Casting process optimization method and device combining numerical simulation and machine learning

Abstract

The invention relates to a numerical simulation and machine learning combined casting process optimization method and device.The method comprises the steps that during numerical simulation low-pressure casting, a mechanism model based on momentum conservation and energy conservation is adopted to solve a flow field and a temperature field, a volume function method is adopted to process a flowing free surface in the mold filling process, and the flow field and the temperature field are calculated; the microscopic segregation degree in the solidification process is determined according to the relation between the solidification solid phase fraction and the temperature, and a solidification temperature field simulation result is obtained; and according to the simulation result, adopting a Xinshan criterion representing the permeation feeding capacity of the solidified interdendritic liquid to predict the shrinkage cavity porosity so as to obtain the shrinkage cavity porosity volume, and adopting a machine learning method based on Gaussian process regression and heredity to predict the shrinkage cavity porosity under different low-pressure casting process conditions. Therefore, the problems that the design cost is high, the period is long, and the adaptability and robustness of the technological parameters are poor due to the fact that the selection of the casting technological parameters mostly depends on experience or repeated tests and a large number of simulation and actual trial casting tests are needed are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of casting process optimization, and particularly relates to a casting process optimization method and device combining numerical simulation and machine learning. BACKGROUND

[0002] In related technologies, in the process of manufacturing aerospace castings, the design of casting process is directly related to the quality and performance stability of the castings. Low-pressure casting process is widely used in the production of aerospace aluminum alloy castings with complex structure and high performance requirements due to its high forming quality and good compactness. By reasonably designing process parameters, the metal flow and solidification process can be effectively controlled to prevent the occurrence of typical casting defects such as shrinkage and porosity.

[0003] However, in related technologies, the selection of casting process parameters mostly depends on experience or repeated experiments, and there is a lack of systematic quantitative methods. In particular, in the new product development stage, the best process window corresponding to different structures, wall thicknesses and alloy types differs significantly. If experience-based rules or trial-and-error optimization are still used, a large number of simulations and actual pouring tests are often required, resulting in high design costs, long cycle times, and poor adaptability and robustness of process parameters, which need to be solved urgently. SUMMARY

[0004] The present application provides a casting process optimization method and device combining numerical simulation and machine learning to solve the problem that in related technologies, the selection of casting process parameters mostly depends on experience or repeated experiments, and in particular in the new product development stage, the best process window corresponding to different structures, wall thicknesses and alloy types differs significantly, often requiring a large number of simulations and actual pouring tests, resulting in high design costs, long cycle times, and poor adaptability and robustness of process parameters.

[0005] The first aspect embodiment of the present application provides a casting process optimization method combining numerical simulation and machine learning, comprising the following steps: determining the casting material, casting process and basic process parameters based on the function of the target aerospace thin-walled casting, casting structure and size requirements; performing three-dimensional solid modeling and mesh division according to the shape of the casting structure, and determining the relationship between the physical property parameters and the temperature according to the casting material, so as to solve the flow field and temperature field by using a mechanism model based on momentum conservation and energy conservation in numerical simulation low-pressure casting, and to process the flow free surface in the filling process by using the volume function method, and to determine the degree of microsegregation in the solidification process according to the relationship between the solidification solid phase fraction and the temperature, so as to obtain the simulation results of the solidification temperature field; predicting shrinkage and porosity by using the new mountain criterion representing the liquid infiltration and compensation ability between solidification dendrites according to the simulation results of the solidification temperature field, so as to obtain the volume of shrinkage and porosity, and predicting the shrinkage and porosity under different low-pressure casting process conditions based on the volume of shrinkage and porosity by using a machine learning method based on Gaussian process regression and genetic algorithm.

[0006] By the above technical means, the shrinkage porosity defects in the low-pressure casting process are predicted by numerical simulation, the distribution of defect volume under different process parameter combinations is obtained, and then the nonlinear mapping relationship between the process parameters and the defect volume is constructed by combining the Gaussian process regression and genetic machine learning, and the optimal casting process parameter combination is obtained by optimization search, so that the process stability can be improved, the defect occurrence probability can be reduced, and the intelligent optimization and precise control of the casting process can be realized under the premise of ensuring the casting quality.

[0007] Optionally, in an embodiment of the present application, the probability density function of the Gaussian process regression and genetic machine learning method is:

[0008]

[0009] Wherein, f(X) is the probability density function; X is a random variable; K is a covariance matrix; n represents that X is an n-dimensional vector; and μ represents a mean vector.

[0010] By the above technical means, the probability density function is established based on the Gaussian process regression and genetic machine learning method, which can effectively depict the nonlinear mapping relationship between the process parameters and the defect response, and quantify the uncertainty of the prediction result, so as to realize the fine control and optimization design of the casting quality, improve the process stability, and reduce the defect risk.

[0011] Optionally, in an embodiment of the present application, the shrinkage porosity under different low-pressure casting process conditions is predicted by using the Gaussian process regression and genetic machine learning method, which comprises: obtaining an fitness function by Gaussian process regression; taking the fitness function as the objective function of genetic algorithm to search for the optimal target parameter combination in the preset random parameter combination range through simulated mutation, genetic, selection, wherein the process of finding the optimal solution is reflected as an adaptive process.

[0012] By the above technical means, the fitness function is constructed by Gaussian process regression, and the function is taken as the objective function of genetic algorithm, and then the optimal process parameter combination is searched, so that the defect occurrence probability can be significantly reduced, the shrinkage porosity and other defects can be precisely inhibited, and the overall process stability and product consistency can be improved under the premise of considering the casting quality and efficiency, thereby reducing the quality fluctuation and cost risk in the manufacturing process.

[0013] Optionally, in an embodiment of the present application, it further comprises: taking the simulation results of orthogonal test as the training set of machine learning, taking the results of full number experiment as the test set, when the machine learning prediction result meets the simulation accuracy requirement, constructing a learning model which can obtain the optimal process scheme.

