Processing method and device for synthetic casting model

Abstract

The invention provides a processing method and device for a synthetic casting model. The method comprises the steps that at least two to-be-processed three-dimensional casting models are obtained; respectively inputting the at least two three-dimensional casting models into a preset casting prediction model for feature extraction analysis to obtain feature maps of different network levels, which are output by the casting prediction model and respectively correspond to the at least two three-dimensional casting models; processing is carried out based on the feature maps of the different network levels, and casting feature sub-models corresponding to the at least two three-dimensional casting models respectively are obtained; and based on the casting feature sub-models corresponding to the at least two three-dimensional casting models, splicing synthesis processing is carried out, and a new casting model is obtained. According to the processing method for synthesizing the casting model, the efficiency of synthesizing the casting model can be effectively improved, so that the workload of synthesizing the casting model is reduced.

Description

Technical Field

[0001] The present invention relates to the field of casting technology, and in particular to a method and device for processing a synthetic casting model, and also to an electronic device, a non-transitory computer-readable storage medium and a computer program product. Background Art

[0002] In recent years, with the continuous development of computer industry technology, various casting models have been used more and more widely in the industrial field. There are many types of casting models, and it is difficult to fully reflect the characteristics of castings by collecting and organizing them. Artificial modeling requires constructing structures out of thin air, and the workload is very huge. Numerical simulation technology is widely used in the casting field, but it has limitations such as cumbersome implementation process and long calculation time, resulting in low efficiency of synthetic casting models. Therefore, how to design an efficient processing solution for synthetic casting models has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The invention provides a processing method for a synthetic casting model, which is used to solve the defect that the synthetic casting model scheme in the prior art has high limitations, resulting in poor efficiency of the actual synthetic casting model.

[0004] The present invention provides a method for processing a synthetic casting model, comprising: obtaining at least two three-dimensional casting models to be processed; The at least two three-dimensional casting models are respectively input into a preset casting prediction model for feature extraction and analysis, and feature graphs of different network levels corresponding to the at least two three-dimensional casting models are obtained, which are output by the casting prediction model; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on sample three-dimensional casting models and feature prediction result labels corresponding to the sample three-dimensional casting models in advance; Based on the feature graphs of different network levels, processing is performed to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively; based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, splicing and synthesis processing is performed to obtain a new casting model.

[0005] According to a method for processing a synthetic casting model provided by the present invention, the processing based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively includes: Screening the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels to obtain target feature graphs of different network levels after the value screening process; Connectivity detection is performed based on the target feature graphs of different network levels after the screening process to obtain models of a preset number of connected areas, and the models of the connected areas are used as the casting feature sub-models.

[0006] According to a processing method for a synthetic casting model provided by the present invention, the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels are screened to obtain target feature graphs of different network levels after the value screening process, specifically comprising: Screening the feature graphs of different network layers to obtain the screened feature graphs; The values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

[0007] According to a method for processing a synthetic casting model provided by the present invention, the target feature graphs of different network levels after the screening process are used to perform connectivity detection to obtain a model of a preset number of connected areas, specifically comprising: Performing connectivity detection based on the target feature graphs of different network levels after the screening process to obtain connectivity detection results of the target feature graphs of different network levels; According to the size of the connected areas in the connectivity detection result, the connected areas are sorted in descending order to obtain a sorting result; models of a preset number of first N connected areas are obtained from the sorting result; wherein N is a positive integer greater than or equal to 2.

[0008] According to a processing method for a synthetic casting model provided by the present invention, the values ​​in the three-dimensional array corresponding to the feature graph after the screening process are screened to obtain target feature graphs of different network levels after the value screening process, specifically comprising: According to the quantitative characteristics of all values ​​in the three-dimensional array corresponding to the feature map after the screening process, a value threshold is determined; based on the value threshold, the values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

[0009] According to a processing method for synthesizing a casting model provided by the present invention, the method of performing splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models to obtain a new casting model specifically includes: Acquire size and position information of casting feature sub-models respectively corresponding to the at least two three-dimensional casting models; Based on the size and position information of the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models, the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are rotated, scaled and / or translated respectively, so that the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are spliced ​​and synthesized to obtain a new casting model.

