, tempering experiment, hardness test after carburizing, and metallographic experiment; the experimental assessment submodule is used to assess and evaluate the user's familiarity with the experimental operation process and related knowledge according to the customized questions output by the experimental AI navigation module, generate assessment results, and judge the correctness of the user's operation through the experimental operation submodule according to the real experimental process, generate operation judgment results, and feed back the assessment results and the operation judgment results to the experimental teaching comprehensive evaluation and achievement display module; The experimental teaching comprehensive evaluation and achievement display module is used to display the assessment results and the operation evaluation results, and output the user's knowledge weaknesses and operation errors, and output the experimental result table; The experimental AI pilot module is used to assist the teaching instruction control module, including analyzing the user's knowledge gaps according to the collected assessment results of the experimental preview submodule and accurately customizing key review knowledge and related knowledge assessment questions, and is also used to receive question instructions output by the user, and realize intelligent response to question instructions through natural language processing technology and knowledge graph technology and display intelligent response results; The multi-task prediction and modeling module based on deep learning includes the facial feature deep learning modeling submodule and the part heat treatment process parameter prediction submodule. The facial feature deep learning modeling submodule is used to extract key features of the original facial image data through a convolutional neural network, construct a facial three-dimensional model and output it to the digital human face customization submodule; the part heat treatment process parameter prediction submodule is used to predict the heat treatment process parameters using a convolutional neural network after obtaining the material, type and target hardness information of the part, and obtain predicted heat treatment process parameter results. The heat treatment process parameter results include carburizing parameters, quenching parameters, tempering time and temperature parameters, and are also used to receive process parameter consultation instructions input by the user, parse the process parameter consultation instructions and output recommended heat treatment process parameter values.

2. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to claim 1, characterized in that: The convolutional neural network used in the facial feature deep learning modeling submodule includes an input layer, three hidden layers and an output layer, wherein the first hidden layer contains 128 neurons, the second hidden layer contains 256 neurons, and the third hidden layer contains 512 neurons, and the ReLU function is used as a nonlinear activation function between these hidden layers.

3. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to claim 2, characterized in that: The process of training the convolutional neural network used in the facial feature deep learning modeling submodule includes the following steps: Collect a number of facial images covering different genders, ages, races, expressions, facial postures, and multiple angles including front, side, upward, and downward views to form a facial image dataset, and the resolution of each facial image is not less than 1024 pixels × 1024 pixels; Performing preprocessing operations on each of the facial images, including size cropping, normalization processing, and grayscale processing; The convolutional neural network was trained using the preprocessed dataset. The training rounds were set to 5000 times, the initial learning rate was 0.001, and the learning rate was decayed according to the exponential law with 1000 rounds as a cycle. The network was optimized using the mean square error as the loss function.

4. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to claim 1, characterized in that: The convolutional neural network used in the part heat treatment process parameter prediction submodule includes an input layer, three hidden layers and an output layer, wherein the first hidden layer contains 64 neurons, the second hidden layer contains 128 neurons, and the third hidden layer contains 256 neurons, and the ReLU function is used as a nonlinear activation function between these hidden layers.

5. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to claim 4, characterized in that: The process of training the convolutional neural network used in the part heat treatment process parameter prediction submodule includes the following steps: Collect a number of sample data of parts that have been processed by different heat treatment processes to build a training data set, wherein the sample data of the parts includes the material, type, heat treatment process parameters and the final hardness value of the parts; Standardize all parts sample data; The convolutional neural network was trained using the standardized data set. The training rounds were set to 3000 times, the initial learning rate was 0.001, and the learning rate was exponentially decayed every 500 rounds according to the training progress. The mean square error was used as the loss function to optimize the network.

6. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to any one of claims 1 to 5, characterized in that: The quenching experiment includes two quenching methods: direct quenching and one-shot quenching. After the experiment, the metallographic test results and hardness test results obtained by the two quenching methods are compared and analyzed.

7. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to any one of claims 1 to 5, characterized in that: The experimental operation submodule combines water flow simulation, flame effect, gas flow display and sound prompts for joint demonstration during the experimental operation.

8. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to any one of claims 1 to 5, characterized in that: The digital workshop also includes a heat treatment area, which includes five virtual experimental equipment: a box furnace, two oil tanks, two water tanks, a carburizing furnace and a carburizing furnace hanger; The sample preparation area includes two virtual experimental equipments, a metallographic sample polishing machine and a sample and reagent storage cabinet; The performance testing area includes two virtual experimental equipments: an electron microscope and a Rockwell hardness tester.

9. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to claim 8, characterized in that: When the user operates the digital human to move near any of the virtual experimental equipment, including the box furnace, metallographic sample polishing machine, carburizing furnace, Rockwell hardness tester and electron microscope, a pop-up window is triggered on the user interface to display equipment-related information, operation methods and precautions.

10. A virtual simulation system for gas carburizing and performance testing integrating deep learning according to any one of claims 1 to 5, characterized in that: The part material is any one of 20CrMo, 20CrMnTi, 18Cr2Ni4WA, and 20MnCr5, and the experimental part is any one of a worm gear, a cylindrical gear, and a bevel gear.

Images