Template-based Inference Method and System for Aeronautical Part Process Planning

Abstract

The present invention relates to the technical field of intelligent manufacturing of aviation parts, and discloses a template-based inference method and system for aviation part process specifications. First, the process features and geometric structure features of the parts to be machined are obtained and stored. Then, the feature data of the parts to be machined, the feature data of the historically machined parts, the process specification template data, and the process knowledge data that need to be used are encoded in the One-hot manner. Next, according to the process features and geometric structure features of the parts to be machined, through a feature-based template matching algorithm, a matching process specification template is obtained, and through a typical process parameter inference and decision-making algorithm, the typical process parameters involved in the preparation of the part process specification based on the process specification template are obtained. The matching process specification template is fused with the calculated typical process parameters to form a part process specification and output. The present invention can quickly and intelligently generate the required part process specifications, greatly improving the design efficiency of the process design link in the production process.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing of aviation parts, and more specifically to a method and system for reasoning about aviation part process specifications based on templates. Background Art

[0002] In the traditional process design process, the preparation of process specifications relies too much on the personal experience and level of process designers. The rationality, operability of process documents, and the length of the preparation time mainly depend on the experience and proficiency of process personnel; at the same time, the preparation cycle is long, there is a lot of repetitive labor, and it is difficult to adapt to the changes of products and manufacturing environments.

[0003] However, the internal structure of an aircraft is complex, the shapes of parts are diverse and the quantity is large, and its development process is very complicated. