A hot isostatic pressing diffusion welding parameter optimization method and system based on deep learning, an electronic device, and a storage medium
By constructing a physical model of hot isostatic diffusion and using deep learning methods, combined with an error compensation model and a PID controller, the welding parameters of hot isostatic diffusion are optimized. This solves the problems of parameter adjustment relying on experience and lacking dynamic optimization in traditional methods, and achieves efficient and accurate welding parameter optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUAINAN NEW ENERGY RES CENT
- Filing Date
- 2025-10-30
- Publication Date
- 2026-04-10
AI Technical Summary
Traditional hot isostatic pressure diffusion welding technology relies on the experience of technicians or trial-and-error experiments, lacks comprehensive consideration of the interaction between multiple parameters, and requires actual measurement to evaluate the performance of the welded joint, lacking pre-prediction and dynamic optimization methods.
A thermal isostatic diffusion physical model was constructed, and welding parameters were optimized using deep learning methods. Iterative optimization was carried out by improving the MLP model and PID controller, and combined with an error compensation model, to achieve dynamic optimization of welding parameters.
It enables precise optimization of welding parameters, reduces material and time losses, improves the accuracy and efficiency of welding quality prediction, and reduces reliance on human experience.
Smart Images
Figure CN121351402B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of heterogeneous material diffusion welding, in particular to a hot isostatic pressing diffusion welding parameter optimization method and system based on deep learning, an electronic device and a storage medium. BACKGROUND
[0002] Hot isostatic pressing, referred to as HIP. Tungsten copper and chromium zirconium copper are composite structural materials commonly used in the divertor of EAST and other fusion devices due to their excellent thermal conductivity, electrical conductivity and mechanical properties. However, the traditional welding method has problems such as loose structure, large interface thermal resistance and micro-cracks at the welded joint, which seriously affect the service life. HIP has become an ideal connection process due to its high density and interface bonding strength. In the field of heterogeneous material diffusion welding, existing patents on hot isostatic pressing diffusion welding mainly focus on the methods and steps of hot isostatic pressing diffusion welding of different metals, and few of them are systematic optimization of process parameters for hot isostatic pressing diffusion welding.
[0003] However, the existing hot isostatic pressing diffusion welding technology has obvious limitations: the hot isostatic pressing welding parameters of tungsten copper and chromium zirconium copper are mostly determined depending on the experience of technicians or a large number of trial-and-error experiments. During parameter adjustment, only a single or a few parameters such as temperature, pressure and holding time are tested, and there is a lack of comprehensive consideration of the interaction between multiple parameters. At the same time, after welding, joint performance evaluation needs to be measured by tensile strength detection, microstructure observation and other experimental methods after welding is completed, and the whole process lacks effective pre-prediction and dynamic optimization means. SUMMARY
[0004] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a hot isostatic pressing diffusion welding parameter optimization method and system based on deep learning, an electronic device and a storage medium, which solves the problem of excessive reliance on personnel experience or trial-and-error experiments and lack of dynamic optimization methods in traditional methods.
[0005] To achieve the above-mentioned purpose, the present application provides the following solutions:
[0006] A hot isostatic pressing diffusion welding parameter optimization method based on deep learning, comprising:
[0007] constructing a hot isostatic pressing diffusion physical model;
[0008] setting a plurality of experimental schemes, inputting each experimental scheme into the hot isostatic pressing diffusion physical model to obtain a theoretical prediction result;
[0009] performing a hot isostatic pressing diffusion welding experiment according to the experimental scheme to obtain an actual detection result;
[0010] extracting errors of the theoretical prediction result and the actual detection result to obtain theoretical actual error data, aligning and integrating the theoretical actual error data, the theoretical prediction result and the experiment scheme to obtain original training data;
[0011] training the improved MLP model by using the original training data to obtain an error compensation model;
[0012] fusing the hot isostatic pressing diffusion physical model and the error compensation model to obtain a result prediction model;
[0013] setting a parameter range, randomly selecting a set of schemes in the parameter range and inputting the set of schemes into the result prediction model to obtain a candidate prediction result;
[0014] iterating the schemes input into the result prediction model by using a preset PID controller until the candidate prediction result meets a preset requirement to stop iteration, and obtaining hot isostatic pressing diffusion welding optimization parameters.
[0015] Preferably, the hot isostatic pressing diffusion physical model is constructed, including:
[0016] constructing an interface atomic diffusion equation; the interface atomic diffusion equation includes: and ; wherein, is an atomic diffusion flux; is a diffusion coefficient; is an element concentration value; denotes a distance; denotes time;
[0017] constructing a temperature dependence equation; the temperature dependence equation is: ; wherein, is a diffusion constant; is a diffusion activation energy; is a gas constant; is an absolute temperature;
[0018] constructing a densification equation; the densification equation is: ; wherein, is a relative density; is a material constant; is a hot isostatic pressing pressure; , are respectively a first fitting index and a second fitting index;
[0019] constructing a free energy equation; the free energy equation is: ; wherein, is a free energy change; is an enthalpy change; is an entropy change;
[0020] Fusing the interface atom diffusion equation, the temperature dependence equation, the densification equation, and the free energy equation obtains the hot isostatic pressing diffusion physical model.