[0014] By means of the above technical means, the numerical simulation results of the orthogonal experiment are taken as the training set of machine learning, the results of the full numerical simulation experiment are taken as the test set, and then the learning model of the optimal process scheme is constructed, the volume and distribution of shrinkage cavity, porosity and other defects under different process parameter combinations can be accurately predicted, the model can be strictly verified by using the full numerical experiment data to ensure that it has good generalization ability and prediction accuracy, and the overfitting phenomenon is avoided, meanwhile, based on the model, intelligent screening and optimization of process parameters can be effectively assisted, so that the test cycle is significantly shortened and the research and development cost is reduced, and the stability of the casting process and the product quality are improved.

[0015] The second aspect embodiment of the application provides a casting process optimization device combining numerical simulation and machine learning, comprising: a determination module configured to determine a casting material, a casting process and basic process parameters based on a target aerospace thin-walled casting function, casting structure and size requirements; a simulation module configured to perform three-dimensional solid modeling and mesh division according to the shape of the casting structure, and determine the relationship between physical parameters and temperature according to the casting material, so as to solve the flow field and temperature field by using a mechanism model based on momentum conservation and energy conservation during numerical simulation of low-pressure casting, and to process the flow free surface during the filling process by using a volume function method, and to determine the degree of microsegregation during solidification according to the relationship between solidification solid phase fraction and temperature during the solidification process, and to obtain a solidification temperature field simulation result; and an optimization module configured to predict shrinkage cavity and porosity by using a new mountain criterion representing the liquid infiltration and shrinkage ability between solidification dendrites according to the solidification temperature field simulation result, to obtain the volume of shrinkage cavity and porosity, and to predict shrinkage cavity and porosity under different low-pressure casting process conditions by using a machine learning method based on Gaussian process regression and genetic algorithm based on the volume of shrinkage cavity and porosity.

[0016] By means of the above technical means, the numerical simulation is used to predict the shrinkage cavity and porosity defects in the low-pressure casting process, the distribution of defect volume under different process parameter combinations is obtained, and then a nonlinear mapping relationship between process parameters and defect volume is constructed by combining Gaussian process regression and genetic machine learning, optimization search is performed to obtain the optimal casting process parameter combination, the process stability can be improved and the defect occurrence probability can be reduced under the premise of ensuring the casting quality, and intelligent optimization and accurate control of the casting process are realized.

[0017] Optionally, in an embodiment of the application, the probability density function of the machine learning method based on Gaussian process regression and genetic algorithm is:

[0018]

[0019] Wherein, f(X) is the probability density function; X is a random variable; K is a covariance matrix; n represents that X is an n-dimensional vector; and μ represents a mean vector.

[0020] By the above technical means, the probability density function is established based on the Gaussian process regression and the genetic machine learning method, the nonlinear mapping relationship between the process parameters and the defect response can be effectively described, the uncertainty of the prediction result is quantified, the fine control and optimization design of the casting quality are realized, the process stability is improved, and the defect risk is reduced.

[0021] Optionally, in an embodiment of the present application, the optimization module comprises: an acquisition unit configured to obtain the fitness function through Gaussian process regression; and an optimization unit configured to take the fitness function as a target function of a genetic algorithm to search for an optimal target parameter combination in a preset random parameter combination range through simulated mutation, genetic, and selection, wherein the process of finding an optimal solution is reflected as an adaptation process.

[0022] By the above technical means, the fitness function is constructed through Gaussian process regression, the function is taken as a target function of a genetic algorithm, and then an optimal process parameter combination is searched, so that the defect occurrence probability can be significantly reduced on the premise of considering the casting quality and efficiency, the defects such as shrinkage holes and porosity can be accurately inhibited, and the overall process stability and product consistency can be improved, and the quality fluctuation and cost risk in the manufacturing process can be reduced.

[0023] Optionally, in an embodiment of the present application, further comprising: a verification module configured to take the simulation results of the orthogonal test as a training set of machine learning, take the results of the full number of experiments as a test set, and construct a learning model that can derive an optimal process scheme when the machine learning prediction result meets the simulation accuracy requirement.

[0024] By the above technical means, the numerical simulation results of the orthogonal test are taken as a training set of machine learning, the results of the full number of simulation experiments are taken as a test set, and then a learning model of an optimal process scheme is constructed, the volume and distribution of defects such as shrinkage holes and porosity under different process parameter combinations can be accurately predicted, the model can be strictly verified by using the full number of experimental data to ensure that it has good generalization ability and prediction accuracy, and the overfitting phenomenon is avoided, and based on the model, intelligent screening and optimization of process parameters can be effectively assisted, so that the test cycle is significantly shortened and the research and development cost is reduced, and the stability of the casting process and the finished product quality are improved.

[0025] The third aspect embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the program to implement the casting process optimization method combining numerical simulation and machine learning as described in the above embodiments.

[0026] The fourth aspect of the present application provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the casting process optimization method combining numerical simulation and machine learning as above.

[0027] The fifth aspect of the present application provides a computer program product comprising a computer program, which is executed to implement the casting process optimization method combining numerical simulation and machine learning as above.

[0028] Additional aspects and advantages of the present application will be made apparent by the following description and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0029] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, wherein:

[0030] Figure 1 A flow chart of a casting process optimization method combining numerical simulation and machine learning according to an embodiment of the present application;

[0031] Figure 2 A flow chart of a casting process optimization genetic algorithm according to an embodiment of the present application;

[0032] Figure 3 A schematic diagram of low pressure casting shrinkage porosity prediction based on numerical simulation before machine learning according to an embodiment of the present application;

[0033] Figure 4 A schematic diagram of low pressure casting shrinkage porosity prediction based on numerical simulation after machine learning according to an embodiment of the present application;

[0034] Figure 5 A flow chart of a casting process optimization design combining numerical simulation and machine learning according to an embodiment of the present application;

[0035] Figure 6 A block diagram of a casting process optimization device combining numerical simulation and machine learning according to an embodiment of the present application;

[0036] Figure 7 A structural diagram of an electronic device according to an embodiment of the present application.

