A foundry process design method and system based on similar process reuse

Abstract

The application provides a foundry process design method and system based on similar process reuse, comprising: obtaining a target foundry three-dimensional model, and performing triangular facetization and attitude normalization processing on the target foundry three-dimensional model; determining the similarity of the target foundry three-dimensional model and each foundry three-dimensional model based on the similarity of the feature vectors of the shape features of the target foundry three-dimensional model and the feature vectors of the shape features of each foundry three-dimensional model in the model library, the similarity of the feature vectors of the domain knowledge of the three-dimensional model and the feature vectors of the domain knowledge of each foundry three-dimensional model, and the weight coefficients of the similarity of each feature vector; sorting each foundry three-dimensional model to select the top k foundry three-dimensional models with high similarity as similar foundry three-dimensional models; and determining the process route of the target foundry based on the process route of the similar foundry three-dimensional models. The application significantly improves the retrieval accuracy, retrieval efficiency and process reuse effect, and can meet the demand of short development cycle of complex foundry.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent manufacturing, and more particularly to a foundry process design method and system based on similar process reuse. BACKGROUND

[0002] The complex castings used in major equipment in the aviation, aerospace, rail transportation and engineering machinery industries have complex process structures such as irregular curved surfaces, blind cavities and multi-dimensional twists, resulting in common problems such as "high difficulty and long cycle of process design" of complex castings, which restricts the speed of three-dimensional model process design of complex castings. The traditional process design method has a long design cycle and a large task amount, and it is difficult to meet the demand for rapid increase in the number and types of castings and the increasingly short development cycle. Similar process reuse is a new path and method to effectively solve this problem.

[0003] When designing the process of a new casting product, the traditional foundry enterprise usually relies on the experience of the designer to search in the process design scheme of the historical casting product already saved, and indirectly find the casting process design scheme of the similar casting product according to the casting product number, batch number and other information. This manual search method requires a high knowledge reserve of the designer and a high familiarity of the designer with the historical products, which leads to a long search process and a low search accuracy, and still cannot meet the demand for a short development cycle of complex castings. SUMMARY

[0004] In view of the defects of the prior art, the purpose of the present application is to provide a foundry process design method and system based on similar process reuse, which aims to solve the problem that the existing foundry process design method based on similar process reuse cannot meet the demand for a short development cycle of complex castings.

[0005] To achieve the above-mentioned purpose, in a first aspect, the present application provides a foundry process design method based on similar process reuse, comprising:

[0006] S101 obtaining a target casting three-dimensional model and performing triangle facetization and attitude normalization processing on the target casting three-dimensional model to obtain a processed target casting three-dimensional model;

[0007] S102 calculating a feature vector of a shape feature of the target casting three-dimensional model and a feature vector of domain knowledge based on the processed target casting three-dimensional model;

[0008] S103 determining the similarity of the target casting three-dimensional model and each casting three-dimensional model based on the similarity of the feature vector of the shape feature of the target casting three-dimensional model and the feature vector of the shape feature of each casting three-dimensional model in the model library, the similarity of the feature vector of the domain knowledge of the three-dimensional model and the feature vector of the domain knowledge of each casting three-dimensional model, and the weight coefficient of each feature vector;

[0009] S104 ranks the casting three-dimensional models based on the similarity of the target casting three-dimensional model and each casting three-dimensional model, to select the top-k casting three-dimensional models with high similarity as similar casting three-dimensional models;

[0010] S105 determines the process route of the target casting based on the process route of the similar casting three-dimensional models.

[0011] In an optional example, the weight coefficient of the similarity of each feature vector is obtained by using a particle swarm algorithm and performing parameter optimization with the maximum top-k retrieval accuracy as the fitness function; the top-k retrieval accuracy is the proportion of the models with the same category as the target casting three-dimensional model in the top-k casting three-dimensional models with high similarity.

[0012] In an optional example, the feature vector of the shape feature of the target casting three-dimensional model includes a D1 operator, a D2 operator and an N2 operator of the target casting three-dimensional model.

[0013] In an optional example, the feature vector of the domain knowledge of the target casting three-dimensional model is obtained by splicing a Mod operator, an Env operator and a Sym operator of the target casting three-dimensional model.

[0014] In a second aspect, the present application provides a casting process design system based on similar process reuse, comprising:

[0015] A model preprocessing module is configured to obtain a target casting three-dimensional model, perform triangle facetization and pose normalization on the target casting three-dimensional model, and obtain a processed target casting three-dimensional model.

[0016] A feature vector calculation module is configured to calculate a feature vector of a shape feature of the target casting three-dimensional model and a feature vector of domain knowledge based on the processed target casting three-dimensional model.