[0021] Preferably, a plurality of sets of experimental schemes are set, each of the experimental schemes is input into the hot isostatic pressing diffusion physical model to obtain theoretical prediction results, including:
[0022] Selecting to-be-controlled process parameters; the to-be-controlled process parameters include: hot isostatic pressing temperature, hot isostatic pressing pressure, holding time, cooling rate, heating rate;
[0023] Determining material attribute parameters; the material attribute parameters include: diffusion constant, diffusion activation energy, and material constant;
[0024] Orthogonal combination processing is performed on the set to-be-controlled process parameters and the material attribute parameters to obtain a plurality of sets of experimental schemes;
[0025] The experimental schemes are input into the hot isostatic pressing diffusion physical model to obtain the theoretical prediction results; the theoretical prediction results include: tensile strength, thermal diffusivity, porosity, and residual stress.
[0026] Preferably, hot isostatic pressing diffusion welding experiments are performed according to the experimental schemes to obtain actual detection results, including:
[0027] Selecting a base material matched with the experimental scheme parameters;
[0028] Sequentially performing cutting, grinding, ultrasonic cleaning, deionized water washing, and drying processing on the base material to obtain an experimental sample;
[0029] Calibrating temperature, pressure, and rate of a hot isostatic pressing device;
[0030] After a pre-set tooling and the experimental sample are placed into a furnace cavity of the hot isostatic pressing device, performing vacuumizing processing on the furnace cavity;
[0031] According to a heating rate provided by the experimental scheme, controlling the hot isostatic pressing device to perform heating, and when the temperature in the furnace cavity meets the requirements of the experimental scheme, performing hot isostatic pressing processing on the experimental sample;
[0032] In a holding and pressure maintaining stage, performing diffusion welding on the experimental sample at a fixed temperature and pressure, and after the holding and pressure maintaining stage ends, according to the experimental scheme, controlling the furnace cavity to perform cooling and pressure reducing processing to obtain a welded sample;
[0033] The welding sample is subjected to mechanical property detection, thermophysical property detection, microscopic quality detection and stress state detection respectively to obtain the actual detection results.
[0034] Preferably, the improved MLP model comprises an input layer, a weighted preprocessing layer, a hidden layer and an output layer connected in sequence.
[0035] Preferably, the scheme input into the result prediction model is iterated by using a preset PID controller until the candidate prediction result meets the preset requirement, and the heat isopressing diffusion welding optimization parameter is obtained, comprising:
[0036] The candidate prediction result is set as a control target, and a PID input variable is set according to the control target; the expression of the PID input variable is: ; wherein, is a control deviation in the i th iteration; is a target index number; is a weight of the j th index; is the j th index output by the result prediction model in the i th iteration; is a preset target value of the j th index;
[0037] The PID controller is constructed; the expression of the PID controller is:
[0038] ;
[0039] wherein, is a parameter adjustment amount; , , are a proportional coefficient, an integral coefficient and a differential coefficient respectively; is a deviation integral in the i th iteration; is a deviation change rate; It is judged whether the PID input variable meets the preset requirement or not to obtain a deviation degree state;
[0040] When the deviation degree state does not meet the requirement, the PID input variable is input into the PID controller to obtain the parameter adjustment amount;
[0041]
[0042] updating the parameter adjustment amount into the scheme of the result prediction model, and returning to the step of "judging whether the PID input variable meets the preset requirement to obtain a deviation degree state", until the deviation degree state is the meeting requirement, outputting the last updated scheme of the result prediction model to obtain the hot isostatic pressing diffusion welding optimization parameter.
[0043] Preferably, the improved MLP model is trained using the original training data to obtain an error compensation model, comprising:
[0044] building the improved MLP model;
[0045] receiving the original training data through the input layer;
[0046] using the weighted preprocessing layer to perform pairwise weighted calculation on the data in each group of data in the original training data to obtain weighted interaction data; the expression of the weighted preprocessing layer is: ; wherein, is the weighted interaction data of the i-th data and the j-th data in the same group of data of the original training data; is the weighted interaction data of the i-th data and the j-th data in the same group of data of the original training data; is a learnable interaction weight matrix; , , are the i-th data and the j-th data in the same group of data of the original training data, respectively; synchronously inputting the weighted interaction data and the original training data into two layers of full connection network connected in turn in the hidden layer for non-linear mapping to obtain extracted features;
[0047] inputting the extracted features into the output layer for multi-index error prediction to obtain model prediction results;
[0048] performing back propagation iterative optimization on the hidden layer and the weighted preprocessing layer according to the model prediction results and the theoretical actual error data to obtain the trained error compensation model.