[0037] REFERENCE NUMERALS

[0038] 10 - casting process optimization device combining numerical simulation and machine learning; 100 - determination module, 200 - simulation module, 300 - optimization module; 701 - memory, 702 - processor, 703 - communication interface. Detailed Implementation

[0039] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0040] The following description, with reference to the accompanying drawings, describes a casting process optimization method and apparatus combining numerical simulation and machine learning according to embodiments of this application. Addressing the technical problem mentioned in the background art, where the selection of low-pressure casting process parameters largely relies on experience or repeated experiments, requiring extensive simulations and actual trial castings, resulting in high design costs, long cycles, and poor adaptability and robustness of process parameters, this application provides a casting process optimization method combining numerical simulation and machine learning. In this method, based on the principles of momentum and energy conservation, a mechanistic model of the low-pressure casting process is established. Numerical simulation is used to obtain the distribution of shrinkage cavities and porosity defects during the casting process. Furthermore, by combining Gaussian process regression and genetic machine learning methods, accurate modeling and prediction of the complex nonlinear relationship between key process parameters and defect indicators in the casting process are achieved. This effectively guides process parameter optimization, improves casting quality, significantly enhances the density and mechanical properties of castings, and strengthens product reliability. Simultaneously, the optimized process parameter combination has stronger adaptability, reducing quality fluctuations caused by parameter fluctuations, achieving stable production, thereby reducing production risks and costs, reducing the need for repeated experiments and process adjustments, shortening the process development cycle, saving manpower and material costs, and possessing broad engineering application prospects. This solves the problem in related technologies where the selection of casting process parameters mostly relies on experience or repeated experiments, requiring a large number of simulations and actual trial casting tests, resulting in high design costs, long cycles, and poor adaptability and robustness of process parameters.

[0041] Specifically, Figure 1 This is a flowchart illustrating a casting process optimization method combining numerical simulation and machine learning, provided in an embodiment of this application.

[0042] like Figure 1 As shown, the casting process optimization method combining numerical simulation and machine learning includes the following steps:

[0043] In step S101, based on the function, structure and size requirements of the target aerospace thin-walled casting, the casting material, casting process and basic process parameters are determined.

[0044] It can be understood that aerospace thin-walled castings usually need to withstand thermal load and force load under extreme and complex working conditions, and the functional requirements usually include high specific strength, high specific stiffness and good thermal stability to ensure the long-term reliable operation of the structure in a high-temperature, high-pressure and vibration environment; in order to meet the above functional requirements, the structure of the casting is usually designed to have multi-functional channels, precise positioning interfaces and complex internal cavity shapes to realize functional integration, weight reduction and assembly difficulty reduction, thereby improving the reliability and stability of the whole machine system; in terms of size, the aerospace thin-walled castings have strict requirements on geometric precision and size consistency, and especially the uniformity of wall thickness needs to be strictly controlled to avoid casting defects such as thermal node, shrinkage hole, crack and the like caused by uneven wall thickness, and to ensure the density and structural integrity of the finished product. The function, structure and size of the aerospace thin-walled casting can be set by a person skilled in the art according to the actual situation, which is not specifically limited here.

[0045] Further, the embodiments of the present application can determine the casting material, casting process and basic process parameters according to the functional requirements of the target aerospace thin-walled casting, casting structure and size. The casting material can include but is not limited to aluminum alloy, magnesium alloy, titanium alloy and high-temperature alloy, and the material needs to consider factors such as mechanical properties, use environment, processing performance and cost; the corresponding casting process can adopt but is not limited to low-pressure casting, precision casting (investment), vacuum casting or semi-solid forming and the like; the basic process parameters can include but are not limited to forming temperature, mold temperature, pouring pressure and pressure holding time and the like.

[0046] For example, in the embodiments of the present application, the height of the cabin body casting can be about 730 mm, the outer diameter can be 380 mm, the minimum wall thickness can be about 10 mm, and the reinforcement rib height can be 15 mm, and the material of the casting can be AC-42100 aluminum alloy.

[0047] To improve the solidification control ability and forming quality of the casting, the low-pressure casting process structure combining with the conformal chill and bottom injection gap gating system can be designed. The conformal chill is customized according to the local thermal node distribution and solidification characteristics of the casting, and is arranged along the complex curved surface or reinforcement rib area, which can effectively improve the local cooling rate, guide the sequential solidification and inhibit the generation of defects such as shrinkage and hot cracking; the bottom injection gap gating system has the advantages of smooth filling, inclusion floating and reducing turbulence by slowly filling the metal liquid from the lower part of the casting mold along the gap type runner, and is especially suitable for the forming of thin-walled structures and complex internal cavity parts. The combined design can further improve the organization density and size stability while ensuring the integrity of the casting filling, thereby effectively meeting the requirements of high-quality casting process for aerospace complex thin-walled components.

[0048] In the embodiments of the present application, the range of the mold temperature in the process parameters can be 250-300°C, the pouring temperature can be 700-780°C, the filling time can be 8-12s, and the filling pressure can be 180-220kPa.

[0049] As a possible implementation manner, the embodiments of the present application can construct a full number of experiment schemes according to the above process parameters, and further carry out numerical simulation to systematically analyze the influence of each parameter on the quality indicators such as porosity defects in the forming process. Specifically, the embodiments of the present application can select three levels with equal intervals within the above four process parameter ranges, and then construct a four-factor three-level orthogonal test scheme (L9(3 4 )) and a full number of experiment schemes (3 4 =81 groups) to provide a data basis for subsequent rule extraction and modeling analysis.

[0050] The embodiments of the present application can use numerical simulation software to simulate the casting process for each group of parameter combinations, and can obtain typical quality evaluation indicators, which can include but are not limited to shrinkage hole distribution, filling integrity, temperature gradient, solidification sequence, etc. Through comparative analysis of the simulation results, the main control factors affecting the formation of defects and their interaction rules can be identified, thereby providing a theoretical basis for process optimization.