[0017] A similarity calculation module is configured to determine the similarity of the target casting three-dimensional model and each casting three-dimensional model based on the similarity of the feature vector of the shape feature of the target casting three-dimensional model and the feature vector of the shape feature of each casting three-dimensional model in the model library, the similarity of the feature vector of the domain knowledge of the three-dimensional model and the feature vector of the domain knowledge of each casting three-dimensional model, and the weight coefficient of the similarity of each feature vector.

[0018] A similarity ranking module is configured to rank each casting three-dimensional model based on the similarity of the target casting three-dimensional model and each casting three-dimensional model, to select the top-k casting three-dimensional models with high similarity as similar casting three-dimensional models.

[0019] A similar process reuse module is configured to determine the process route of the target casting based on the process route of the similar casting three-dimensional models.

[0020] In an optional example, the weight coefficient of the similarity of each feature vector in the similarity calculation module is obtained by using a particle swarm algorithm and parameter optimization with the maximum top-k retrieval accuracy as the fitness function; the top-k retrieval accuracy is the proportion of the same model category in the first k casting three-dimensional models with high similarity to the target casting three-dimensional model.

[0021] In an optional example, the feature vector of the shape feature of the target casting three-dimensional model in the feature vector calculation module comprises a D1 operator, a D2 operator and an N2 operator of the target casting three-dimensional model.

[0022] In an optional example, the feature vector of the domain knowledge of the target casting three-dimensional model in the feature vector calculation module is obtained based on a Mod operator, an Env operator and a Sym operator of the target casting three-dimensional model.

[0023] Overall, compared with the prior art, the above technical solutions conceived by the present application have the following beneficial effects:

[0024] The present application provides a casting process design method and system based on similar process reuse, which combines the geometric shape features of the three-dimensional model with the casting process design features represented by the domain knowledge, and gives different weight coefficients to the features by using an optimization algorithm, to calculate the similarity of the target casting three-dimensional model and each casting three-dimensional model, realize multi-feature extraction of complex casting three-dimensional models and similar three-dimensional model retrieval driven by shape structure and domain knowledge, and perform process reuse based on the process route of similar casting three-dimensional models, significantly improve the retrieval accuracy, retrieval efficiency and process reuse effect, and meet the short development cycle demand of complex castings such as mine racks and impellers. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart of the casting process design method provided by the present application;

[0026] Figure 2 is an iteration process diagram of the top-3 retrieval accuracy of the three-dimensional model of the multi-domain data set under different meta-heuristic algorithms provided by the present application;

[0027] Figure 3 is an architecture diagram of the casting process design system provided by the present application. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0029] In view of the defects of the prior art, the present application provides a foundry process design method based on similar process reuse, aiming to solve the problem that the mature foundry process is difficult to reuse, and mainly relies on the design personnel to indirectly find the casting process design scheme of similar foundry according to historical experience, foundry product number, batch number and the like, resulting in long retrieval process, low efficiency, high dependence on manual experience, low intelligentization degree, low retrieval accuracy, limited query results and the like.

[0030] Figure 1 is the flowchart of the foundry process design method provided by the present application, as shown in Figure 1 , the method comprises the following steps:

[0031] Step S101, a target foundry three-dimensional model is obtained, and triangle facetization and attitude normalization processing are performed thereon to obtain a processed target foundry three-dimensional model.

[0032] Here, the target foundry is the foundry that needs to be designed. After obtaining the three-dimensional model of the target foundry, the software can be used to perform triangle facetization on the three-dimensional model. Then, considering that the spatial attitude of the three-dimensional model before feature processing is different, resulting in poor feature extraction capability of some feature operators for the three-dimensional model and the like. Therefore, the present application performs attitude normalization preprocessing on the three-dimensional model before feature processing, so that for the same or similar three-dimensional models, their reference coordinate system and the relative position of the model are consistent.

[0033] Step S102, based on the processed target foundry three-dimensional model, the feature vector of the shape feature of the target foundry three-dimensional model and the feature vector of the domain knowledge are calculated;

[0034] Step S103, based on the similarity of the feature vector of the shape feature of the target foundry three-dimensional model and the feature vector of the shape feature of each foundry three-dimensional model in the model library, the similarity of the feature vector of the domain knowledge of the three-dimensional model and the feature vector of the domain knowledge of each foundry three-dimensional model, and the weight coefficient of each feature vector, the similarity of the target foundry three-dimensional model and each foundry three-dimensional model is determined;

[0035] Step S104, based on the similarity of the target foundry three-dimensional model and each foundry three-dimensional model, each foundry three-dimensional model is sorted to select the top k foundry three-dimensional models with high similarity as similar foundry three-dimensional models;

[0036] Step S105, based on the process route of the similar foundry three-dimensional model, the process route of the target foundry is determined.

