CN117972813A

CN117972813A - Intelligent process method, system, equipment and medium for machining parts

Info

Publication number: CN117972813A
Application number: CN202410361359.5A
Authority: CN
Inventors: 李建勋; 邱慧慧; 魏亮; 梅敬成; 邵梦; 刘斌; 李龙传; 齐浩杰; 靳化振; 张金龙; 姜勇泉; 杨飞; 邵梅; 周超
Original assignee: Shandong Hoteam Software Co ltd
Current assignee: Shandong Hoteam Software Co ltd
Priority date: 2024-03-28
Filing date: 2024-03-28
Publication date: 2024-05-03
Anticipated expiration: 2044-03-28
Also published as: CN117972813B

Abstract

The invention belongs to the technical field of machined parts, and provides an intelligent process method, a system, equipment and a medium for machining parts, which are used for classifying the parts based on a preset classification model by combining the identification characteristics of the parts and a three-dimensional model of the parts, so as to solve the problem that the parts cannot be expressed in an omnibearing manner and automatically; comparing the identification characteristics of the parts and the three-dimensional model of the parts with the parts in the knowledge base, and judging the parts to be compared, namely similar parts and processes thereof; synchronizing the identified process characteristic information and the process information in the process knowledge base into a knowledge graph, and carrying out structural management and display; the whole process expression of the parts is realized, the structured data is managed, and the production efficiency is improved; the technology reasoning method based on deep learning realizes the mining of the existing technology data and the reuse of the technology data, and realizes technology data reasoning based on data driving.

Description

Intelligent process method, system, equipment and medium for machining parts

Technical Field

The invention belongs to the technical field related to machined parts, and particularly relates to an intelligent process method, system, equipment and medium for machined parts.

Background

The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.

Parts are the basic building blocks in the manufacturing process and are also the basic objects of machining. The process design is used as a bridge for connecting product design and product manufacture, and has important influence on the parts, especially on the aspects of ensuring the quality of the parts, improving the production efficiency, promoting the upgrading and improvement of the parts and the objects. The traditional part process design process requires engineers to analyze and classify the parts, then the process scheme of the parts is formulated according to the parts to be processed, and the process scheme is input into a CAPP or PDM system and the like.

The patent CN201911026801.4 discloses an intelligent manufacturing process system of a gearbox body, which realizes feature identification, process reasoning and production scheduling, but the method is mainly based on a rule method and does not contain a part classification module, a knowledge graph module and the like. The prior proposal is as follows: the method comprises three steps of modeling and management based on machining process knowledge, skeleton process generation based on knowledge matching, and accurate process generation based on process constraint. The machining process design system based on the three-dimensional model is researched and focused on feature recognition and knowledge reuse, a system framework is established, and a machining feature recognition algorithm and a process knowledge base are developed. The intelligent decision and the process reuse technology are discussed, the system provides key contents such as feature process chain decision, part potential alternative process route generation, process route comprehensive decision and the like, and the intelligent decision of the system is realized through methods such as fuzzy evaluation, complex network modeling, similarity measurement and the like. Three-dimensional part information often requires a great deal of human intervention in the process of transferring from the design process to the process, and lacks corresponding support in terms of existing reuse methods of process design data. The prior art, however, mainly shows the following points: 1) The identification and classification are needed to be carried out manually; 2) The characteristics of the method need to be identified manually so as to support the follow-up process design; 3) Artificial or semi-artificial process design; 4) The existing mining and reusing method of the process data is missing; 5) The existing process data is structured and has poor visual effect. And the existing researches are mainly aimed at a certain aspect, and a complete set of intelligent process management and design methods for machined parts are lacked.

In summary, in the machining field of the discrete manufacturing industry, how to unify the three-dimensional design of the parts to the process design of the parts, realize the omnibearing expression of the parts, reuse and excavation of the existing process information, and further realize the automation and visualization of the process design process is a problem to be solved urgently at present.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides an intelligent process method, system, equipment and medium for machining parts, which realize feature identification of the parts, automatic classification of the parts, automatic matching of the parts process, management of structured data by constructing a knowledge graph and automatic reasoning of the parts process, and improve the production efficiency.