[0051] In step S102, three-dimensional solid modeling and mesh division are performed according to the shape of the casting structure, and the relationship between the physical property parameters and the temperature is determined according to the casting material, so that in numerical simulation of low-pressure casting, a mechanism model based on momentum conservation and energy conservation is used to solve the flow field and temperature field, and a volume function method is used to process the flow free surface in the filling process, and in the solidification process, the degree of microsegregation in the solidification process is determined according to the relationship between the solidification solid phase fraction and the temperature, and the simulation result of the solidification temperature field is obtained.

[0052] It can be explained that before numerical simulation, the embodiments of the present application can first perform three-dimensional solid modeling on the geometric shape of the casting structure, construct a complete three-dimensional geometric modeling according to the actual size, structural features and key functional areas, and truly reflect the actual forming conditions; further, the embodiments of the present application can perform mesh division on the three-dimensional modeling to provide a basis for subsequent numerical calculation of the thermal flow field, flow field and solidification process. The mesh type usually adopts tetrahedral or hexahedral elements, and local encryption processing can be performed on key areas (such as thin-walled parts, corners or hot spot prone areas) according to the complexity of the casting structure and the simulation accuracy requirement, thereby improving the accuracy and stability of the simulation result.

[0053] The physical property parameters can include, but are not limited to, density, specific heat capacity, thermal conductivity, viscosity, surface tension, latent heat of solidification, and solid-liquid phase line temperature, and the parameters can change nonlinearly with temperature. In the embodiments of the present application, the thermal physical property parameters of the casting material in different temperature ranges can be determined, which can provide accurate physical basis for numerical simulation and temperature field analysis of the solidification process, and help to realize the coupling calculation of the subsequent heat flow field and solidification behavior, thereby improving the reliability and engineering guidance value of the simulation results.

[0054] The flow field mainly reflects the flow path, velocity distribution and turbulent flow characteristics of the molten metal in the cavity, and can be used to identify the filling dead angle, backflow area and high-risk area where inclusions or pores may occur. The temperature field is used to represent the temperature distribution of the metal liquid and the mold at different times, and is a key basis for predicting the solidification sequence, hot spot position and solidification defects. Reasonable flow field distribution helps to achieve smooth filling and avoid turbulent flow, gas entrainment and cold separation; and uniform and controllable temperature field distribution is conducive to realizing sequential solidification and reducing the probability of shrinkage and hot cracking.

[0055] It should be noted that the free surface refers to the interface between the molten metal and the gas medium (such as air) formed inside the cavity during the filling process. The free surface has high dynamicity, and its morphological change can reflect the flow stability and filling quality of the filling process. The free surface is prone to wave, roll or entrain gas in the rapid flow, turning or flow convergence area, thereby causing casting defects such as gas entrapment, porosity and cold separation. The solidification solid fraction can be used to represent the proportion of the alloy that has been converted from liquid to solid at a certain time and a certain position, and is an important physical parameter for evaluating the progress and behavior of the solidification process. The value range is usually 0 to 1, where 0 can represent complete liquid state, 1 can represent complete solidification, and 0-1 can represent liquid-solid coexistence state. Microsegregation refers to the phenomenon of local enrichment or depletion of composition on a microscale due to uneven distribution of solute elements at the solid-liquid interface during alloy solidification. High microsegregation degree easily leads to uneven organization between grains, local melting point reduction, and even formation of low melting point eutectic, thereby causing hot cracking, increased corrosion sensitivity or mechanical property degradation. The solidification temperature field simulation result refers to the visualized and quantified analysis result of the spatiotemporal variation law of the temperature distribution during the process from the liquid metal filling to the completion of the solidification of the casting.

[0056] The embodiment of the application can obtain the solidification temperature field distribution result of the casting by the mechanism model of momentum conservation and energy conservation, and accurately track the liquid-gas free surface by combining the VOF (Volume of Fluid), so as to accurately reflect the filling and solidification behavior of the metal liquid, thereby providing an important basis for subsequent process optimization, defect prediction and casting quality improvement.

[0057] In step S103, the shrinkage porosity is predicted according to the solidification temperature field simulation result by using a Ny criterion representing the liquid penetration and feeding ability between solidification dendrites, to obtain the shrinkage porosity volume, and based on the shrinkage porosity volume, a machine learning method based on Gaussian process regression and genetic algorithm is used to predict the shrinkage porosity under different low-pressure casting process conditions.

[0058] The shrinkage porosity mainly refers to the voids or pores formed due to the volume shrinkage of the metal during the solidification process and unable to be supplemented by sufficient liquid metal, and the factors causing the shrinkage porosity can include but are not limited to too fast solidification speed, insufficient temperature gradient, narrow interdendritic gap, etc. Such defects can seriously affect the compactness and mechanical properties of the casting, and even cause the service life to be shortened or invalid.

[0059] As a specific example, the Ny (Niyama) criterion representing the liquid penetration and feeding ability between solidification dendrites can be used to predict the shrinkage porosity, which can be expressed as:

[0060]

[0061] Wherein, G is the temperature gradient, with the unit of ℃ / cm; and ε is the cooling speed, with the unit of ℃ / min.

[0062] It should be noted that the larger the Ny criterion value is, the greater the temperature gradient of the solidification zone and the slower the cooling rate, which is beneficial to the liquid metal feeding, thereby reducing the risk of shrinkage formation; on the contrary, a lower Ny criterion value means insufficient feeding ability, which is prone to shrinkage defects.

[0063] The embodiment of the application predicts the shrinkage porosity risk in the casting, and then obtains the volume evolution and spatial distribution characteristics in the solidification process, which can provide key support for subsequent defect evaluation, process parameter optimization and structure reliability analysis.