[0037] Specifically, after obtaining the target casting three-dimensional model processed in step S101, the feature vector of the shape feature of the target casting three-dimensional model and the feature vector of the domain knowledge can be calculated. The shape feature includes the geometric shape feature and the concave-convex degree feature of the casting to be retrieved, and the domain knowledge includes the hot spot feature, the symmetry plane feature and the envelope size of the casting to be retrieved. In addition, the feature vector of the shape feature of each casting three-dimensional model in the model library and the feature vector of the domain knowledge can be calculated in a similar manner. The model library here is constructed according to the three-dimensional models of a plurality of historical casting products.

[0038] By the feature vector of the shape feature of the target casting three-dimensional model and the feature vector of the domain knowledge, a search is performed in the model library to find a similar casting three-dimensional model. On this basis, the process route of the target casting can be determined according to the process route of the similar casting three-dimensional model, so that the casting process design of the target casting is quickly completed through similar process reuse.

[0039] The retrieval process of the target casting three-dimensional model can be specifically that the similarity of the feature vector of the shape feature of the target casting three-dimensional model and the feature vector of the shape feature of each casting three-dimensional model in the model library is calculated, the similarity of the feature vector of the domain knowledge of the three-dimensional model and the feature vector of the domain knowledge of each casting three-dimensional model is calculated, and then the similarity of the target casting three-dimensional model and each casting three-dimensional model is obtained by weighted summation according to the weight coefficients corresponding to the similarity of each feature vector. The weight coefficients can be obtained by parameter optimization using meta-heuristic algorithms such as particle swarm, genetic and ant colony algorithms. Subsequently, according to the similarity of the target casting three-dimensional model and each casting three-dimensional model, each casting three-dimensional model is sorted in descending order of similarity, so that the top k casting three-dimensional models with high similarity are selected as similar casting three-dimensional models.

[0040] The method provided by the embodiment of the application combines the geometric shape feature of the three-dimensional model with the casting process design feature represented by the domain knowledge, and assigns different weight coefficients to the features by using an optimization algorithm to calculate the similarity of the target casting three-dimensional model and each casting three-dimensional model, so as to realize the multi-feature extraction of the complex casting three-dimensional model and the similar three-dimensional model retrieval driven by the shape structure and the domain knowledge, and perform process reuse based on the process route of the similar casting three-dimensional model. The retrieval accuracy, retrieval efficiency and process reuse effect are significantly improved, and the demand for short development cycle of complex castings such as mine racks and impellers can be met.

[0041] Based on the above embodiment, considering the weight coefficient optimization problem for multi-feature fusion, the particle swarm algorithm has the strongest search ability and the fastest convergence speed, and therefore, in the embodiment of the present application, the weight coefficient of the similarity of each feature vector is obtained by using the particle swarm algorithm and parameter optimization with the maximum top-k retrieval accuracy as the fitness function; the top-k retrieval accuracy is the proportion of models of the same category as the target casting three-dimensional model in the top-k casting three-dimensional models with high similarity.

[0042] In the formula, Ra represents the number of three-dimensional models of the same category as the model to be retrieved in the top-k casting three-dimensional models.

[0043] top-k = Ra / k

[0044] In the formula, Ra represents the number of three-dimensional models of the same category as the model to be retrieved in the top-k casting three-dimensional models.

[0045] It should be noted that the existing casting three-dimensional model retrieval method usually adopts a step-by-step elimination mechanism based on feature operators. However, the step-by-step elimination mechanism has great limitations, and the screening order needs to be determined manually according to the characteristics of different operators, and the proportion of elimination at each screening needs to be determined according to the size of the model library, which leads to low retrieval accuracy if the number and types of models in the model library are large, and therefore, the method is not suitable for scenarios with a large number and types of models in the model library in the fields of rail transit and energy equipment.

[0046] The present application adopts the particle swarm algorithm and parameter optimization with the maximum top-k retrieval accuracy as the fitness function to obtain the weight coefficient of the similarity of each feature vector, and calculates the similarity between the target casting three-dimensional model and each casting three-dimensional model based on this, so as to determine the similar casting three-dimensional model. This casting three-dimensional model retrieval method not only improves the retrieval efficiency, but also improves the retrieval accuracy. Moreover, the existing method usually uses precision and recall to evaluate the retrieval result, and if the model library is small, the precision and recall cannot fully represent it, but the top-k retrieval accuracy of the present application can avoid this problem.

[0047] Through experimental data, it is found that the method provided by the present application can improve the retrieval accuracy by 5.24% for multi-fields, by 12.12% for the field of rail transit, by 0.03% for the field of engineering machinery (the model library in this field is small), and by 4.17% for the field of energy equipment.