[0049] Preferably, a hot isostatic pressing diffusion welding parameter optimization system based on deep learning, comprising:
[0050] a physical model building module for building a hot isostatic pressing diffusion physical model;
[0051] a theoretical data collection module for setting a plurality of experimental schemes, inputting each of the experimental schemes into the hot isostatic pressing diffusion physical model to obtain a theoretical prediction result;
[0052] a theoretical data collection module for setting a plurality of experimental schemes, inputting each of the experimental schemes into the hot isostatic pressing diffusion physical model to obtain a theoretical prediction result;
[0053] An actual data collection module is configured to perform a hot isostatic pressing diffusion welding experiment according to the experiment scheme to obtain actual detection results.
[0054] A training data construction module is configured to extract errors of the theoretical prediction results and the actual detection results to obtain theoretical actual error data, and align and integrate the theoretical actual error data, the theoretical prediction results and the experiment scheme to obtain original training data.
[0055] A compensation model construction module is configured to train an improved MLP model by using the original training data to obtain an error compensation model.
[0056] A model fusion module is configured to fuse the hot isostatic pressing diffusion physical model and the error compensation model to obtain a result prediction model.
[0057] A result prediction module is configured to set a parameter range, randomly select a set of schemes in the parameter range and input the set of schemes into the result prediction model to obtain candidate prediction results.
[0058] A parameter optimization module is configured to use a preset PID controller to iteratively input schemes into the result prediction model until the candidate prediction results meet preset requirements to stop iteration and obtain hot isostatic pressing diffusion welding optimization parameters.
[0059] Preferably, an electronic device comprises at least one processor and a memory connected with the processor in communication; the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to implement the aforementioned hot isostatic pressing diffusion welding parameter optimization method based on deep learning.
[0060] Preferably, a non-transitory computer readable storage medium stores computer instructions for enabling a computer to implement the aforementioned hot isostatic pressing diffusion welding parameter optimization method based on deep learning.
[0061] The present application discloses the following technical effects:
[0062] The present application provides a hot isostatic pressing diffusion welding parameter optimization method and system based on deep learning, an electronic device and a storage medium, which solves the defect of large prediction error of a traditional linear fitting model by constructing a hot isostatic pressing diffusion physical model and an error compensation model, realizes correction of a physical model deviation, improves the model compensation precision by improving an MLP model, solves the problem of lack of consideration of data interaction characteristics in a traditional MLP model, and realizes rapid search for optimal process parameter combination by using a result prediction model through a PID controller, solves the problem of excessive dependence on artificial experience or high trial and error cost in a traditional method, and realizes rapid search for optimal process parameter combination by using a result prediction model. BRIEF DESCRIPTION OF DRAWINGS
[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.
[0064] Figure 1 A deep learning-based hot isostatic pressing diffusion welding parameter optimization process schematic diagram is provided for the embodiments of the present application.
[0065] Figure 2 A step flowchart is provided for the embodiments of the present application.
[0066] Figure 3 A physical model construction flowchart is provided for the embodiments of the present application.
[0067] Figure 4 A PID parameter optimization flowchart is provided for the embodiments of the present application.
[0068] Figure 5 A network architecture diagram is provided for the embodiments of the present application. DETAILED DESCRIPTION
[0069] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0070] The purpose of the present application is to provide a deep learning-based hot isostatic pressing diffusion welding parameter optimization method, system, electronic device and storage medium, which solves the problem that the traditional method excessively relies on personnel experience or trial and error experiments and lacks dynamic optimization methods.
[0071] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0072] Figure 1 A deep learning-based hot isostatic pressing diffusion welding parameter optimization process schematic diagram is provided for the embodiments of the present application, Figure 2 A step flowchart is provided for the embodiments of the present application, as shown in Figure 1 and Figure 2 The present application provides a deep learning-based hot isostatic pressing diffusion welding parameter optimization method, which comprises:
[0073] Step 100: constructing a hot isostatic pressing diffusion physical model;
[0074] Step 200: setting a plurality of experimental schemes, inputting each of the experimental schemes into the hot isostatic pressing diffusion physical model to obtain a theoretical prediction result;
[0075] Step 300: performing a hot isostatic pressing diffusion welding experiment according to the experimental scheme to obtain an actual detection result;
[0076] Step 400: extracting an error of the theoretical prediction result and the actual detection result to obtain theoretical actual error data, aligning and integrating the theoretical actual error data, the theoretical prediction result and the experimental scheme to obtain original training data;
[0077] Step 500: training an improved MLP model using the original training data to obtain an error compensation model;
[0078] Step 600: fusing the hot isostatic pressing diffusion physical model and the error compensation model to obtain a result prediction model;
[0079] Step 700: setting a parameter range, randomly selecting a scheme in the parameter range and inputting the scheme into the result prediction model to obtain a candidate prediction result;
[0080] Step 800: using a pre-set PID controller to iterate the scheme input into the result prediction model until the candidate prediction result meets a pre-set requirement to stop iteration, and obtaining a hot isostatic pressing diffusion welding optimization parameter.