[0064] It can be understood that the Gaussian process regression is a probability model integrating machine learning and statistics, applying Bayesian inference, without presetting linear equation or polynomial function form, can be completely data-driven, and can perform non-parametric regression analysis on data under the assumption of Gaussian process, and can give the posterior distribution of the predicted sample. In particular, in the high-dimensional nonlinear and small sample data scene, the Gaussian process regression can fully utilize the information of limited data, and provide accurate and robust interpolation prediction results. Genetic machine learning method generally refers to genetic algorithm and its integration with machine learning, which can iteratively search for optimal solution from search space by simulating "selection, crossover, mutation" and other operations in biological evolution process, without function derivative information, and can be applied to complex, nonlinear, high-dimensional, non-derivable and even discontinuous problems.

[0065] The embodiments of the present application can combine the advantages of Gaussian process regression and genetic machine learning method, model the complex nonlinear relationship between key parameters and defect indicators in the casting process, thereby realizing the prediction of shrinkage porosity occurrence probability and distribution, and fully mining the potential information of data under the condition of small sample data, thereby improving the reliability and generalization ability of prediction.

[0066] As a possible implementation manner, the embodiments of the present application can perform solidification numerical simulation based on the above designed orthogonal experiment scheme. Specifically, the embodiments of the present application can record the pouring temperature as TP, the mold preheating temperature as TM, the filling time as t, the holding pressure as p, and the shrinkage porosity volume as v, and then can obtain 9 groups of sample data from the orthogonal test L9(3 4 )4, each group of sample data can be composed of a four-dimensional input vector and an output scalar.

[0067] The input vector can be expressed as:

[0068] x i =(TP i ,TM i ,t i ,p i )∈R 4 ,

[0069] In the formula, TP is the recorded pouring temperature; TM is the mold preheating temperature; t is the filling time; p is the holding pressure; x i represents the input vector composed of process parameters, R 4 represents that the input vector is a four-dimensional vector.

[0070] The output scalar can be expressed as:

[0071]

[0072] In the formula, y i =vi ; y i represents an output scalar; v i represents a given process parameter combination x i , D represents a loose volume obtained by the following formula; D = f(x i , y i ) represents a vector (x i , y i ) combined by the output scalar y 9 ; R j represents that y is a nine-dimensional vector.

[0073] Further, the prediction results of the numerical simulation of the nine groups of loose volumes under the combination of the four factors and the three levels in the embodiments of the present application can be shown in Table 1. Table 1 is a low-pressure casting orthogonal test design scheme and a shrinkage cavity loose table based on numerical simulation prediction.

[0074] Table 1

[0075]

[0076] In some cases, in order to eliminate the influence of dimensional differences of each dimension on numerical stability, the embodiments of the present application can use z-score for standardization:

[0077]

[0078] Wherein: μ j , σ j are the sample mean and standard deviation of the jth column of x respectively; similarly, μ y , σ y correspond to y; x ik is the item of the ith row and the jth column of the matrix x; represents the standardized x ok ; y o represents the ith item of the vector y; represents the standardized y o .

[0079] Through the standardization processing, the embodiments of the present application can ensure that each dimension feature has the same importance in the model training process, and avoid that some features have a dominant influence on the model results due to the too large dimension.

[0080] Further, in the embodiments of the present application, for any dimensional random variable X (x1, x2,..., x n ), its Gaussian distribution can be written as X ~ N(μ, K), where K is the covariance matrix and μ is the mean vector. Thus, the embodiments of the present application can further obtain its corresponding probability density function, which can be used to describe the probability characteristics of the random variable at each value point.

[0081] Optionally, in an embodiment of the present application, the probability density function of the machine learning method based on Gaussian process regression and genetic algorithm can be expressed as:

[0082]

[0083] Wherein, f(X) is the probability density function; X is the random variable; K is the covariance matrix; n represents that X is an n-dimensional vector; μ represents the mean vector.

[0084] The embodiment of the present application can obtain a typical process parameter combination based on orthogonal experimental design, predict the corresponding shrinkage porosity volume through numerical simulation, model the parameter-defect relationship by using the machine learning method based on Gaussian process regression and genetic algorithm, and then search for the optimal solution, thereby further improving the prediction accuracy and optimization efficiency. The method can perform probabilistic modeling on the nonlinear relationship between the process parameters and the porosity volume by using the Gaussian process regression, fully characterize the uncertainty of the prediction result, and construct a fitness function based on this, so as to realize global optimization of the process parameters by combining the genetic algorithm, thereby improving the prediction accuracy and the robustness of process control.

[0085] Optionally, in an embodiment of the present application, the machine learning method based on Gaussian process regression and genetic algorithm is used to predict the shrinkage porosity under different low-pressure casting process conditions, including: obtaining a fitness function by using Gaussian process regression; taking the fitness function as the objective function of the genetic algorithm, so as to search for the optimal target parameter combination in the preset random parameter combination range through simulation mutation, genetic, and selection, wherein the process of finding the optimal solution is reflected as an adaptation process.

[0086] The embodiment of the present application can construct a fitness function based on the probability distribution function obtained by the Gaussian process regression, and then take the fitness function as the objective function of the genetic algorithm, which can be used to guide the optimization search of the process parameters, thereby realizing accurate prediction and effective control of the shrinkage porosity defect in the low-pressure casting process.

[0087] It should be noted that the genetic algorithm searches for the optimal target parameter combination in a certain random parameter combination range by simulating mutation, genetic, and selection, reflects the process of finding the optimal solution as an adaptation process, so as to realize global optimization of complex problems.

[0088] As shown in FIG. 1, Figure 2 The following will take a specific example to describe the process of the genetic algorithm of the embodiment of the present application in detail. The embodiment of the present application can include the following steps:

[0089] In step S201, the genetic algorithm starts;

[0090] In step S202, a certain number of initial solutions (individuals) are randomly generated to form an initial population;

[0091] In step S203, for each individual in the population, a fitness value is calculated according to the objective function to evaluate its pros and cons;

[0092] In step S204, it is judged whether a termination condition (fitness reaches a threshold value) is met, and if so, step S208 is executed; otherwise, step S205 is executed.