[0048] Based on any of the above embodiments, the feature vector of the shape feature of the target casting three-dimensional model includes a D1 operator, a D2 operator and an N2 operator of the target casting three-dimensional model.

[0049] The D1 operator and the D2 operator describe global features, and the N2 operator describes surface features, and is used to measure the relationship between surfaces of the three-dimensional model.

[0050] It should be noted that the existing casting three-dimensional model retrieval method usually only uses a shape feature operator, and the shape feature operator only uses a single D1 operator or a single D2 operator, while the present application uses a shape feature operator combining the D1 operator, the D2 operator and the N2 operator, and a process feature operator combining domain knowledge, and the casting three-dimensional model retrieval method combining multiple features greatly improves the accuracy of retrieval. And through experimental verification, the combination of D1 and D2 features has a greater top-k retrieval accuracy than a single D1 operator or a single D2 operator, and the combination of D1 and D2 features is the optimal combination of the shape feature algorithm for complex casting three-dimensional model retrieval.

[0051] Based on any of the above embodiments, the feature vector of the domain knowledge of the target casting three-dimensional model is obtained by splicing a Mod operator, an Env operator and a Sym operator of the target casting three-dimensional model.

[0052] Specifically, the Mod operator describes the volume, wall thickness and other process features of the casting, so as to ensure that the retrieved casting three-dimensional model is similar to the new casting product in terms of process features such as wall thickness, modulus and volume; the Env operator calculates a minimum envelope, representing the three-dimensional size of the casting; and the Sym operator represents the symmetry surface feature of the casting three-dimensional model, and symmetry can improve the efficiency of process design. Since the modulus (Mod) operator, the minimum envelope (Env) operator and the surface symmetry degree (Sym) operator for representing the casting process design features of the complex casting have a low feature dimension, separate feature fusion has little significance in considering the coupling effect between features, and therefore the above three feature vectors are spliced into a six-dimensional feature vector, which is called a process feature operator (Casting process, Cast), that is, a feature vector of domain knowledge.

[0053] The method provided in the embodiments of the present application realizes reduction of the difficulty of particle swarm iteration, while reducing feature loss as much as possible, and the six-dimensional feature vector needs more iteration space.

[0054] Based on any of the above embodiments, in order to perform similar casting retrieval, the similarity between the target casting three-dimensional model and each casting three-dimensional model can be calculated in the following manner:

[0055] SimK=K D1 *imD1 + D2 *im D2 + N2 *im N2 + cast *im cast

[0056] wherein, Sim D1 , Sim D2 , Sim N2 , Sim cast are feature vector similarities of two casting three-dimensional models obtained under D1, D2, N2, Cast operators respectively, the feature vector similarity can be obtained by calculating the Euclidean distance of two feature vectors, K D1 , K D2 , K N2 , K cast are weight coefficients of four feature vector similarities, wherein, K D1 , K D2 , K N2 , K cast range is [0, 1].

[0057] Based on any one of the above embodiments, the application provides a multi-feature casting three-dimensional model retrieval method, comprising the following steps:

[0058] 1. Obtain a three-dimensional model of a casting to be retrieved, and perform triangle facetization and attitude normalization processing thereon;

[0059] 2. Based on the three-dimensional model of the casting to be retrieved after triangle facetization, calculate the shape feature and domain knowledge of the casting to be retrieved;

[0060] 3. Use particle swarm, genetic, and ant colony three meta-heuristic algorithms to optimize feature weight parameters of the shape feature and process feature, and measure the performance of the three complex casting three-dimensional model retrieval algorithms based on top-k retrieval accuracy to find the optimization algorithm and the best feature combination most suitable for the optimization problem of the application;

[0061] 4. The results show that the D1 and D2 feature combination is the optimal combination of the shape feature algorithm for complex casting three-dimensional model retrieval; the N2 operator realizes the quantitative representation of the position relationship between the surfaces of the complex casting three-dimensional model; the Mod operator, the Env operator, and the Sym operator well describe the key process features in the casting process design process; for the weight coefficient optimization problem of multi-feature fusion, the particle swarm algorithm has the strongest search ability and the fastest convergence speed.

[0062] 5. In summary, the application proposes a method of combining the particle swarm algorithm to fuse D1, D2, N2, Mod, Env, and Sym six operators to represent the casting three-dimensional model and retrieve and recommend similar models.

[0063] wherein the shape features include geometric shape features and concave-convex degree features of the cast to be retrieved, and the domain knowledge includes hot spot features, symmetry plane features and envelope size of the cast to be retrieved.

[0064] Preferably, the shape features include D1 operator, D2 operator and N2 operator of the three-dimensional model.