[0081] Reference Figure 3 , constructing a hot isostatic pressing diffusion physical model, comprising:
[0082] Step 101: constructing an interface atomic diffusion equation; the interface atomic diffusion equation comprises: and ; wherein, is an atomic diffusion flux; is a diffusion coefficient; is an element concentration value; denotes distance; denotes time;
[0083] Step 102: constructing a temperature-dependent equation; the temperature-dependent equation is: ; wherein, is a diffusion constant; is a diffusion activation energy; is a gas constant; is an absolute temperature;
[0084] Step 103: constructing a densification equation; the densification equation is: ; wherein, is a relative density; is a material constant; is a hot isostatic pressing pressure; , are respectively a first fitting index, a second fitting index;
[0085] Step 104: constructing a free energy equation; the free energy equation is: ; wherein, is a free energy change; is an enthalpy change; is an entropy change;
[0086] Step 105: fusing the interface atom diffusion equation, the temperature dependence equation, the densification equation, and the free energy equation to obtain the hot isostatic pressing diffusion physical model.
[0087] Specifically, a plurality of sets of experimental schemes are set, each of the experimental schemes is input into the hot isostatic pressing diffusion physical model to obtain a theoretical prediction result, including:
[0088] selected to be controlled process parameters; the process parameters to be controlled include: hot isostatic pressing temperature, hot isostatic pressing pressure, holding time, cooling rate, heating rate;
[0089] determining material attribute parameters; the material attribute parameters include: diffusion constant, diffusion activation energy, material constant;
[0090] orthogonal combination processing is performed on the set process parameters to be controlled and the material attribute parameters to obtain a plurality of sets of experimental schemes;
[0091] the experimental scheme is input into the hot isostatic pressing diffusion physical model to obtain the theoretical prediction result; the theoretical prediction result includes: tensile strength, thermal diffusivity, porosity, residual stress.
[0092] Further, according to the experimental scheme, a hot isostatic pressing diffusion welding experiment is performed to obtain an actual detection result, including:
[0093] selecting a base material matched with the experimental scheme parameters;
[0094] cutting, grinding, ultrasonic cleaning, deionized water washing, and drying the base material in sequence to obtain an experimental sample;
[0095] calibrating the temperature, pressure, and rate of the hot isostatic pressing equipment;
[0096] After the preset tooling and the experimental sample are placed into the furnace cavity of the hot isostatic pressing device, the furnace cavity is subjected to vacuumizing treatment;
[0097] The heating rate provided by the experimental scheme controls the heating of the hot isostatic pressing device, and when the temperature in the furnace cavity meets the requirements of the experimental scheme, the experimental sample is subjected to hot isostatic pressing treatment;
[0098] The experimental sample is subjected to diffusion welding at a fixed temperature and pressure in the holding and pressure maintaining stage, and after the holding and pressure maintaining stage ends, the furnace cavity is controlled to be subjected to temperature and pressure reduction treatment according to the experimental scheme, thereby obtaining a welded sample;
[0099] The welded sample is subjected to mechanical property detection, thermophysical property detection, microscopic quality detection, and stress state detection, respectively, thereby obtaining the actual detection results.
[0100] Preferably, the improved MLP model comprises an input layer, a weighted preprocessing layer, a hidden layer, and an output layer connected in sequence.
[0101] Reference Figure 4 , the scheme input into the result prediction model is iterated by using a preset PID controller until the candidate prediction result meets the preset requirements, and the iteration is stopped, thereby obtaining hot isostatic pressing diffusion welding optimization parameters, including:
[0102] Step 801: The candidate prediction result is set as a control target, and a PID input variable is set according to the control target; the expression of the PID input variable is: ; wherein, is a control deviation at the i-th iteration; is the number of target indicators; is the weight of the j-th indicator; is the j-th indicator output by the result prediction model at the i-th iteration; is the preset target value of the j-th indicator; Step 802: The PID controller is constructed; the expression of the PID controller is: ;
[0103] Step 802: The PID controller is constructed; the expression of the PID controller is:
[0104] ;
[0105] wherein, is a parameter adjustment amount; , , are proportional coefficient, integral coefficient, and differential coefficient, respectively; is a deviation integral of the secondary iteration; a deviation change rate;
[0106] Step 803: determining whether the PID input variable meets the preset requirement to obtain a deviation degree state;
[0107] Step 804: when the deviation degree state does not meet the requirement, inputting the PID input variable into the PID controller to obtain a parameter adjustment amount;
[0108] Step 805: updating the parameter adjustment amount into a scheme of the result prediction model, and returning to step “determining whether the PID input variable meets the preset requirement to obtain a deviation degree state” until the deviation degree state meets the requirement, outputting a last updated scheme of the result prediction model to obtain the hot isostatic pressing diffusion welding optimization parameter.