[0010] The present invention also provides a processing device for a synthetic casting model, comprising: A casting model obtaining unit, used for obtaining at least two three-dimensional casting models to be processed; A casting model feature extraction unit, used for inputting the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtaining feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on sample three-dimensional casting models and feature prediction result labels corresponding to the sample three-dimensional casting models in advance; The casting model processing unit is used to perform processing based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively; and perform splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively to obtain a new casting model.

[0011] According to a processing device for a synthetic casting model provided by the present invention, the casting model processing unit is specifically used for: Screening the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels to obtain target feature graphs of different network levels after the value screening process; Connectivity detection is performed based on the target feature graphs of different network levels after the screening process to obtain models of a preset number of connected areas, and the models of the connected areas are used as the casting feature sub-models.

[0012] According to a processing device for a synthetic casting model provided by the present invention, the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels are screened to obtain target feature graphs of different network levels after the value screening process, specifically comprising: Screening the feature graphs of different network layers to obtain the screened feature graphs; The values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

[0013] According to a processing device for a synthetic casting model provided by the present invention, the connectivity detection is performed based on the target feature graphs of different network levels after the screening process to obtain a model of a preset number of connected areas, specifically including: Performing connectivity detection based on the target feature graphs of different network levels after the screening process to obtain connectivity detection results of the target feature graphs of different network levels; According to the size of the connected areas in the connectivity detection result, the connected areas are sorted in descending order to obtain a sorting result; models of a preset number of first N connected areas are obtained from the sorting result; wherein N is a positive integer greater than or equal to 2.

[0014] According to a processing device for a synthetic casting model provided by the present invention, the filtering process is performed on the values ​​in the three-dimensional array corresponding to the feature graph after the filtering process to obtain target feature graphs of different network levels after the value filtering process, specifically comprising: According to the quantitative characteristics of all values ​​in the three-dimensional array corresponding to the feature map after the screening process, a value threshold is determined; based on the value threshold, the values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

[0015] According to a processing device for a synthetic casting model provided by the present invention, the casting model processing unit is specifically used for: Acquire size and position information of casting feature sub-models respectively corresponding to the at least two three-dimensional casting models; Based on the size and position information of the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models, the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are rotated, scaled and / or translated respectively, so that the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are spliced ​​and synthesized to obtain a new casting model.

[0016] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for processing a synthetic casting model as described in any one of the above items is implemented.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processing method for synthesizing a casting model as described in any one of the above items is implemented.

[0018] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the processing method for synthesizing a casting model as described in any one of the above items is implemented.

[0019] The processing method of the synthetic casting model provided by the present invention obtains at least two three-dimensional casting models to be processed, inputs the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, obtains feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models respectively, processes based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, and performs splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, so as to quickly obtain a new casting model, which can effectively improve the efficiency of synthesizing the casting model, thereby reducing the workload of synthesizing the casting model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 It is one of the flow diagrams of the method for processing the synthetic casting model provided by the present invention.

[0022] Figure 2 This is the second flow chart of the method for processing the synthetic casting model provided by the present invention.

[0023] Figure 3 It is a schematic diagram of a three-dimensional casting model provided by the present invention.

[0024] Figure 4 It is a schematic diagram of another three-dimensional casting model provided by the present invention.

[0025] Figure 5 It is a schematic diagram of the positions of two three-dimensional casting models provided by the present invention.

[0026] Figure 6 It is a schematic diagram of the new casting model provided by the present invention.

[0027] Figure 7 It is the feature map of the third convolutional layer channel 5 and its reference value provided by the present invention.

[0028] Figure 8 It is the feature map of the 8th convolutional layer channel 2 and its reference value provided by the present invention.

[0029] Fig. 9It is the feature map of the 8th convolutional layer channel 15 and its reference value provided by the present invention.

[0030] Fig.10 It is the casting feature sub-model and its reference value of the 8th convolutional layer channel 2 provided by the present invention.

[0031] Fig.11 It is a structural schematic diagram of a processing device for a synthetic casting model provided by the present invention.

[0032] Fig.12 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] Combine the following Figure 1-Figure 12 The processing method and device of the composite casting model of the present invention are described, and the embodiments thereof are described in detail.