[0065] The calculation method of the D1 operator is as follows:

[0066] The D1 operator represents the three-dimensional model of the cast by statistically representing the distance distribution between a single sampling point and the geometric center of the model.

[0067] Randomly select several points on the surface of the cast to be retrieved, and calculate the distance between each point and the geometric center of the model to obtain an array containing the distances of all points, and normalize the array;

[0068] Set the statistical interval, count the number of point pair distances in the normalized array in each interval, and form the feature vector as the D1 operator of the cast to be retrieved;

[0069] The calculation method of the D2 operator is as follows:

[0070] The D2 operator represents the three-dimensional model of the cast by statistically representing the distance distribution between any two sampling points.

[0071] Randomly select several point pairs on the surface of the cast to be retrieved, and calculate the distance between the two points of each point pair to obtain an array containing the distances of all point pairs, and normalize the array;

[0072] Set the statistical interval, count the number of point pair distances in the normalized array in each interval, and form the feature vector as the D2 operator of the cast to be retrieved;

[0073] The calculation method of the N2 operator is as follows:

[0074] The N2 operator represents the three-dimensional model of the cast by statistically representing the cosine value distribution of the included angle between two random triangular facet vectors.

[0075] Randomly select several facet pairs on the three-dimensional model of the cast to be retrieved, calculate the cosine value of the included angle between the two facet vectors of each facet pair to obtain an array containing all the cosine values;

[0076] Set the statistical interval, count the number of included angle cosine values in the array in each interval, and form the feature vector as the N2 operator of the cast to be retrieved;

[0077] Preferably, the domain knowledge includes Mod operator, Env operator and Sym operator of the three-dimensional model.

[0078] The calculation method of the Mod operator is:

[0079] The Mod operator represents the process characteristics of the three-dimensional model of the casting by describing the volume and wall thickness of the casting.

[0080] The sum of the triangular areas of the casting to be searched is taken as the heat dissipation area of the three-dimensional model of the casting to be searched;

[0081] The volume of the three-dimensional model of the casting to be searched is divided by the heat dissipation area to obtain the Mod operator;

[0082] The calculation method of the Env operator is:

[0083] The Env operator represents the process characteristics of the three-dimensional model of the casting by describing the three-dimensional size of the casting.

[0084] The maximum and minimum values of the three-dimensional model on the X, Y and Z axes in the Cartesian coordinate system are determined by traversing all triangular patches of the three-dimensional model of the casting to be searched, and the envelope size of the three-dimensional model of the casting to be searched along the coordinate principal axis direction is obtained;

[0085] The three-dimensional feature vector is obtained by calculating the width-length ratio, width-height ratio and height-length ratio of the envelope size, and the three-dimensional feature vector is the Env operator.

[0086] The calculation method of the Sym operator is:

[0087] The Sym operator represents the three-dimensional model of the casting by describing the symmetry degree characteristics of the coordinate surface.

[0088] The relationship between the center of gravity of each patch of the casting to be searched and the coordinate surface is compared, and the three-dimensional model of the casting to be searched is segmented along the YOZ surface, the XOZ surface and the XOY surface.

[0089] The surface area error of the two parts of the three-dimensional model segmented along each coordinate surface is calculated as the symmetry degree of the coordinate surface, and the symmetry degrees of the three coordinate surfaces are calculated in turn to form the feature vector of the three-dimensional model, and the feature vector is the Sym operator.

[0090] Since the Mod operator, the Env operator and the Sym operator representing the casting process design characteristics of the complex casting have low feature dimension, separate feature fusion has little significance considering the coupling effect between features, therefore the above three feature vectors are spliced into a six-dimensional feature vector, which is called the process feature operator.

[0091] In the formula, Sim D1 , Sim D2 , Sim N2 , Sim cast are the feature vector similarity of the two three-dimensional models of the casting obtained under the D1, D2, N2 and Cast operators respectively, K D1 , KD2 , K N2 , K cast is the weight coefficient of the similarity of 4 feature vectors, wherein, K D1 , K D2 , K N2 , K cast The range of is [0, 1]. SimK is the total similarity of two three-dimensional models after linear combination, which is used for similar casting retrieval.

[0092] SimK = K D1 *im D1 + D2 *im D2 + N2

[0093] *im N2 + cast *im cast

[0094] top-k represents the accuracy of the top-k three-dimensional models in the retrieval result, which is the same as the category of the model to be retrieved, and the formula is defined as follows:

[0095] top-k = Ra / k

[0096] In the formula, Ra represents the number of three-dimensional models in the retrieval result which are the same as the category of the model to be retrieved, and k represents the total number of three-dimensional models participating in the accuracy calculation. The following takes top-k as top-3 for detailed description.