[0109] Reference Figure 5 , training the improved MLP model by using the original training data to obtain an error compensation model, comprising:
[0110] constructing the improved MLP model;
[0111] receiving the original training data through the input layer;
[0112] performing pairwise weighted calculation on data in each group of data in the original training data by using the weighted preprocessing layer to obtain weighted interaction data; an expression of the weighted preprocessing layer is: ; wherein, is the weighted interaction data of the i-th data and the j-th data in the same group of data of the original training data; is the weighted interaction data of the i-th data and the j-th data in the same group of data of the original training data; is a learnable interaction weight matrix; , , are the i-th data and the j-th data in the same group of data of the original training data, respectively; synchronously inputting the weighted interaction data and the original training data into two layers of full connection networks connected in sequence in the hidden layer to perform nonlinear mapping to obtain extracted features;
[0113] inputting the extracted features into the output layer to perform multi-index error prediction to obtain a model prediction result;
[0114]
[0115] The hidden layer and the weighted pretreatment layer are iteratively optimized in a backward propagation manner according to the model prediction result and the theoretical actual error data, so that the error compensation model is trained.
[0116] Specifically, the embodiment provides a hot isostatic pressing diffusion welding parameter optimization method. First, a hot isostatic pressing diffusion physical model is constructed, then a plurality of experiments are set, errors of physical model outputs and actual measured values are extracted, and an error compensation model is constructed in this way; a result prediction model is obtained by fusing the physical model and the compensation model, and a PID controller is used to iteratively optimize parameters to find an optimal parameter combination in the subsequent process. In the model construction process, not only welding process related parameters are considered, but also material properties are considered.
[0117] Preferably, the hot isostatic pressing diffusion welding mainly promotes interfacial atomic diffusion of materials to be welded, pore elimination and joint densification through the synergistic effect of high temperature and isostatic pressure. Based on this, the physical model part of the embodiment integrates diffusion kinetics, interfacial thermodynamics, pressure-densification coupling and the like, and realizes theoretical prediction of welding effect in combination with process parameters and material properties. Before simulation, the values of some parameters need to be determined, including diffusion activation energy, diffusion constant, welding temperature, holding time, initial interfacial concentration, interfacial roughness and the like. With reference to existing literature, the expressions of the HIP diffusion welding physical model constructed in the embodiment include:
[0118] Fick diffusion law:
[0119]
[0120]
[0121] In the formula, J is an atomic diffusion flux; D is a diffusion coefficient; C is an element concentration value; r represents a distance; t represents time. Arrhenius equation:
[0122]
[0123] In the formula, D is a diffusion constant;
[0124] E is a diffusion activation energy; R is a gas constant; T is an absolute temperature. HIP densification equation:
[0125]
[0126]
[0127] wherein, is the relative density; is the material constant; is the hot isostatic pressing pressure; , is the first fitting index, and the second fitting index, respectively.
[0128] Gibbs free energy equation:
[0129]
[0130] wherein, is the free energy change; is the enthalpy change; is the entropy change.
[0131] Specifically, the following describes the hot isostatic pressing diffusion welding of tungsten copper-chromium zirconium copper as a specific embodiment. The process parameters related to the hot isostatic pressing diffusion welding are selected, and reasonable value ranges are set based on material properties. The process parameters include hot isostatic pressing temperature, hot isostatic pressing pressure, holding time, heating rate, and cooling rate. The material attribute parameters of tungsten copper and chromium zirconium copper are determined, including diffusion constant, diffusion activation energy, and material constant. The experimental scheme is constructed by the orthogonal experimental design method, and the scheme is sequentially input into the hot isostatic pressing diffusion physical model for simulation, and the theoretical prediction results are output, including tensile strength, thermal diffusivity, porosity, and residual stress.
[0132] Further, the tungsten copper base material and the chromium zirconium copper base material matching the material attribute parameters in the experimental scheme are selected, and the base material size specifications are unified, with the size set to 50-100 mm in length, 30-50 mm in width, and 5-10 mm in thickness. The base material is cut by a wire cutting device, the welding interface of the sample to be welded is gradiently polished, 800 mesh, 1200 mesh, and 2000 mesh silicon carbide sandpaper are used in sequence, and the polishing is performed by a mechanical polisher until the interface roughness is not greater than 0.8 μm. The polished sample is placed in an ultrasonic cleaning tank, anhydrous ethanol is used as a cleaning agent, and ultrasonic cleaning is performed after setting the ultrasonic power and cleaning time to remove residual grinding dust and oil stains on the interface. The rinsed sample is placed in a vacuum drying oven and completely dried at a constant temperature.