[0035] The following is a detailed description of an embodiment of the method for processing a synthetic casting model according to the present invention. Figure 1 As shown, it is one of the flow diagrams of the method for processing the synthetic casting model provided by the present invention, and the specific implementation process includes the following steps: Step 101: Obtain at least two three-dimensional casting models to be processed.

[0036] In the embodiment of the present invention, before executing this step, it is necessary to perform model training in advance to obtain a trained casting prediction model. Specifically, firstly obtain a sample three-dimensional casting model, and determine the feature prediction result label corresponding to the sample three-dimensional casting model, and then iteratively train the initial casting prediction model based on the sample three-dimensional casting model and the feature prediction result label corresponding to the sample three-dimensional casting model. When the loss function meets the preset conditions, the trained casting prediction model is obtained. Figure 2 The sample three-dimensional casting model shown may include casting model 1, casting model 2, and casting model 3. The initial casting prediction model may be as follows: Figure 2 The portion of the convolutional neural network shown.

[0037] It should be noted that the sample three-dimensional casting model can select multiple existing typical three-dimensional casting models, and set a three-dimensional array with the same side length in three directions, and discretize the multiple typical three-dimensional casting models in the three-dimensional space corresponding to the three-dimensional array. Set the position of the three-dimensional casting model in the three-dimensional array and other positions to different values. Establish an initial casting prediction model (i.e., an initial convolutional neural network model), use multi-level convolution operations to extract features, input the three-dimensional casting model, and normalize the data of the three-dimensional casting model during input. Specifically, supervised or unsupervised learning can be selected. Supervised learning can divide known castings into multiple categories, and the labels are set to the serial numbers of the casting categories. Unsupervised learning can set the labels to the input itself. In the model training process, the sample three-dimensional casting model can be selected to train the above-mentioned initial casting prediction model, and the training is performed until the loss function reaches a smaller value. After the training is completed, the casting prediction model is obtained. In the actual implementation process, a typical three-dimensional casting model a can be selected as a sample three-dimensional casting model. A three-dimensional array with side lengths of b in three directions is set, and the shape of the three-dimensional casting model is discretized into a grid pattern in the three-dimensional space corresponding to the three-dimensional array and scaled so that the three-dimensional casting model can be included in the three-dimensional array. The value of the corresponding position in the three-dimensional array is set to c according to the three-dimensional space occupied by the shape of different three-dimensional casting models, and the remaining positions are set to d, and the positions of c and d represent the casting and the mold respectively. The initial casting prediction model is trained using the sample three-dimensional casting model until the loss function reaches a smaller value. For example, when the loss function is the MSE (Mean Squared Error) loss function, the maximum threshold of the loss function can be set to q. After the training is completed, the casting prediction model is obtained. Among them, the value ranges of the above parameters can be as follows: a range of 10-100, b range of 16-256, k range of 1-4, m range of 2-10, n range of 1-5, q range of 0.1-0.0001, p range of 2-10.

[0038] For example, 10 existing typical three-dimensional casting models can be selected. Rotate them to obtain three upward forms in the x, y, and z directions, respectively, and obtain 30 three-dimensional casting models. Copy them once to obtain data of 60 three-dimensional casting models. Convert the three-dimensional casting model into a three-dimensional array with a side length of 128, set the outside of the three-dimensional casting model as a casting mold, and take the values ​​of 9 and 1 for the casting part and the casting mold part respectively. Establish an initial convolutional neural network model based on U-net, using 3 encoder and decoder structures, each structure includes 2 convolution operations, the input and output are both casting models, the validation set ratio is 0.5, and the data is normalized when input. Use the existing typical three-dimensional casting model as a sample three-dimensional casting model to train the above initial convolutional neural network model, train 80 epochs, and the loss value of the loss function can be 0.0733.

[0039] In an embodiment of the present invention, the at least two three-dimensional casting models to be processed are three-dimensional casting models of different shapes. When there are two three-dimensional casting models, the three-dimensional casting models may include a casting model A and a casting model B. The casting model A is specifically as follows: Figure 3 As shown, the casting model B is as follows Figure 4 As shown, the relative positions of casting model A and casting model B are as follows Figure 5 In actual implementation, the three-dimensional casting model can be a three-dimensional casting model of a blade and a three-dimensional casting model of a cylinder respectively.

[0040] Step 102: respectively input the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtain feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on sample three-dimensional casting models and feature prediction result labels corresponding to the sample three-dimensional casting models.