[0097] The present application uses a multi-field casting three-dimensional model data set of 14 kinds of complex casting three-dimensional models, takes the top-3 three-dimensional model retrieval accuracy maximum as the fitness function, adopts particle swarm, genetic and ant colony algorithms for algorithm training, so as to find the optimization algorithm and the best feature combination most suitable for the optimization problem of the present application, and the top-3 retrieval accuracy iterative process of the multi-field data set three-dimensional model under different heuristic algorithms is as shown in Figure 2

[0098] Figure 2 ​The results show that: for the weight coefficient optimization problem of the multi-feature fusion of the application, the performance of the particle swarm algorithm is optimal, the particle swarm algorithm converges after about 25 iterations, the top-3 retrieval accuracy of the 14 classes and 83 kinds of complex casting three-dimensional models of the multi-field data set is 90%, which is the optimal solution of the three optimization algorithms, and the global search ability of the particle swarm algorithm is the strongest. The optimization ability of the ant colony algorithm is second, and the retrieval accuracy reaches the same level as the particle swarm algorithm at about the 50th generation, but the accuracy fluctuates between 77% and 86% in the process of 100 iterations, the convergence speed is slow, and the algorithm does not converge in 100 iterations. The performance of the genetic algorithm is the worst, the retrieval accuracy of the optimal solution of the weight found in the iteration process is 86.3%, and the accuracy decreases continuously in the iteration process, which shows that the global search ability is the worst, and it is not suitable for solving the weight coefficient optimization problem of the application. Therefore, for the multi-feature fusion weight coefficient optimization problem of the application, the search ability of the particle swarm algorithm is the strongest, and the convergence speed is the fastest.

[0099] In summary, the three-dimensional model multi-feature retrieval method combining the particle swarm algorithm and fusing D1, D2, N2, Mod, Env and Sym six operators can better represent the characteristics of the casting three-dimensional model, and can retrieve and recommend similar models with high accuracy.

[0100] The application proposes a three-dimensional model multi-feature retrieval method combining the particle swarm optimization for the "similar process reuse" problem of complex castings in the typical application field of casting; the geometric shape features of the three-dimensional model are extracted through D1 operator, D2 operator, N2 operator and other shape distribution operators; the casting process design features of the three-dimensional model are extracted through Mod operator, Env operator, Sym operator and other process feature operators. Meanwhile, the geometric shape features and the casting process design features of the three-dimensional model are combined to realize the multi-feature extraction of the complex casting three-dimensional model and the similar three-dimensional model retrieval driven by shape structure and field knowledge.

[0101] In the feature combination of D1 operator, D2 operator, N2 operator, Mod operator, Env operator and Sym operator, the particle swarm algorithm is used to give different weight coefficients to the features, the top-3 three-dimensional model retrieval accuracy is taken as the optimization target, the best feature weight coefficient most suitable for the optimization problem of the application is found, and the most similar complex casting three-dimensional model is obtained, which significantly improves the retrieval accuracy and the process reuse effect compared with the traditional indirect search of similar three-dimensional model casting process scheme according to historical experience, casting product number, batch number and other information.

[0102] Based on any of the above embodiments, the spatial pose of the three-dimensional model before feature processing is different, resulting in poor feature extraction capability of some feature operators for the three-dimensional model. Therefore, the three-dimensional model is subjected to pose normalization preprocessing before feature processing, so that for the same or similar three-dimensional models, their reference coordinate system and the relative position of the model are consistent.

[0103] The pose normalization of the three-dimensional model mainly includes two parts: ① translating the coordinate system origin to the center of gravity of the three-dimensional model to realize the translational invariance of the three-dimensional model; and ② determining the position of the main shaft to realize the rotational invariance of the three-dimensional model. However, it should be ensured that the relative positions of the main shafts of the casting models of the same structure type are consistent, so as to ensure the segmentation effect of the subsequent model.

[0104] Let the total number of vertices of the three-dimensional model after faceting be P, and the total number of triangular facets be N, then the vertices of each triangular facet are A i (x i ,y i ,z i ), and the center of gravity of the three-dimensional model is denoted as G M (G xm ,G ym ,G zm ), then the translation matrix is denoted as M T .

[0105]

[0106]

[0107] In order to ensure that the relative positions of the main shafts of the casting models of the same structure type are consistent after rotational normalization, the principal component analysis algorithm considering the area of the facet is adopted to determine the rotation matrix of the model. The algorithm finds the principal component direction of the spatial distribution of the vertices of the three-dimensional model as the new main shaft direction of the model, that is, it ensures that the dispersion degree of all vertices projected onto the principal component direction is maximum, and the correlation between the principal component directions is minimum. Let the area of each facet of the three-dimensional model be S i , and the covariance matrix of the vertices be M, then the rotation matrix M R of the three-dimensional model is composed of three unitized eigenvectors [a i ,b i ,c i ] of M. In particular, the three eigenvectors of the rotation matrix are arranged in ascending order of eigenvalues in the present application, so the dispersion degrees of the vertices in the main shaft directions are Z axis, Y axis and X axis in turn.

[0108]

[0109]

[0110] The translation matrix and the rotation matrix are applied to the three-dimensional model, and the preprocessed three-dimensional model is obtained.

[0111] Based on any of the above embodiments, in order to illustrate that D1 and D2 are better than D1 or D2 alone, the present application carries out shape feature fusion retrieval experiment and effect analysis:

[0112] The present application only uses two feature vectors obtained by D1 and D2 shape feature operators to quantitatively represent the three-dimensional model of the casting, and uses particle swarm, genetic and ant colony algorithms to optimize the two weight coefficients K D1 、K D2 in the similarity formula, and Table 1, Table 2 and Table 3 are super parameters used by the above three optimization algorithms.

[0113] Table 1: Particle swarm algorithm super parameter setting

[0114]

[0115] Table 2: Genetic algorithm super parameter setting

[0116]

[0117] Table 3: Ant colony algorithm super parameter setting

[0118]

[0119] Table 4: D1, D2 shape feature combination complex casting three-dimensional model retrieval algorithm accuracy

[0120]

[0121] In the four three-dimensional model data sets of the engineering machinery field, the rail transit field, the energy equipment field and multiple fields, the particle swarm, the genetic and the ant colony algorithms are used to optimize the weight coefficients of the two shape features D1 and D2 by taking the top-3 retrieval accuracy of the three-dimensional model as the fitness function, and the top-3 retrieval accuracy of the three-dimensional model of each data set has no obvious change, so it is considered that the weight coefficients of the two shape features D1 and D2 do not need to be considered when the two shape features are combined. Table 4 is the optimal retrieval accuracy of the complex casting three-dimensional model in the four data sets under the combination of the D1 and D2 operator features, and the results show that the combination of the D1 and D2 operators can improve the retrieval accuracy of the complex casting three-dimensional model to a certain extent. Compared with the single feature retrieval algorithm, the top-3 retrieval accuracy of the casting three-dimensional model in multiple fields is improved by about 18.5%. Without considering the optimization of the shape feature algorithm for the complex casting surface shape feature and the process design feature, the combination of the D1 and D2 features is the optimal combination of the shape feature algorithm for the retrieval of the complex casting three-dimensional model.

[0122] Based on any one of the above embodiments, the application provides a casting process design system based on similar process reuse, Figure 3 is the architecture diagram of the casting process design system provided by the application, as Figure 3 shown, the system comprises:

[0123] A model preprocessing module 310 is configured to acquire a target casting three-dimensional model, perform triangle facetization and pose normalization on the target casting three-dimensional model, and obtain a processed target casting three-dimensional model.

[0124] A feature vector calculation module 320 is configured to calculate a feature vector of a shape feature of the target casting three-dimensional model and a feature vector of domain knowledge based on the processed target casting three-dimensional model.

[0125] A similarity calculation module 330 is configured to determine the similarity between the target casting three-dimensional model and each casting three-dimensional model based on the similarity between the feature vector of the shape feature of the target casting three-dimensional model and the feature vector of the shape feature of each casting three-dimensional model in the model library, the similarity between the feature vector of the domain knowledge of the three-dimensional model and the feature vector of the domain knowledge of each casting three-dimensional model, and the weight coefficient of the similarity of each feature vector.

[0126] A similarity sorting module 340 is configured to sort each casting three-dimensional model based on the similarity between the target casting three-dimensional model and each casting three-dimensional model, so as to select the top k casting three-dimensional models with high similarity as similar casting three-dimensional models.

[0127] A similar process reuse module 350 is configured to determine the process route of the target casting based on the process route of the similar casting three-dimensional model.

[0128] The system provided by the embodiment of the present application combines the geometric feature of the three-dimensional model with the casting process design feature represented by the domain knowledge, and adopts an optimization algorithm to assign different weight coefficients to the features, to calculate the similarity of the target casting three-dimensional model and each casting three-dimensional model, to realize the shape structure and domain knowledge driven three-dimensional model multi-feature extraction and similar three-dimensional model retrieval of the complex casting, and to perform process reuse based on the process route of the similar casting three-dimensional model, thereby significantly improving the retrieval accuracy, retrieval efficiency and process reuse effect, and meeting the short development cycle requirement of the complex castings such as mine racks and impellers.

[0129] It can be understood that the detailed function implementation of each module can refer to the description in the foregoing method embodiments, and will not be described herein.

[0130] In addition, the embodiment of the present application provides another casting process design device based on similar process reuse, which comprises a memory and a processor.

[0131] The memory is used to store a computer program.

[0132] The processor is used to implement the method in the foregoing embodiments when executing the computer program.

[0133] In addition, the present application also provides a computer readable storage medium, and the storage medium stores a computer program, and when the computer program is executed by a processor, the method in the foregoing embodiments is implemented.

[0134] Based on the method in the foregoing embodiments, the embodiment of the present application provides a computer program product, and when the computer program product is run on a processor, the processor executes the method in the foregoing embodiments.

[0135] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not used to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A foundry process design method based on similar process reuse, characterized by, The method comprises the following steps: S101: obtaining a target casting three-dimensional model, and performing triangular facetization and attitude normalization processing on the target casting three-dimensional model to obtain a processed target casting three-dimensional model; S102: calculating a feature vector of a shape feature of the target casting three-dimensional model and a feature vector of domain knowledge based on the processed target casting three-dimensional model; S103: determining a similarity between the target casting three-dimensional model and each casting three-dimensional model based on a similarity between the feature vector of the shape feature of the target casting three-dimensional model and a feature vector of a shape feature of each casting three-dimensional model in a model library, a similarity between the feature vector of the domain knowledge of the three-dimensional model and a feature vector of domain knowledge of each casting three-dimensional model, and a weight coefficient of the similarity of each feature vector; the weight coefficient is used for weighted summation; S104: sorting each casting three-dimensional model based on the similarity between the target casting three-dimensional model and each casting three-dimensional model, so as to select the first k casting three-dimensional models with large similarity as similar casting three-dimensional models; S105: determining a process route of the target casting based on a process route of the similar casting three-dimensional model; The weight coefficient of the similarity of each feature vector is obtained by using a particle swarm algorithm and performing parameter optimization on a top-k retrieval accuracy maximum as an adaptive function; the top-k retrieval accuracy is a proportion of models with the same category as the target casting three-dimensional model in the first k casting three-dimensional models with large similarity; The particle swarm algorithm is found based on the top-k retrieval accuracy among three meta-heuristic algorithms, and the three meta-heuristic algorithms comprise a particle swarm algorithm, an ant colony algorithm and a genetic algorithm.

2. The method of claim 1, wherein, The feature vector of the shape feature of the target casting three-dimensional model comprises a D1 operator, a D2 operator and an N2 operator of the target casting three-dimensional model.

3. The method of claim 1, wherein, The feature vector of the domain knowledge of the target casting three-dimensional model is obtained by splicing a Mod operator, an Env operator and a Sym operator of the target casting three-dimensional model.

4. A foundry process design system based on similar process reuse, characterized by, The method comprises the following steps: The model preprocessing module is used for obtaining a target casting three-dimensional model, and performing triangular facetization and attitude normalization processing on the target casting three-dimensional model to obtain a processed target casting three-dimensional model; The feature vector calculation module is used for calculating a feature vector of a shape feature of the target casting three-dimensional model and a feature vector of domain knowledge based on the processed target casting three-dimensional model; The similarity calculation module is used for determining a similarity between the target casting three-dimensional model and each casting three-dimensional model based on a similarity between the feature vector of the shape feature of the target casting three-dimensional model and a feature vector of a shape feature of each casting three-dimensional model in a model library, a similarity between the feature vector of the domain knowledge of the three-dimensional model and a feature vector of domain knowledge of each casting three-dimensional model, and a weight coefficient of the similarity of each feature vector; the weight coefficient is used for weighted summation; The similarity sorting module is used for sorting each casting three-dimensional model based on the similarity between the target casting three-dimensional model and each casting three-dimensional model, so as to select the first k casting three-dimensional models with large similarity as similar casting three-dimensional models; The similar process reuse module is used for determining a process route of the target casting based on a process route of the similar casting three-dimensional model. The weight coefficient of the similarity of each feature vector in the similarity calculation module is obtained by using a particle swarm algorithm and parameter optimization with a maximum top-k retrieval accuracy as a fitness function; the top-k retrieval accuracy is a proportion of models of the same category as the target casting three-dimensional model in the first k casting three-dimensional models with high similarity; The particle swarm algorithm is found based on the top-k retrieval accuracy in three meta-heuristic algorithms, including a particle swarm algorithm, an ant colony algorithm and a genetic algorithm.

5. The system of claim 4, wherein, The feature vector of the shape feature of the target casting three-dimensional model in the feature vector calculation module includes a D1 operator, a D2 operator and an N2 operator of the target casting three-dimensional model.

6. The system of claim 4, wherein, The feature vector of the domain knowledge of the target casting three-dimensional model in the feature vector calculation module is obtained based on splicing of a Mod operator, an Env operator and a Sym operator of the target casting three-dimensional model.

Images