[0133] Optionally, before conducting the experiment, the hot isostatic pressing (HIP) equipment needs to be precisely calibrated. Three to five temperature measuring points should be evenly arranged within the furnace cavity, with the calibration temperature range covering the HIP temperature set in the experimental design. A standard pressure sensor should be used to calibrate the pressure range covering the HIP pressure set in the experimental design. The heating and cooling rates should be calibrated using a temperature acquisition instrument, and the pressure increase rate should be calibrated using a pressure acquisition instrument. A graphite fixture compatible with the sample material should be selected, and the pretreated experimental sample should be fixed inside the fixture. The fixture and sample should be placed together into the HIP furnace cavity. After closing the furnace door, the vacuum system should be activated to evacuate the furnace cavity. The heating rate of the equipment was controlled according to the experimental plan. When the temperature inside the furnace reached 90% of the target hot isostatic pressure temperature, the pressure system was started, and the pressure was slowly increased to the target pressure at the set pressure increase rate. After the temperature and pressure reached the values set in the experimental plan, the heat preservation and pressure holding stage was entered. During this stage, the temperature and pressure of each temperature measuring point in the furnace were monitored in real time to ensure that the parameter fluctuations were within the calibration accuracy range. The heat preservation and pressure holding time was executed according to the experimental plan. After the heat preservation and pressure holding was completed, the temperature was reduced according to the cooling rate set in the experimental plan. After the temperature dropped below 50°C and the pressure dropped to normal pressure, the furnace was opened and the welded sample was taken out.
[0134] Furthermore, the error between the theoretical prediction results and the actual detection results is extracted, and this error data, theoretical prediction results, and experimental scheme are aligned and integrated to form the original training data.
[0135] Specifically, this embodiment improves upon the original MLP model. The improved MLP model in this embodiment includes an input layer, a weighted preprocessing layer, a hidden layer, and an output layer connected in sequence. The function of each network layer is as follows:
[0136] 1) Input layer: Receives and preprocesses the original training data, which includes experimental scheme parameters, theoretical prediction results, and theoretical and actual error data. Before input, the data needs to be processed by Min-Max normalization to map the feature values to the interval [0, 1].
[0137] 2) Weighted preprocessing layer: Used to extract the coupling relationships between different features in the original data. Through a learnable interaction weight matrix, weighted interaction calculations are performed on any two features of the same set of input data, as shown in the following formula:
[0138]
[0139] In the formula, The first data in the same set of original training data The data and the first Weighted interactive data of individual data; This is a learnable interaction weight matrix; , The first data point in the same set of original training data is the first... first data and the second data. first data and the second data.
[0140] All feature pairs are traversed to generate a weighted interaction data set.
[0141] 3) Hidden layer: composed of two layers of full connection networks connected in turn, the number of neurons of the first layer of full connection network is set to twice the sum of the original input feature number and the weighted interaction feature number, and the number of neurons of the second layer of full connection network is set to half of the first layer.
[0142] 4) Output layer: the number of neurons is consistent with the number of indicators of the theoretical actual error data (the number of neurons of the output layer of the embodiment is 4, corresponding to tensile strength, thermal diffusivity, porosity, residual stress respectively), and the final output is the multi-index error compensation value predicted by the model.
[0143] Preferably, the multi-dimensional welding quality indicators output by the result prediction model are control targets, including: tensile strength, thermal diffusivity, porosity and residual stress of tungsten-copper-chromium-zirconium copper welded joints; and the input variables of the PID are determined.
[0144]
[0145] In the formula, is the control deviation at the i-th iteration; is the target index number; is the weight of the j-th index; is the j-th index output by the result prediction model at the i-th iteration; is the preset target value of the j-th index. An independent PID sub-controller is designed for each parameter, and the PID output of a single parameter to be controlled is:
[0146] In the formula, is the parameter adjustment amount;
[0147] , , are proportional coefficient, integral coefficient and differential coefficient respectively;
[0148] is the deviation integral of the i-th iteration; is the rate of change of the deviation.
[0149] The initial values of the proportional coefficient, the integral coefficient and the differential coefficient are set based on the experience of the hot isostatic pressing process, and the parameter adjustment constraint, that is, the boundary of the parameter, is defined. If the comprehensive control deviation is less than the allowed maximum deviation and the comprehensive control deviation is stable below the allowed maximum deviation for three consecutive iterations, it is judged that the preset requirement is met, and the iteration is stopped. If the comprehensive control deviation is greater than the allowed maximum deviation, it is judged that the preset requirement is not met, and the parameter adjustment is required.
[0150] As an optional implementation, the embodiment also provides a hot isostatic pressing diffusion welding parameter optimization system based on deep learning, comprising:
[0151] A physical model construction module is configured to construct a hot isostatic pressing diffusion physical model.
[0152] A theoretical data collection module is configured to set a plurality of experimental schemes, input each of the experimental schemes into the hot isostatic pressing diffusion physical model, and obtain a theoretical prediction result.
[0153] An actual data collection module is configured to perform a hot isostatic pressing diffusion welding experiment according to the experimental schemes, and obtain an actual detection result.
[0154] A training data construction module is configured to extract an error of the theoretical prediction result and the actual detection result, obtain theoretical actual error data, and align and integrate the theoretical actual error data, the theoretical prediction result and the experimental schemes to obtain original training data.
[0155] A compensation model construction module is configured to train an improved MLP model by using the original training data, and obtain an error compensation model.