[0041] In an embodiment of the present invention, the at least two three-dimensional casting models are respectively input into a preset casting prediction model for feature extraction and analysis, and for any three-dimensional casting model (ie, casting), feature maps at different network levels and different channels are obtained.

[0042] Step 103: Processing is performed based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively; splicing and synthesis processing is performed based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively to obtain a new casting model.

[0043] In an embodiment of the present invention, when processing based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models, the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels can be screened to obtain target feature graphs of different network levels after the value screening process, and connectivity detection can be performed based on the target feature graphs of different network levels after the screening process to obtain models of a preset number of connected areas, and the models of the connected areas can be used as the casting feature sub-models. In addition, when splicing and synthesizing the casting feature sub-models corresponding to the at least two three-dimensional casting models to obtain a new casting model, the size and position information of the casting feature sub-models corresponding to the at least two three-dimensional casting models can be obtained; based on the size and position information of the casting feature sub-models corresponding to the at least two three-dimensional casting models, the casting feature sub-models corresponding to the at least two three-dimensional casting models can be rotated, scaled and / or translated, so that the casting feature sub-models corresponding to the at least two three-dimensional casting models can be spliced ​​and synthesized to obtain a new casting model. Figure 2 As shown, the casting feature sub-model may include a feature model Mk1 and a feature model Mk2.

[0044] Among them, the values ​​in the three-dimensional array corresponding to the feature maps of the different network levels are screened to obtain the target feature maps of the different network levels after the value screening process. The corresponding implementation process includes: screening the feature maps of the different network levels to obtain the feature maps after the screening process; screening the values ​​in the three-dimensional array corresponding to the feature maps after the screening process to obtain the target feature maps of the different network levels after the value screening process. Specifically, in the process of screening the feature maps of the different network levels to obtain the feature maps after the screening process, the specific screening method adopted is, for example: for the feature maps of different channels of a certain level, a series of values ​​{Pi} are obtained by performing global average pooling on each feature map, and these values ​​are normalized by the softmax function to represent the weights of different channels. For feature maps of different levels, different series of values ​​are obtained by independent calculation, and the p feature maps with the highest values ​​are selected in each series.

[0045] It should be noted that, in the process of filtering the values ​​in the three-dimensional array corresponding to the feature graph after the filtering process to obtain the target feature graphs of different network levels after the value filtering process, the value threshold can be determined according to the quantity characteristics of all values ​​in the three-dimensional array corresponding to the feature graph after the filtering process; the values ​​in the three-dimensional array corresponding to the feature graph after the filtering process are filtered based on the value threshold to obtain the target feature graphs of different network levels after the value filtering process. In addition, the target feature graphs of different network levels after the filtering process are subjected to connectivity detection to obtain the models of the preset number of connected areas, and the corresponding implementation process includes: performing connectivity detection based on the target feature graphs of different network levels after the filtering process to obtain the connectivity detection results of the target feature graphs of different network levels; sorting the connected areas in the order from large to small according to the size of the connected areas in the connectivity detection results to obtain the sorting results; obtaining the models of the preset number of the first N connected areas from the sorting results; wherein N is a positive integer greater than or equal to 2.

[0046] Among them, in the process of determining the value threshold according to the quantitative characteristics of all values ​​in the three-dimensional array corresponding to the feature graph after the screening process, the values ​​of different positions of the three-dimensional array of each feature graph can be arranged from small to large, and the value of the most number in the corresponding array is found, and the value is set as the value threshold W1. By detecting the number of numbers above / below the threshold in the casting part of the array, the side with more numbers is selected. Subsequently, by finding the position of the first n values ​​with the largest mutation in this sequence, or finding the values ​​of the first n positions of the numbers in the array except W1. For example, the difference x1 and x2 between the values ​​corresponding to the numbers at a certain position and the numbers on the left and right can be compared, and the first n values ​​with the largest absolute value of the other value when one of the two is zero can be found. Any one of the n values ​​is used as the threshold W2, and the side with more numbers is selected. Then, connectivity detection is performed. If there are multiple connected areas, the models of the largest first N connected areas are retained, and the models of these connected areas obtained are the casting feature submodels of the three-dimensional casting model, that is, the {Mki} model. Furthermore, multiple casting MKi models of different shapes (i.e., casting feature sub-models of the three-dimensional casting model) are selected as sub-models for synthesizing a new casting model. The algorithm ensures that these can contact each other. For example, find the coordinates (a1, a2, a3), (b1, b2, b3) of the center points of the two casting feature sub-models A and B, translate all the coordinates of the casting feature sub-model B by (a1-b1, a2-b2, 2a3) units, move the casting feature sub-model B up and down so that it contacts the casting feature sub-model A and is directly above A, and finally splice them together to form a new casting model. The setting standard of the value threshold is that the local or overall features of the model can be clearly seen.

[0047] In a complete embodiment of the present invention, the three-dimensional casting model of the blade and the three-dimensional casting model of the cylinder can be input into the casting prediction model for feature extraction and analysis, and the feature graphs of different network levels during prediction are output. The feature graphs of different network levels are screened. Figure 7 The feature map of the 3rd convolutional layer channel 5 of casting A (i.e., one of the 3D casting models) is shown in Figure 8 The feature map of channel 2 of the 8th convolutional layer of casting A is shown in Fig. 9 The feature map of channel 15 of the 8th convolutional layer of casting A is shown in Fig.10 The casting feature submodel of the 8th convolutional layer channel 2 of casting A is shown in the figure. The values ​​at different positions of the three-dimensional array of each feature map are arranged from small to large, and the value with the most digits in the corresponding three-dimensional array is found. The value is set as the threshold, and the part above the threshold is retained, and then the same screening method is repeated several times. For example, the final threshold of the 8th convolutional layer channel 2 is 0.3291. Then, a connectivity check is performed. If there are multiple connected areas, the models of the largest connected areas are retained. The models of these connected areas are the casting feature submodels of the three-dimensional casting model. The visualization graph of the feature submodel of the 8th convolutional layer channel 2 is shown. Select two casting feature sub-models of different shapes as the sub-models of the new synthetic casting model, find the coordinates of the center points of the two casting feature sub-models A and B (7, 7, 4), (3, 3, 7), translate all the coordinates of the casting feature sub-model B by (4, 4, 8) units, move the casting feature sub-model B up and down so that it contacts the casting feature sub-model A and is directly above A, and finally splice them together to form a new casting model. Figure 4 As shown, it is based on the splicing and synthesis processing of the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively to obtain a new casting model, and the new casting model can be as follows Figure 6 shown.

[0048] The embodiment of the present invention provides a casting heat transfer simulation method, device, medium and equipment based on deep learning, including the steps of: selecting a typical three-dimensional casting model and converting it into a three-dimensional array, establishing a convolutional neural network (i.e., an initial convolutional neural network model), using a typical three-dimensional casting model to train the convolutional neural network, inputting at least two three-dimensional casting models to be processed to obtain feature graphs of different network levels, screening the feature graphs in terms of network level, channel, value, and connectivity to obtain a core feature model (i.e., a casting feature sub-model), processing the casting feature sub-model and splicing it into a new casting model. The method can effectively and quickly generate a casting model. The present invention is suitable for feature extraction of existing three-dimensional casting models, can realize the extraction of basic features of three-dimensional casting models, and reconstruct and synthesize new casting models based on these basic features, thereby realizing the generation of massive new casting models, solving the problem of building massive models of training sets required for simulation based on artificial intelligence, and can realize the automatic generation of a large number of three-dimensional casting models to meet the deep learning training set.

[0049] In summary, the processing method of the synthetic casting model provided by the present invention obtains at least two three-dimensional casting models to be processed, and inputs the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, obtains feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models respectively, processes based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, and performs splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, to quickly obtain a new casting model, which can effectively improve the efficiency of synthesizing the casting model, thereby reducing the workload of synthesizing the casting model.

[0050] The following is a description of the processing device for the synthetic casting model provided by the present invention. The processing device for the synthetic casting model described below and the processing method for the synthetic casting model described above can be referred to in correspondence with each other. Fig.11 As shown, it is a schematic diagram of the structure of the processing device of the synthetic casting model provided by the present invention. The processing device of the synthetic casting model of the present invention specifically includes the following parts: The casting model obtaining unit 1101 is used to obtain at least two three-dimensional casting models to be processed.

[0051] The casting model feature extraction unit 1102 is used to input the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtain feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models respectively; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on a sample three-dimensional casting model and a feature prediction result label corresponding to the sample three-dimensional casting model.

[0052] The casting model processing unit 1103 is used to perform processing based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively; and perform splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively to obtain a new casting model.

[0053] According to a processing device for a synthetic casting model provided by the present invention, the casting model processing unit is specifically used for: Screening the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels to obtain target feature graphs of different network levels after the value screening process; Connectivity detection is performed based on the target feature graphs of different network levels after the screening process to obtain models of a preset number of connected areas, and the models of the connected areas are used as the casting feature sub-models.

[0054] According to a processing device for a synthetic casting model provided by the present invention, the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels are screened to obtain target feature graphs of different network levels after the value screening process, specifically comprising: Screening the feature graphs of different network layers to obtain the screened feature graphs; The values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

[0055] According to a processing device for a synthetic casting model provided by the present invention, the connectivity detection is performed based on the target feature graphs of different network levels after the screening process to obtain a model of a preset number of connected areas, specifically including: Performing connectivity detection based on the target feature graphs of different network levels after the screening process to obtain connectivity detection results of the target feature graphs of different network levels; According to the size of the connected areas in the connectivity detection result, the connected areas are sorted in descending order to obtain a sorting result; models of a preset number of first N connected areas are obtained from the sorting result; wherein N is a positive integer greater than or equal to 2.

[0056] According to a processing device for a synthetic casting model provided by the present invention, the filtering process is performed on the values ​​in the three-dimensional array corresponding to the feature graph after the filtering process to obtain target feature graphs of different network levels after the value filtering process, specifically comprising: According to the quantitative characteristics of all values ​​in the three-dimensional array corresponding to the feature map after the screening process, a value threshold is determined; based on the value threshold, the values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

[0057] According to a processing device for a synthetic casting model provided by the present invention, the casting model processing unit is specifically used for: Acquire size and position information of casting feature sub-models respectively corresponding to the at least two three-dimensional casting models; Based on the size and position information of the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models, the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are rotated, scaled and / or translated respectively, so that the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are spliced ​​and synthesized to obtain a new casting model.

[0058] The processing device for synthetic casting models provided by the present invention obtains at least two three-dimensional casting models to be processed, inputs the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, obtains feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models respectively, performs processing based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, and performs splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, to quickly obtain a new casting model, which can effectively improve the efficiency of synthesizing casting models, thereby reducing the workload of synthesizing casting models.

[0059] Fig.12 An example of a physical structure diagram of an electronic device is shown in FIG. Fig.12As shown, the electronic device (i.e., a power grid real-time monitoring system or a power grid real-time monitoring device) may include: a processor (processor) 1201, a communication interface (Communications Interface) 1204, a memory (memory) 1202 and a communication bus 1203, wherein the processor 1201, the communication interface 1204, and the memory 1202 communicate with each other via the communication bus 1203. The processor 1201 can call the logic instructions in the memory 1202 to execute a processing method for a synthetic casting model, which includes: obtaining at least two three-dimensional casting models to be processed; inputting the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtaining feature maps of different network levels output by the casting prediction model corresponding to the at least two three-dimensional casting models; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on a sample three-dimensional casting model and a feature prediction result label corresponding to the sample three-dimensional casting model; processing based on the feature maps of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models; performing splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models to obtain a new casting model.

[0060] In addition, the logic instructions in the above-mentioned memory 1202 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0061] On the other hand, the present application also provides a computer program product, which includes a computer program, which can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the processing method of the synthetic casting model provided by the above-mentioned methods, which includes: obtaining at least two three-dimensional casting models to be processed; inputting the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtaining feature maps of different network levels output by the casting prediction model corresponding to the at least two three-dimensional casting models; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on a sample three-dimensional casting model and a feature prediction result label corresponding to the sample three-dimensional casting model; processing is performed based on the feature maps of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models; splicing and synthesis processing is performed based on the casting feature sub-models corresponding to the at least two three-dimensional casting models to obtain a new casting model.

[0062] On the other hand, the present application also provides a computer-readable storage medium, the computer-readable storage medium includes a stored program, wherein the program, when running, executes the processing method of the synthetic casting model provided by the above-mentioned methods, the method comprising: obtaining at least two three-dimensional casting models to be processed; inputting the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtaining feature maps of different network levels output by the casting prediction model corresponding to the at least two three-dimensional casting models; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on a sample three-dimensional casting model and a feature prediction result label corresponding to the sample three-dimensional casting model; processing is performed based on the feature maps of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models; splicing and synthesis processing is performed based on the casting feature sub-models corresponding to the at least two three-dimensional casting models to obtain a new casting model.

[0063] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0064] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for processing a synthetic casting model, characterized in that: include: obtaining at least two three-dimensional casting models to be processed; The at least two three-dimensional casting models are respectively input into a preset casting prediction model for feature extraction and analysis, and feature graphs of different network levels corresponding to the at least two three-dimensional casting models are obtained, which are output by the casting prediction model; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on sample three-dimensional casting models and feature prediction result labels corresponding to the sample three-dimensional casting models in advance; Based on the feature graphs of different network levels, processing is performed to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively; based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively, splicing and synthesis processing is performed to obtain a new casting model.

2. The method for processing a synthetic casting model according to claim 1, characterized in that: The processing based on the feature graphs of different network levels to obtain the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively specifically includes: Screening the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels to obtain target feature graphs of different network levels after the value screening process; Connectivity detection is performed based on the target feature graphs of different network levels after the screening process to obtain models of a preset number of connected areas, and the models of the connected areas are used as the casting feature sub-models.

3. The method for processing a synthetic casting model according to claim 2, characterized in that: The filtering process of the values ​​in the three-dimensional array corresponding to the feature graphs of different network levels to obtain the target feature graphs of different network levels after the value filtering process specifically includes: Screening the feature graphs of different network layers to obtain the screened feature graphs; The values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

4. The method for processing a synthetic casting model according to claim 2, characterized in that: The target feature graphs at different network levels after the screening process are used to perform connectivity detection to obtain a model of a preset number of connected areas, specifically including: Performing connectivity detection based on the target feature graphs of different network levels after the screening process to obtain connectivity detection results of the target feature graphs of different network levels; According to the size of the connected areas in the connectivity detection result, the connected areas are sorted in descending order to obtain a sorting result; models of a preset number of first N connected areas are obtained from the sorting result; wherein N is a positive integer greater than or equal to 2.

5. The method for processing a synthetic casting model according to claim 3, characterized in that: The filtering process is performed on the values ​​in the three-dimensional array corresponding to the feature graph after the filtering process to obtain target feature graphs of different network levels after the value filtering process, specifically including: According to the quantitative characteristics of all values ​​in the three-dimensional array corresponding to the feature map after the screening process, a value threshold is determined; based on the value threshold, the values ​​in the three-dimensional array corresponding to the feature map after the screening process are screened to obtain target feature maps of different network levels after the value screening process.

6. The method for processing a synthetic casting model according to claim 1, characterized in that: The step of performing splicing and synthesis processing based on the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models to obtain a new casting model specifically includes: Acquire size and position information of casting feature sub-models respectively corresponding to the at least two three-dimensional casting models; Based on the size and position information of the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models, the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are rotated, scaled and / or translated respectively, so that the casting feature sub-models respectively corresponding to the at least two three-dimensional casting models are spliced ​​and synthesized to obtain a new casting model.

7. A processing device for a synthetic casting model, characterized in that: include: A casting model obtaining unit, used for obtaining at least two three-dimensional casting models to be processed; A casting model feature extraction unit, used for inputting the at least two three-dimensional casting models into a preset casting prediction model for feature extraction and analysis, and obtaining feature graphs of different network levels output by the casting prediction model and corresponding to the at least two three-dimensional casting models; wherein the casting prediction model is a convolutional neural network model obtained by iterative training based on sample three-dimensional casting models and feature prediction result labels corresponding to the sample three-dimensional casting models in advance; The casting model processing unit is used to perform processing based on the feature graphs of different network levels to obtain casting feature sub-models corresponding to the at least two three-dimensional casting models respectively; and perform splicing and synthesis processing based on the casting feature sub-models corresponding to the at least two three-dimensional casting models respectively to obtain a new casting model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the method for processing a synthetic casting model according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for processing a synthetic casting model according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for processing a synthetic casting model according to any one of claims 1 to 6 is implemented.

Images