To achieve the above object, a first aspect of the present invention provides an intelligent process for machining parts, comprising:

acquiring identification characteristics of the part and a corresponding three-dimensional model of the part;

Inputting the identification characteristics of the parts and the corresponding three-dimensional models of the parts into a preset part classification model to obtain the classification result of the parts; the preset part classification model specifically comprises the following steps: respectively extracting semantic relation of part identification features and key information of a part three-dimensional model; respectively processing the semantic relation of the extracted part identification features and the key information of the part three-dimensional model, then carrying out information fusion, and carrying out classification identification based on the fused information;

determining a part to be compared based on a classification result of the part, and performing part process matching by calculating feature similarity and three-dimensional model similarity between the part and the part to be compared based on identification features of the part and a corresponding three-dimensional model of the part;

Constructing a knowledge graph according to the obtained classification and identification result of the parts and the matching result of the part process;

And carrying out process data reasoning on the parts based on a process reasoning model according to the constructed knowledge graph.

A second aspect of the invention provides a smart process system for machining parts, comprising:

The acquisition module is used for: acquiring identification characteristics of the part and a corresponding three-dimensional model of the part;

and a classification module: inputting the identification characteristics of the parts and the corresponding three-dimensional models of the parts into a preset part classification model to obtain the classification result of the parts; the preset part classification model specifically comprises the following steps: respectively extracting semantic relation of part identification features and key information of a part three-dimensional model; respectively processing the semantic relation of the extracted part identification features and the key information of the part three-dimensional model, then carrying out information fusion, and carrying out classification identification based on the fused information;

and a process matching module: determining a part to be compared based on a classification result of the part, and performing part process matching by calculating feature similarity and three-dimensional model similarity between the part and the part to be compared based on identification features of the part and a corresponding three-dimensional model of the part;

And (3) constructing a knowledge graph module: constructing a knowledge graph according to the obtained classification and identification result of the parts and the matching result of the part process;

the process reasoning module: and carrying out process data reasoning on the parts based on a process reasoning model according to the constructed knowledge graph.

A third aspect of the present invention provides a computer apparatus comprising: the system comprises a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, when the computer device runs, the processor and the memory are communicated through the bus, and the machine-readable instructions are executed by the processor to execute an intelligent process method facing to a machined part.

A fourth aspect of the invention provides a computer readable storage medium having a computer program stored thereon, which when executed by a processor performs a smart process for machining parts.

The one or more of the above technical solutions have the following beneficial effects:

In the invention, the identification characteristics of the parts and the three-dimensional model of the parts are combined, and the parts are classified based on a preset part classification model; comparing the identification characteristics of the parts and the three-dimensional model of the parts with the parts in the knowledge base, and judging the parts to be compared, namely similar parts and processes thereof; synchronizing the identified process characteristic information and the process information in the process knowledge base into a knowledge graph, and carrying out structural management and display; the whole process expression of the parts is realized, the knowledge graph is constructed to manage the structured data, the production efficiency is improved, and the process reasoning is carried out on the parts according to the constructed knowledge graph, so that the omnibearing expression of the parts is realized.

Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.

FIG. 1 is a flow chart of an intelligent process method for machining parts according to a first embodiment of the invention;

FIG. 2 is a diagram showing a preferred method of identifying processing features based on start surface classification in accordance with a first embodiment of the present invention;

FIG. 3 is a model of a part class in accordance with a first embodiment of the present invention;

FIG. 4 is a multi-modal deep learning classification model according to an embodiment of the invention;

FIG. 5 is a three-dimensional model contrast algorithm model according to an embodiment of the invention;

FIG. 6 is a schematic diagram of an exemplary process reasoning model in accordance with an embodiment of the present invention;

FIG. 7 is a schematic diagram of a sequential process reasoning model in accordance with an embodiment of the present invention;

FIG. 8 is a schematic diagram of an intelligent process system for machining parts according to a first embodiment of the present invention;

FIG. 9 is a schematic view of a pin in accordance with a first embodiment of the present invention;

FIG. 10 is a schematic view of a pin similar to that of the first embodiment of the present invention;

Fig. 11 is a diagram illustrating a knowledge graph of a pin in accordance with an embodiment of the present invention.