[0156] A model fusion module is configured to fuse the hot isostatic pressing diffusion physical model and the error compensation model, and obtain a result prediction model.
[0157] A result prediction module is configured to set a parameter range, randomly select a scheme in the parameter range and input the scheme into the result prediction model to obtain a candidate prediction result.
[0158] A parameter optimization module is configured to use a preset PID controller to iteratively input the scheme into the result prediction model until the candidate prediction result meets a preset requirement, and obtain a hot isostatic pressing diffusion welding optimization parameter.
[0159] As an optional implementation, the embodiment also provides an electronic device, comprising at least one processor and a memory connected with the processor in communication; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to perform the foregoing hot isostatic pressing diffusion welding parameter optimization method based on deep learning.
[0160] As an optional implementation, the embodiment also provides a non-transient computer readable storage medium storing computer instructions, the computer instructions being used for enabling a computer to execute the aforementioned deep learning based hot isostatic pressing diffusion welding parameter optimization method.
[0161] The beneficial effects of the present application are as follows:
[0162] The present application can accurately correct the theoretical deviation of the physical model by constructing a hot isostatic pressing diffusion physical model and an error compensation model, thereby reducing material loss and time loss; the degree of utilization of the original data is optimized by improving the MLP model, the coupling features between the data are extracted, and the compensation accuracy of the model is improved; the optimal process parameter combination can be quickly searched by the result prediction model through the PID controller, thereby reducing the parameter optimization difficulty.
[0163] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between the embodiments can be mutually referred to.
[0164] The principles and implementation modes of the present application are described by applying specific examples in the present application, and the above description of the embodiments is only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation mode and application range of the present application will be changed according to the idea of the present application. In conclusion, the content of the specification should not be understood as the limitation of the present application.
Claims
1. A deep learning-based hot isostatic pressing diffusion welding parameter optimization method, characterized in that, The method comprises the following steps: constructing a hot isostatic pressing diffusion physical model; setting a plurality of experimental schemes, inputting each of the experimental schemes into the hot isostatic pressing diffusion physical model to obtain theoretical prediction results; performing hot isostatic pressing diffusion welding experiments according to the experimental schemes to obtain actual detection results; extracting errors of the theoretical prediction results and the actual detection results to obtain theoretical actual error data, and aligning and integrating the theoretical actual error data, the theoretical prediction results and the experimental schemes to obtain original training data; training an improved MLP model by using the original training data to obtain an error compensation model; fusing the hot isostatic pressing diffusion physical model and the error compensation model to obtain a result prediction model; setting a parameter range, randomly selecting a scheme in the parameter range and inputting the scheme into the result prediction model to obtain a candidate prediction result; iterating the scheme input into the result prediction model by using a pre-set PID controller until the candidate prediction result meets a pre-set requirement, thereby obtaining hot isostatic pressing diffusion welding optimization parameters; constructing a hot isostatic pressing diffusion physical model, comprising: constructing an interface atomic diffusion equation; the interface atomic diffusion equation comprising: and ; wherein, is an atomic diffusion flux; is a diffusion coefficient; is an element concentration value; denotes a distance; denotes a time; constructing a temperature dependence equation; the temperature dependence equation is: ; wherein, is a diffusion constant; is a diffusion activation energy; is a gas constant; is an absolute temperature; constructing a densification equation; the densification equation is: ; wherein, is a relative density; is a material constant; is a hot isostatic pressing pressure; , is a first fitting exponent, a second fitting exponent, respectively; constructing a free energy equation; the free energy equation is: ; wherein, is a change in free energy; is a change in enthalpy; is a change in entropy; fusing the interface atomic diffusion equation, the temperature-dependent equation, the densification equation and the free energy equation to obtain the hot isostatic pressing diffusion physical model; setting a plurality of experimental schemes, inputting each of the experimental schemes into the hot isostatic pressing diffusion physical model to obtain theoretical prediction results, comprising: selecting process parameters to be controlled; the process parameters to be controlled include hot isostatic pressing temperature, hot isostatic pressing pressure, holding time, cooling rate and heating rate; determining material attribute parameters; the material attribute parameters include diffusion constant, diffusion activation energy and material constant; orthogonally combining the process parameters to be controlled and the material attribute parameters to obtain a plurality of experimental schemes; inputting the experimental schemes into the hot isostatic pressing diffusion physical model to obtain the theoretical prediction results; the theoretical prediction results include tensile strength, thermal diffusivity, porosity and residual stress.