[0093] In step S205, according to the fitness value, some excellent individuals are selected as parents for generating the next generation;

[0094] In step S206, the selected parent individuals are subjected to gene crossover with a certain probability to generate new offspring individuals to increase population diversity;

[0095] In step S207, the genes of the newly generated individuals are subjected to random mutation with a certain probability to prevent falling into local optimum, and the individuals after crossover and mutation are used to form a new generation population;

[0096] In step S208, the individual with the highest fitness is output.

[0097] In step S209, the genetic algorithm ends.

[0098] The embodiments of the present application can perform global optimization search in a multi-dimensional parameter space through a genetic algorithm, evaluate and select candidate solutions in combination with a fitness function, and then gradually approach and finally obtain an optimal solution that satisfies a preset performance index, thereby realizing accurate optimization and defect control of casting process parameters.

[0099] Optionally, in an embodiment of the present application, the orthogonal test simulation results are used as a training set for machine learning, the full number of experiments is used as a test set, and when the machine learning prediction result meets the simulation accuracy requirement, a learning model that can derive an optimal process scheme is constructed.

[0100] To realize accurate prediction and optimization, the numerical simulation results obtained by orthogonal test design can be used as a training set for the machine learning model in the embodiments of the present application, to ensure that the model can effectively learn the complex nonlinear relationship between the process parameters and the shrinkage volume. Meanwhile, to verify the generalization ability and prediction accuracy of the model, the simulation results obtained by full number of experiments design can be used as a test set to strictly test the trained proxy model.

[0101] In the embodiments of the present application, when the prediction result of the machine learning model on the test set meets the preset accuracy requirement, it can be considered that the model has good generalization ability and reliability, and based on the model, the corresponding optimal process scheme, i.e., the parameter combination with the smallest shrinkage volume, can be obtained.

[0102] In addition, to ensure the accuracy and engineering feasibility of the optimization result, several optimal process schemes predicted by the machine learning can be selected for rechecking in numerical simulation to ensure that the prediction result is highly consistent with the actual simulation result, thereby verifying the effectiveness and practical value of the machine learning method in process parameter optimization.

[0103] The following enumerated embodiments illustrate the effect of the process parameter optimization after machine learning of the embodiments of the present application.

[0104] As shown in Figure 3 , before the method is used, the embodiments of the present application predict and analyze the shrinkage porosity defects in the low-pressure casting process based on numerical simulation, wherein under the conditions of a mold preheating temperature of 250°C, a pouring temperature of 700°C, a filling time of 8s and a holding pressure of 180kPa, the porosity volume is 2.98cm 3 . It can be seen that in the embodiments of the present application, there is obvious shrinkage porosity defect in the barrel part of the casting, indicating that the current casting process parameters need to be adjusted and optimized to improve the compactness and mechanical properties of the casting, thereby reducing defects and ensuring the service life and reliability of the casting.

[0105] As shown in Figure 4 , after the process is optimized by the method, the embodiments of the present application predict the shrinkage porosity volume. The optimized process parameters are: a pouring temperature of 762.34°C, a mold temperature of 288.04°C, a filling time of 10.08s and a holding pressure of 196.57kPa, and the porosity volume is 1.7607cm 3 . It can be seen that the porosity volume after optimization is significantly smaller than that before optimization, and therefore the process optimization effect based on machine learning in the embodiments of the present application is better than that of the orthogonal test design method. Moreover, in the embodiments of the present application, the calculation time of one prediction by the machine learning model is only about 1ms, which is much lower than the calculation amount of several hours based on numerical simulation, greatly improving the calculation efficiency.

[0106] As can be seen from the above comparison results, the casting process optimization method combining numerical simulation and machine learning proposed in the embodiments of the present application not only improves the efficiency and precision of the casting process optimization, but also effectively reduces the consumption of computing resources, and has good generalization ability and applicability.

[0107] As shown in Figure 5 , the following illustrates the process of the casting process optimization method combining numerical simulation and machine learning of the embodiments of the present application with a specific example.

[0108] In step S501, k key process factors and m levels are selected for experimental design to construct an orthogonal test scheme and a full-number experimental scheme, so as to systematically explore the process parameter space and obtain corresponding experimental data;

[0109] In step S502, the material property parameters of the casting entity are taken as input, and a mechanism model based on momentum conservation and energy conservation is adopted to numerically simulate the flow field and temperature field in the casting process, so as to accurately predict the solidification behavior and defect formation of the casting;

[0110] In step S503, the temperature gradient and cooling rate in the numerical simulation result are analyzed by adopting the Xinshan criterion, and the defect regions such as shrinkage cavity and porosity that may appear in the casting are predicted, so as to guide the process optimization and defect control;

[0111] In step S504, Gaussian process regression is adopted, combined with genetic algorithm for machine learning, to model and optimize the numerical simulation prediction result, so as to seek the best process parameter combination, thereby realizing the optimization of the casting process;

[0112] In step S505, the numerical simulation results obtained by orthogonal test design are taken as the training set, and the simulation results of full number experiment design are taken as the test set, to train and verify the machine learning model;

[0113] In step S506, the prediction result of the machine learning model is compared with the numerical simulation, if the prediction error meets the accuracy requirement, step S507 is entered, if not, step S504 is entered;

[0114] In step S507, the output optimal process parameter combination is output.

[0115] According to the casting process optimization method combining numerical simulation and machine learning proposed in the embodiments of the present application, a mechanism model of the low-pressure casting process is established based on the principles of momentum conservation and energy conservation, the distribution of shrinkage cavity and porosity defects in the casting process is obtained through numerical simulation, and further combined with the machine learning method of Gaussian process regression and genetic algorithm, the complex nonlinear relationship between the key process parameters and the defect indicators in the casting process is accurately modeled and predicted, which can effectively guide the optimization of process parameters, improve the casting quality, significantly improve the density and mechanical properties of the casting, enhance the product reliability, and the optimized process parameter combination has stronger adaptability, which can reduce the quality fluctuation caused by parameter fluctuation, realize stable production, thereby reducing the production risk and cost, reducing the demand for repeated experiments and process adjustment, shortening the process development cycle, saving manpower and material cost, and having wide engineering application prospect.