Detailed Description

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention.

Embodiments of the invention and features of the embodiments may be combined with each other without conflict.

Example 1

The embodiment discloses an intelligent process method for machining parts, which comprises the following steps:

Constructing a part type model, and inputting the identification characteristics of the part and the corresponding part three-dimensional model into a preset part classification model to obtain a part classification result; the preset part classification model specifically comprises the following steps: respectively extracting semantic relation of part identification features and key information of a part three-dimensional model; respectively processing the semantic relation of the extracted part identification features and the key information of the part three-dimensional model, then carrying out information fusion, and carrying out classification identification based on the fused information;

Carrying out process data reasoning on the part based on a process reasoning model according to the constructed knowledge graph;

And constructing two deep learning methods of visual operation by using the acquired process data, and realizing typical process reasoning and sequential process reasoning.

The technical scheme of the embodiment is as follows: an intelligent process method for machining parts is constructed: feature recognition of the three-dimensional part model is realized; the identified part characteristics and the part three-dimensional model are combined, and the part is correspondingly classified into a part category model; comparing the identified part characteristics and the three-dimensional model of the part with the parts in the knowledge base, and judging similar parts and processes thereof; synchronizing the identified process characteristic information and the process information in the process knowledge base to a knowledge graph for structural management and display; and inputting the process data in the knowledge base and the atlas into the deep learning model to realize the reasoning of the process procedure information.

An intelligent process method for machining parts according to this embodiment is described in detail with reference to fig. 1 and 8, and specifically includes:

Step 1: part feature identification

As shown in fig. 2, the method specifically includes:

(1) And realizing the surface-edge topological relation analysis of the part Brep based on SVMAN-M to obtain the surface-edge set of the part S.

(2) For the public edges of any two sides, based on the direction vector of the public edge and the normal information of the local areas of the adjacent two sides near the public edge, the included angle formed by the two sides at the public edge and the concave-convex information of the edge are calculated in sequence.

(3) Deleting all edges of the convex attributes from the part attribute adjacency graph AAG to obtain a concave sub-graph set MAAG = { MAAG }, i=1, 2, …, m.

(4) One sub-graph MAAG is taken from MAAG; taking the surface with the largest adjacent surface in the subgraph as a starting surface initial value CGF;

(5) Feature start surface optimization algorithms such as grooves, holes (lines), outer circles, planes and the like are respectively applied to the CGF ₁, the start surfaces are optimized, and the feature types of the CGF are determined.

(6) The feature type of the CGF is determined, if so, to (7), and if not, to (8).

(7) The CGF, corresponding feature type, is recorded and all start surface sets MAAGA of the determined feature types are added to (9).

(8) Subgraph MAAG is performed; and adding the rest subgraph set MAAG _B to (9).

(9) It is determined whether the set MAAG is all traversed, if so, then go to (10), and if not, then go to (4).

(10) Based on SVMAN-M single feature recognition method, sequentially searching single features such as grooves, holes, excircles, planes and the like for the starting surfaces recorded in MAAG _A, extracting feature surfaces corresponding to the starting surfaces, creating corresponding feature nodes, and adding the recognized feature surface set FeatureSrf.

(11) For the concave subgraphs in the rest subgraph set MAAG _b, feature nodes of unidentified features on the feature tree can be temporarily listed, and then feature types can be determined by isomorphic comparison of the concave subgraphs in MAAG _b and feature subgraphs predefined by a user, and nodes of corresponding feature types can be created.

(12) And outputting the characteristic node tree to a knowledge base based on the SVMAN-M process management class interface.

Step2: part classification

1) And constructing a part category model. As shown in fig. 3, first, the parts are classified into a first class, and the parts are classified into a second class according to the revolving body and the non-revolving body. The class of revolution bodies can be sub-classified into three classes, including: wheel discs, bushings, pin shafts, gears, special-shaped pieces and special-purpose pieces; the revolution solid can be sub-classified into three classes, including: bars, plates, brackets, and box shells. Each sub-category can be further subdivided into four categories according to actual requirements.

2) A multi-modal deep learning classification model is used as the part classification model. Training the deep learning model by using the marked data set, inputting feature information and the part model obtained by feature recognition into the deep learning model at the same time, and outputting the specific classification condition of the part in the constructed part type model.

As shown in fig. 4, a multiple-input deep learning model is employed, while processing information acquired by the three-dimensional model of the part and feature recognition, and using them for a classification task. The deep learning model comprises five parts, namely a part model input layer, a characteristic information input layer, a part model processing layer, a characteristic information processing layer, a characteristic fusion layer and an output layer.

1. And inputting a three-dimensional model of the part into the layer. The 3DCNN is used as an input layer of the three-dimensional model of the part to process the data of the three-dimensional model of the part, and the network comprises operations such as convolution, pooling and flattening, so that extraction of key information in the model is realized. The part three-dimensional model input layer is connected to the part three-dimensional model processing layer.

2. And a characteristic information input layer. The identification features of the part in step 1 are processed by the feature information input layer. Specifically, embedding, i.e., the embedding layer, is used to map words to a high-dimensional vector space. The Embedding layer converts discrete words or categories into a dense low-dimensional vector representation. The input integer sequence is mapped to the embedding matrix, and the corresponding word embedding vector is output. The word embedding vector representation may capture semantic relationships between feature information. The feature information input layer is connected to the feature information processing layer.

3. And processing a layer of the three-dimensional model of the part. The input data is mapped to the high-dimensional information representation space through a full connection layer and then output to the information fusion layer through a similar full connection layer. The first full-connection layer maps the three-dimensional model information to a high-dimensional information representation space and is used for capturing the three-dimensional model information, the structure of the full-connection layer comprises a weight matrix and a bias vector, the structure is a traditional full-connection layer, and each input node is connected with each output node. The second fully-connected layer further processes this high-dimensional information representation, mapping the three-dimensional model information to a fixed dimension, the fully-connected layer also comprising a weight matrix and a bias vector, again being a conventional fully-connected layer, differing from the fully-connected layer described above in that the output dimension of the fully-connected layer is fixed.

4. And a characteristic information processing layer. The characteristic information maps the input data to a high-dimensional representation space through a full connection layer, and then outputs the input data to the information fusion layer through a similar full connection layer. The first full-connection layer maps the characteristic information to a high-dimensional information representation space and is used for capturing the characteristic information, the structure of the full-connection layer comprises a weight matrix and a bias vector, the structure is a traditional full-connection layer, and each input node is connected with each output node. The second fully-connected layer further processes this high-dimensional representation of information, mapping the feature information to a fixed dimension, the fully-connected layer also comprising a weight matrix and a bias vector, again being a conventional fully-connected layer, differing from the fully-connected layer described above in that the output dimension of the fully-connected layer is fixed.

5. And an information fusion layer. The output of the part three-dimensional model processing layer and the characteristic information processing layer is fused through an information fusion layer. The information fusion layer adopts a full connection layer, the part model processing layer and the characteristic information processing layer are output and spliced, and then mapped to a space suitable for task dimension through the full connection layer, and a subsequent characteristic representation is obtained after an activation function is added.

6. And an output layer. The fused information representation is mapped to the output layer by a fully connected layer. The output layer uses a softmax activation function, is applicable to multi-classification problems, and outputs probability distribution of each class so that the model can classify the input.

Training the deep learning model by using a marked data set, wherein the data set needs to cover all classification corresponding parts in the part class model, and cross entropy loss is used as a loss function of the deep learning model. After training is completed, feature information and a three-dimensional model of the part, which are obtained by feature recognition, are simultaneously input into the deep learning model, and the classification condition of the part, which is specifically located in the constructed part type model, can be output.

Step 3: parts process matching

A similarity calculation method based on combination of feature similarity and model similarity is used. According to the method, the comprehensive similarity CS of the process part to be matched and other parts in the selected range is calculated to obtain the similarity between different parts.

The comparison selection range of the part process is divided into primary classification, secondary classification, tertiary classification and quaternary classification according to the classification condition of the part, and corresponds to the part class model position of the part. If the first-level classification is used, the parts are compared with other parts in the first-level classification, and the second-level classification, the third-level classification and the fourth-level classification are the same.

The comprehensive similarity CS is obtained by adding the feature similarity FS and the three-dimensional model similarity MS after a certain weight is respectively given.

The calculation formula of the comprehensive similarity CS is as follows:

Wherein FS is feature similarity, MS is model similarity, a is a weight coefficient of feature similarity, and b is a model similarity weight coefficient. a. b is adjusted according to the actual application condition of the algorithm.

The feature similarity FS is obtained by a feature information comparison algorithm. The feature information comparison algorithm calculates the similarity FS of the two parts by comparing the attribute information of the two parts with the features and the feature attribute information obtained by feature recognition.

Specifically, the feature information comparison algorithm is applied based on the BM25 algorithm, and can be divided into the following 3 processes.

1. And establishing an index. Selecting the selected comparison within the selection range asN is the total number of parts to be compared, and D is a certain part to be compared containing information such as characteristic attribute and the like.

2. The BM25 score was calculated. The BM25 algorithm is used to calculate the relevance scores of the query part and the part to be compared. The calculation formula is as follows:

Where f (Q _i, D) is the number of occurrences of query term Q _i in part D, where Q _i is a specific attribute of query part Q, such as information including characteristics, materials, quality, etc., inherited by the characteristics recognition, part classification, and part design process; n (q _i) is the number of parts that contain the query term q _i; len (D) is the number of all properties that part D contains; avg_len is the average number of attributes of the part; k ₁ and b are adjustment parameters.

3. Obtaining scores of a query part and a plurality of parts to be compared。

The model similarity MS is obtained by a three-dimensional model comparison algorithm. The three-dimensional model comparison algorithm calculates the similarity MS by comparing the similarity of the three-dimensional models of the two parts.

As shown in fig. 5, a 3 DCNN-based connected deep learning network was used. The method can be concretely divided into the following parts: an input layer, a shared 3DCNN feature extraction layer, a similarity distance calculation layer and a similarity output layer.

1. An input layer. There are two input layers corresponding to the three-dimensional models of the query part and the part to be compared, respectively. The two input layers define the input shape of each part three-dimensional model, ensuring that the input part three-dimensional model matches the shape desired by the network. The dimensions of the input data are the length, width and height in three axes.

2. Shared 3DCNN feature extraction layer. There are two 3DCNN networks for feature extraction of two parts to be compared, which share weights. And respectively inputting the query part and the part to be compared which are output by the input layer into two 3DCNN networks. The 3DCNN is used to extract features from a three-dimensional model of a part.

Specifically, features are firstly extracted from input data through a 3D convolution layer, convolution operation is carried out on the input data in three dimensions through a sliding convolution kernel, a series of feature graphs are generated, each feature graph represents different spatial features, and values in the feature graphs represent response degrees of the features at corresponding positions; then reducing the space dimension of the feature map by using a pooling layer, retaining the most important feature information, reducing the parameter quantity of a model by using average pooling, and having a certain invariance to translation and scaling of input data; the nonlinear characteristic is introduced through the ReLU activation function, so that the network can learn the nonlinear relation; and finally, using a full connection layer for mapping the extracted features to a final output space, connecting all the features together by the full connection layer, and performing linear transformation through a weight matrix to generate the output of the 3DCNN network.

3. And a similarity distance calculating layer. And inputting the two 3DCNN network output results, namely the two feature representations, into a similarity distance calculation layer, and measuring the similarity between the two feature representations through Euclidean distance calculation.