2. The method of claim 1, wherein the method is based on deep learning. performing hot isostatic pressing diffusion welding experiments according to the experimental schemes to obtain actual detection results, comprising: selecting a base material matched with the experimental scheme parameters; sequentially cutting, polishing, ultrasonic cleaning, deionized water washing and drying the base material to obtain an experimental sample; calibrating temperature, pressure and rate of a hot isostatic pressing device; after placing a pre-set tooling and the experimental sample into a furnace cavity of the hot isostatic pressing device, performing vacuumizing treatment on the furnace cavity; controlling the hot isostatic pressing device to heat at a heating rate provided by the experimental scheme, and performing hot isostatic pressing treatment on the experimental sample when the temperature in the furnace cavity meets the requirement of the experimental scheme; diffusion welding the experimental sample at a fixed temperature and pressure in a holding and pressure maintaining stage, and controlling the furnace cavity to perform cooling and pressure reducing treatment according to the experimental scheme after the holding and pressure maintaining stage ends, thereby obtaining a welded sample; respectively performing mechanical property detection, thermal physical property detection, microscopic quality detection and stress state detection on the welded sample to obtain the actual detection results.
3. The method of claim 1, wherein the method is based on deep learning. The improved MLP model comprises an input layer, a weighted preprocessing layer, a hidden layer and an output layer connected in sequence.
4. The method of claim 1, wherein the method is based on deep learning. The scheme input into the result prediction model is iterated by using a preset PID controller until the candidate prediction result meets the preset requirement, and the hot isostatic pressing diffusion welding optimization parameter is obtained, including: The candidate prediction results are set as the control target, and the PID input variables are set according to the control target; the expression of the PID input variables is: ;in, For the first Control deviation during the next iteration; The number of target indicators; For the first The weight of each indicator; For the first The result prediction model output during the iteration is the first... One indicator; For the first The preset target value for each indicator; The PID controller is constructed, and an expression of the PID controller is: ； wherein, is a parameter adjustment amount; , , are a proportional coefficient, an integral coefficient, and a differential coefficient, respectively; is is a bias integral of the nth iteration; is a bias change rate; It is judged whether the PID input variable meets the preset requirement, and a deviation degree state is obtained; When the deviation degree state does not meet the requirement, the PID input variable is input into the PID controller, and the parameter adjustment amount is obtained; The parameter adjustment amount is updated into the scheme of the result prediction model, and the step of "judging whether the PID input variable meets the preset requirement and obtaining a deviation degree state" is returned until the deviation degree state meets the requirement, and the last updated scheme of the result prediction model is output, and the hot isostatic pressing diffusion welding optimization parameter is obtained.
5. The deep learning-based hot isostatic pressing diffusion bonding parameter optimization method according to claim 3, characterized in that, The original training data is used to train the improved MLP model, and an error compensation model is obtained, including: The improved MLP model is constructed; The original training data is received by the input layer; The weighted preprocessing layer performs pairwise weighted calculations on the data in each group of the original training data to obtain weighted interactive data; the expression for the weighted preprocessing layer is: ;in, The first data in the same set of the original training data The data and the first The weighted interactive data of each data point; This is a learnable interaction weight matrix; , The first data point in the same group of the original training data is respectively the second data point in the original training data. The data and the first One data point; The weighted interaction data and the original training data are synchronously input into two full connection networks connected in sequence in the hidden layer for non-linear mapping, and extraction features are obtained; The extraction features are input into the output layer for multi-index error prediction, and a model prediction result is obtained; According to the model prediction result and the theoretical actual error data, the hidden layer and the weighted preprocessing layer are iteratively optimized by back propagation, and the trained error compensation model is obtained.
6. A deep learning-based hot isostatic pressing diffusion welding parameter optimization system, characterized in that, A hot isostatic pressing diffusion welding parameter optimization method based on deep learning is used to realize the method of claim 1, and the system comprises: A physical model construction module is used to construct a hot isostatic pressing diffusion physical model; A theoretical data collection module is used to set a plurality of experimental schemes, input each experimental scheme into the hot isostatic pressing diffusion physical model, and obtain a theoretical prediction result; An actual data collection module is used to perform a hot isostatic pressing diffusion welding experiment according to the experimental scheme, and obtain an actual detection result; A training data construction module is used to extract errors of the theoretical prediction result and the actual detection result, obtain theoretical actual error data, and align and integrate the theoretical actual error data, the theoretical prediction result and the experimental scheme to obtain original training data; A compensation model construction module is used to train an improved MLP model by using the original training data, and obtain an error compensation model; A model fusion module is used to fuse the hot isostatic pressing diffusion physical model and the error compensation model, and obtain a result prediction model; A result prediction module is used to set a parameter range, randomly select a scheme in the parameter range, input the scheme into the result prediction model, and obtain a candidate prediction result; A parameter optimization module is configured to use a preset PID controller to iteratively optimize the scheme input to the result prediction model until the candidate prediction result meets a preset requirement, and obtain the HIP diffusion welding optimization parameter.
7. An electronic device, comprising: The method comprises the following steps: At least one processor and a memory connected to the processor in communication; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to perform the deep learning-based HIP diffusion welding parameter optimization method in any one of claims 1 to 5.
8. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the deep learning-based HIP diffusion welding parameter optimization method in any one of claims 1 to 5.
Citation Information
Patent Citations
Systems and methods for additive manufacturing and repair of metal components
CN109937387A
Fabricating a recursive flow gas distribution stack using multiple layers
US20220056584A1