[0116] Secondly, the casting process optimization device combining numerical simulation and machine learning according to the embodiments of the present application is described with reference to the accompanying drawings.

[0117] Figure 6 is the block schematic diagram of the casting process optimization device combining numerical simulation and machine learning according to the embodiments of the present application.

[0118] As shown in the combination of numerical simulation and machine learning casting process optimization device 10 includes: determination module 100, simulation module 200 and optimization module 300. Figure 6

[0119] The determination module 100 is configured to determine the casting material, the casting process and the basic process parameters based on the target aerospace thin-walled casting function, the casting structure and the size requirements.

[0120] The simulation module 200 is configured to perform three-dimensional entity modeling and meshing according to the shape of the casting structure, determine the relationship between the physical parameters and the temperature according to the casting material, and solve the flow field and the temperature field by using a mechanism model based on the conservation of momentum and the conservation of energy during numerical simulation of low-pressure casting, and process the flow free surface during the filling process by using the volume function method, and determine the degree of microsegregation during solidification according to the relationship between the solidification solid phase fraction and the temperature during the solidification process, and obtain the simulation results of the solidification temperature field.

[0121] The optimization module 300 is configured to predict shrinkage porosity by using the newkam criterion representing the liquid infiltration and feeding ability between solidification dendrites according to the simulation results of the solidification temperature field, to obtain the shrinkage porosity volume, and to predict the shrinkage porosity under different low-pressure casting process conditions by using a machine learning method based on Gaussian process regression and genetic algorithm.

[0122] Optionally, in an embodiment of the present application, the probability density function of the machine learning method based on Gaussian process regression and genetic algorithm is:

[0123]

[0124] Wherein, f(X) is the probability density function; X is a random variable; K is the covariance matrix; n represents that X is an n-dimensional vector; μ represents the mean vector.

[0125] Optionally, in an embodiment of the present application, the optimization module 300 includes an acquisition unit and an optimization unit.

[0126] The acquisition unit is configured to obtain the fitness function by Gaussian process regression.

[0127] The optimization unit is configured to use the fitness function as the objective function of the genetic algorithm, to search for the optimal target parameter combination within the preset random parameter combination range by simulating mutation, genetic, selection, and to reflect the process of finding the optimal solution as an adaptive process.

[0128] Optionally, in an embodiment of the present application, it further includes a verification module.

[0129] ​The verification module is configured to use the simulation results of the orthogonal experiment as a training set for machine learning, use the results of the full number of experiments as a test set, and construct a learning model that can derive an optimal process scheme when the result predicted by the machine learning meets the simulation accuracy requirement.

[0130] It should be noted that the foregoing explanation of the embodiment of the casting process optimization method combining numerical simulation and machine learning is also applicable to the embodiment of the casting process optimization device combining numerical simulation and machine learning, and will not be repeated here.

[0131] The casting process optimization device combining numerical simulation and machine learning provided by the embodiment of the application is based on the principle of conservation of momentum and energy, establishes a mechanism model of the low-pressure casting process, obtains the distribution of shrinkage and porosity defects in the casting process through numerical simulation, further combines the machine learning methods of Gaussian process regression and genetic algorithm, realizes accurate modeling and prediction of the complex nonlinear relationship between the key process parameters and the defect indicators in the casting process, can effectively guide the optimization of process parameters, can improve the casting quality, significantly improve the density and mechanical properties of the casting, and enhance the product reliability. At the same time, the optimized process parameter combination has stronger adaptability, can reduce the quality fluctuation caused by parameter fluctuation, realizes stable production, thereby reduces the production risk and cost, reduces the demand for repeated experiments and process adjustment, shortens the process development cycle, can save manpower and material costs, and has wide engineering application prospects.

[0132] Figure 7 The structure schematic diagram of the electronic device provided by the embodiment of the application is provided. The electronic device can include:

[0133] The memory 701, the processor 702, and the computer program stored in the memory 701 and executable on the processor 702.

[0134] The processor 702 implements the casting process optimization method combining numerical simulation and machine learning provided in the above embodiments when executing the program.

[0135] Further, the electronic device further includes:

[0136] The communication interface 703 is configured to communicate between the memory 701 and the processor 702.

[0137] The memory 701 is configured to store the computer program executable on the processor 702.

[0138] The memory 701 can include a high-speed RAM memory, and can also include a non-volatile memory, for example, at least one disk memory.

[0139] If the memory 701, the processor 702 and the communication interface 703 are implemented independently, the communication interface 703, the memory 701 and the processor 702 can be connected with each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For convenience of representation, Figure 7 Only one thick line is used to represent the bus in the figure, but it does not mean that there is only one bus or only one type of bus.

[0140] Optionally, in a specific implementation, if the memory 701, the processor 702 and the communication interface 703 are integrated on a chip, the memory 701, the processor 702 and the communication interface 703 can complete communication between each other through an internal interface.

[0141] The processor 702 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement one or more embodiments of the present application.

[0142] The embodiment also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the casting process optimization method combined with numerical simulation and machine learning.

[0143] The embodiment of the present application further provides a computer program product, which includes a computer program. The computer program can run computer instructions. The computer instructions are executed by a processor to implement the casting process optimization method combined with numerical simulation and machine learning provided by the embodiment of the present application.

[0144] In the description of the application, reference to "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that a particular feature, structure, material, or characteristic being described is included in at least one embodiment or example of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment or example. Furthermore, the described specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples. In addition, the usage of "N" means at least two, for example, two, three or the like, unless explicitly stated otherwise.

[0145] Furthermore, the terms "first", "second", or the like, are used merely as a designation of certain elements or features, and do not imply or connote relative importance or a specific order of categorization thereof. Accordingly, features described as "first" or "second" can be explicitly or implicitly included in at least one of the features. In the description of the application, the meaning of "N" is at least two, for example, two, three, etc., unless explicitly specified otherwise.