4. And an output layer. Outputting Euclidean distance calculation results by using identity mapping to obtain model similarity scores。

Step 4: construction of technological knowledge graph

And constructing a knowledge graph model layer. And converting the existing part model, process model and preset attributes corresponding to the model in the knowledge base into a form of triples, and synchronizing the triples to the knowledge graph. And supplementing the relation among the models, and taking the relation as a model layer of the knowledge graph.

The part examples and the process examples in the knowledge base are converted into knowledge graph data layer data according to the organization mode of the parts and the processes in the mode layer.

Step 5: art data reasoning

(1) And (5) data acquisition. And acquiring the parts and the related attributes thereof and the corresponding processes of the parts from the knowledge base and the knowledge map respectively, and taking the parts and the related attributes as the training data and the reasoning data of the reasoning model.

(2) And configuring a process reasoning model. The intelligent model on-line construction function of the process reasoning is provided, a user can configure and train the intelligent model facing to the personalized process data of the user enterprise by using an advanced artificial intelligent algorithm without writing codes, and interactive selection and display of information such as super parameters, training sets, training results, cost diagrams and the like are provided.

(3) A typical process reasoning model. Training learning is performed on the model using the training data. The model takes the whole process of the part as a label, and matches the process applicable to the part in a knowledge base or a knowledge graph for the new part by analyzing the information contained in the part and the related attribute thereof.

As shown in fig. 6, a typical process reasoning model can be specifically divided into: an input layer, a feature extraction layer, an attention layer and an output layer.

1. An input layer. The part related attribute information is encoded into a vector form by a word embedding mode and is used as the input of a network.

2. And a feature extraction layer. This layer may use RNN, GRU, LSTM, biLSTM network forms in the process inference model configuration process. Sequence data is processed and timing information is captured. The extraction of each attribute of the part into this layer is expressed as a hidden state of forward and backward.

3. Attention layer. This layer may be used in the process inference model configuration process selection. The degree of interest of the model in different parts of the sequence of part attributes is enhanced. The weight of each input position is obtained through learning, and the attention degree of the position is represented.

4. And an output layer. The weighted context vector is received as an input and a final output is generated by a full join operation. A prediction of the final process label is generated.

(4) And (5) a sequential process reasoning model. Training learning is performed on the model using the training data. The model splits the process of the part into process sequences according to the working procedures or the working steps, and generates new process sequences according to the working procedures or the working steps by analyzing the information contained in the part and the related attributes thereof.

As shown in fig. 7, the sequential process reasoning model comprises three parts: the encoder is the feature extraction layer, the attention mechanism, the decoder is the output layer.

1. An encoder. This layer may use RNN, GRU, LSTM, biLSTM network forms in the process inference model configuration process. The method specifically aims at mapping the input part attribute sequence into a context vector and capturing important information of the input sequence. Taking the hidden state of the last time step results in a context vector that will contain the key information of the input sequence.

2. Attention mechanisms. This layer may be used in the process inference model configuration process selection. The attention weight is calculated using the hidden state of the current time step of the decoder and the hidden states of all time steps of the encoder. Different parts of the input sequence are dynamically focused in the decoding process to improve the utilization of the input information by the model.

3. And a decoder. This layer may use RNN, GRU, LSTM network forms in the process inference model configuration process. A part process sequence is generated based on the context vector of the encoder and the output of the previous time step. The weighted context vector and decoder output of the previous time step are received, yielding an output of the current time step and a hidden state. Finally, the output is mapped to the target tag space using the full connectivity layer and softmax activation function.

The process of this embodiment is performed using the pin part shown in fig. 9, and the characteristics of the pin obtained by the characteristic recognition module are shown in table 1:

Table 1:

Sequence number	Feature type
		1	Plane 1 in general
2	Plane 2 in general
		3	Cylinder column
4	Chamfer 1
		5	Chamfer 2
6	Round hole 1
		7	Round hole 2

The pin shafts are classified by the part classification module, and the classification result is pin shaft class.

The parts with the highest comprehensive similarity are obtained through a part process matching model as shown in fig. 10, and the part process is used as a matching process, and the matching process is shown in table 2.

Table 2:

Sequence number	Matching process
		1	Discharging
2	Coarse and fine turning
		3	Scribing line
4	Drilling machine
		5	Clamp
6	Inspection and detection
		7	Quenching
8	Finish turning
		9	Fine grinding
10	Inspection and detection
		11	Zinc plating passivation

Taking the pin shaft and part of the data of the characteristics contained in the pin shaft as an example, the constructed knowledge graph is shown in fig. 11.

Two sets of processes are respectively obtained by a typical process reasoning model and a sequential process reasoning model through a process data reasoning module, and the two model obtaining processes are shown in table 3.

Table 3:

Sequence number	Typical process reasoning model	Sequence process reasoning model
			1	Discharging	Discharging
2	Heat of the body	Tempering
			3	Coarse and fine turning	School and school
4	Inspection and detection	Coarse and fine turning
			5	Scribing line	Inspection and detection
6	Drilling machine	Scribing line
			7	Clamp	Drilling machine
8	Inspection and detection	Clamp
			9	Zinc plating passivation	Inspection and detection
10		Zinc plating passivation

Example two

It is an object of this embodiment to provide an intelligent process system for machining parts comprising:

The system constructed by the embodiment comprises parts characteristic identification, parts classification, parts process matching, process knowledge graph, process data reasoning and knowledge base.

Part feature identification: a method for realizing automatic identification of part processing features based on a visual interaction interface and a functional algorithm provided by a domestic autonomous controllable three-dimensional machining process planning design software system SVMAN-M and a product full life cycle management system INFORCENTER. After a three-dimensional model of a part is input by a user, the three-dimensional model feature is identified by using a processing feature identification method based on the starting surface classification optimization, and the identification result is stored through SVMAN-M and INFORCENTER unified knowledge base.

Part classification: and a visual interaction interface and a functional algorithm are constructed based on the product full life cycle management system INFORCENTER, so that a part classification function is realized. After the part category model is built in the INFORCENTER knowledge base, the part classification module acquires the three-dimensional model of the part and the characteristic information identified by the part characteristic identification module, and the multi-mode deep learning classification model is used for classifying the part. The acquired part classification information is stored through INFORCENTER unified knowledge base.

Matching the part technology: and constructing a visual interaction interface and a functional algorithm based on the product full life cycle management system INFORCENTER to realize part process matching. After the feature identification information, the part classification information and the three-dimensional model information are acquired, the comprehensive similarity is calculated by using a feature information comparison algorithm and a three-dimensional model comparison algorithm, and the corresponding process of the part is matched. The obtained matching process data is stored through a unified knowledge base.

Technological knowledge graph: and constructing a process knowledge graph visualization interaction interface and a functional algorithm based on the product full life cycle management system INFORCENTER and the knowledge graph. And constructing a bidirectional mapping mode of the unified knowledge base and the knowledge graph by using a rule-based method, and realizing synchronous updating from the knowledge base to the graph database.

And (3) process data reasoning: and constructing a process data reasoning visual interaction interface and a functional algorithm based on the product full life cycle management system INFORCENTER and the knowledge graph. And matching a knowledge base or a process suitable for the part in a knowledge graph for the new part through a typical process reasoning model. And generating a new process sequence according to the working procedures or the working steps through a sequence process reasoning model. And storing the process data obtained by reasoning through a unified knowledge base and synchronizing with the process knowledge graph.

Example III

It is an object of the present embodiment to provide a computing device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which processor implements the steps of the method described above when executing the program.

Example IV

An object of the present embodiment is to provide a computer-readable storage medium.

A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the above method.

The steps involved in the devices of the second, third and fourth embodiments correspond to those of the first embodiment of the method, and the detailed description of the embodiments can be found in the related description section of the first embodiment. The term "computer-readable storage medium" should be taken to include a single medium or multiple media including one or more sets of instructions; it should also be understood to include any medium capable of storing, encoding or carrying a set of instructions for execution by a processor and that cause the processor to perform any one of the methods of the present invention.

It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented by general-purpose computer means, alternatively they may be implemented by program code executable by computing means, whereby they may be stored in storage means for execution by computing means, or they may be made into individual integrated circuit modules separately, or a plurality of modules or steps in them may be made into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.

While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.

Claims

1. An intelligent process method for machining parts is characterized by comprising the following steps:

2. The intelligent process for machining parts according to claim 1, wherein the identification features of the parts are obtained by:

Analyzing the surface-edge topological relation of the part;

For the common edge of any two sides of the part, calculating an included angle formed by the two sides at the common edge and the concave-convex property of the edge based on the direction vector of the common edge and the normal vector at the common edge;

Deleting all edges of the convex attributes from the part attribute adjacency graph to obtain a concave sub-graph set;

selecting the most adjacent surfaces as starting surfaces for each concave sub-graph in the concave sub-graph set, carrying out a characteristic starting surface optimization algorithm on the starting surfaces, and determining the characteristic type of each starting surface;

and sequentially carrying out single feature retrieval on the starting surfaces with the determined feature types, and extracting the feature surfaces of all the starting surfaces.

3. The intelligent process for machining parts according to claim 1, wherein the preset part classification model specifically comprises:

extracting key information of a three-dimensional model of the part by using a three-dimensional convolutional neural network, and processing the extracted key information of the three-dimensional model of the part through sequentially connected full-connection layers;

Capturing semantic relations among the part identification features by utilizing the embedded layer, and processing the captured semantic relation features among the part identification by sequentially connected full-connection layers;

and splicing and fusing the semantic relation of the processed part identification features and key information of the part three-dimensional model, and using the spliced and fused result for part classification and identification.

4. The intelligent process method for machining parts according to claim 1, wherein the feature similarity between the parts and the parts to be compared is calculated by adopting a BM25 algorithm, specifically:

determining the number of parts to be compared;

calculating a correlation score of the part and the part to be compared by using a BM25 algorithm;

And adding the correlation scores of the parts and each part to be compared to obtain the feature similarity.

5. The intelligent process method for machining parts according to claim 1, wherein the three-dimensional model similarity between the parts and the parts to be compared is calculated by using a preset connected deep learning network, specifically:

Extracting features of the three-dimensional model of the part and the three-dimensional model of the part to be compared based on the parallel feature extraction layers;

calculating the similarity between the features through Euclidean distance for the output of the parallel feature extraction layers;

And normalizing the calculated similarity to obtain the similarity of the three-dimensional model.

6. The intelligent process method for machining parts according to claim 1, wherein the process data reasoning is performed based on a typical process reasoning model based on the constructed knowledge graph, and specifically comprises the following steps:

acquiring relevant attributes of the parts according to the knowledge graph;

encoding the relevant attributes of the parts into a vector form through an input layer in a word embedding mode;

extracting time sequence information of the vector after the input layer coding by utilizing a feature extraction layer to obtain a context vector;

And performing full connection operation on the obtained context vector to obtain a process label corresponding to the part.

7. The intelligent process method for machining parts according to claim 6, wherein the process sequence reasoning is performed on the parts through a sequence process reasoning model, specifically:

Mapping the part related attribute into a context vector by using an encoder;

a part process sequence is generated using the decoder based on the context vector obtained by the encoder and the output of the previous time step.

8. An intelligent process system for machining parts, comprising:

And a classification module: inputting the identification characteristics of the parts and the corresponding three-dimensional models of the parts into a preset classification model to obtain classification results of the parts; the preset classification model specifically comprises the following steps: respectively extracting semantic relation of part identification features and key information of a part three-dimensional model; respectively processing the semantic relation of the extracted part identification features and the key information of the part three-dimensional model, then carrying out information fusion, and carrying out classification identification based on the fused information;

9. A computer device, comprising: a processor, a memory and a bus, said memory storing machine readable instructions executable by said processor, said processor and said memory in communication via the bus when the computer device is running, said machine readable instructions when executed by said processor performing a smart process for machining parts according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, performs a smart process for machining parts according to any of claims 1 to 7.