[0146] Any process or method descriptions or blocks in flow charts or otherwise described herein represent embodiments which can be managed as one or more modules, segments, or portions of code which include one or more executable instructions for implementing specific logic functions or steps, and alternate implementations are possible. In some embodiments, the processes and methods described can be executably encoded on a machine- readable medium in a data signal embodied in an electromagnetic signal, a wireless signal, or a propagated signal.

[0147] The logic and / or steps represented in the flowcharts and / or described herein, for example, can be considered as a sequence of executable instructions stored in a computer readable medium, which can be executed by an instruction execution system, apparatus or device, such as a computer-based system, a processor-based system, or other system that can fetch the instructions from the instruction execution system, apparatus or device and execute the instructions, or a combination of the above. For the purposes of this specification, a "computer readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus or device. The computer readable medium can be a computer readable storage medium or a computer readable signal medium. The computer readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a propagation medium. The computer readable signal medium can include, but is not limited to, a computer readable medium that facilitates transfer of the program from one place to another. A specific example of a computer readable medium is a non-transitory computer-readable storage medium. A specific example of a computer readable signal medium is a source or destination of the computer readable medium. Another specific example of a computer readable signal medium is a computer readable signal travelling through space. Thus, a computer readable medium can take many forms of hardware to carry out the program for use by or in connection with the instruction execution system, apparatus or device.

[0148] It should be understood that aspects of the application can be implemented in hardware, software, firmware or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, the hardware can be implemented using any or a combination of the following technologies, which are each well known in the art: a discrete logic circuit(s) having logic gates for implementing logic functions upon an application of data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), etc.

[0149] Those of skill in the art would understand that the steps of the methods carried out above can be carried out wholly or partly by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, and when executed, includes one or a combination of the steps of the method embodiments.

[0150] In addition, each of the functional units in the various embodiments of the present application can be integrated in one processing module, or each of the units can be physically present separately, or two or more units can be integrated in one module. The integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0151] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A casting process optimization method combining numerical simulation and machine learning, characterized in that, Includes the following steps: Based on the functional, structural, and dimensional requirements of the target aerospace thin-walled casting, the casting material, casting process, and basic process parameters are determined. Based on the shape of the casting structure, a three-dimensional solid model and mesh generation are performed, and the relationship between the physical properties and temperature is determined according to the material of the casting. In the numerical simulation of low-pressure casting, a mechanism model based on the conservation of momentum and energy is used to solve the flow field and temperature field. The volume function method is used to handle the free surface of the flow during the filling process. In the solidification process, the degree of microsegregation during solidification is determined according to the relationship between the solidification solid fraction and temperature, and the simulation results of the solidification temperature field are obtained. Based on the solidification temperature field simulation results, the Shinzan criterion, which represents the liquid penetration and feeding capacity between solidification dendrites, is used to predict shrinkage porosity to obtain the shrinkage porosity volume. Based on the shrinkage porosity volume, a machine learning method based on Gaussian process regression and genetics is used to predict shrinkage porosity under different low-pressure casting process conditions.

2. The method according to claim 1, characterized in that, The probability density function of the machine learning method based on Gaussian process regression and genetics is: Where f(x) is the probability density function; X is a random variable; K is the covariance matrix; n indicates that X is an n-dimensional vector; and μ represents the mean vector.

3. The method according to claim 1, characterized in that, The method of predicting shrinkage porosity under different low-pressure casting process conditions using machine learning methods based on Gaussian process regression and genetics includes: The fitness function is obtained through Gaussian process regression. The fitness function is used as the objective function of the genetic algorithm to search for the optimal combination of objective parameters within a preset range of random parameter combinations through simulated mutation, heredity, and selection. The process of finding the optimal solution is reflected as the adaptation process.

4. The method according to claim 1, characterized in that, Also includes: The simulation results of orthogonal experiments are used as the training set for machine learning, and the results of all experiments are used as the test set. When the machine learning prediction results meet the simulation accuracy requirements, a learning model that can derive the optimal process scheme is constructed.

5. A casting process optimization device combining numerical simulation and machine learning, characterized in that, include: The determination module is used to determine the casting material, casting process and basic process parameters based on the functional, structural and dimensional requirements of the target aerospace thin-walled casting. The simulation module is used to perform three-dimensional solid modeling and mesh generation based on the shape of the casting structure, and to determine the relationship between physical property parameters and temperature according to the casting material. In the numerical simulation of low-pressure casting, the flow field and temperature field are solved by a mechanism model based on the conservation of momentum and energy. The volume function method is used to process the free surface of the flow during the filling process. In the solidification process, the degree of microsegregation during solidification is determined according to the relationship between the solid phase fraction and temperature, and the simulation results of the solidification temperature field are obtained. The optimization module is used to predict shrinkage porosity based on the solidification temperature field simulation results using the Shinsan criterion, which represents the liquid penetration and feeding capacity between solidification dendrites, to obtain the shrinkage porosity volume. Based on the shrinkage porosity volume, a machine learning method based on Gaussian process regression and genetics is used to predict shrinkage porosity under different low-pressure casting process conditions.

6. The apparatus according to claim 5, characterized in that, The probability density function of the machine learning method based on Gaussian process regression and genetics is: Where f(X) is the probability density function; X is a random variable; K is the covariance matrix; n indicates that X is an n-dimensional vector; and μ represents the mean vector.

7. The apparatus according to claim 5, characterized in that, The optimization module includes: The acquisition unit is used to obtain the fitness function through Gaussian process regression. An optimization unit is used to take the fitness function as the objective function of a genetic algorithm to search for the optimal combination of objective parameters within a preset range of random parameter combinations through simulated mutation, heredity, and selection. The process of finding the optimal solution is reflected as an adaptation process.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the casting process optimization method combining numerical simulation and machine learning as described in any one of claims 1-4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the casting process optimization method combining numerical simulation and machine learning as described in any one of claims 1-4.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed to implement the casting process optimization method combining numerical simulation and machine learning as described in any one of claims 1-4.

Images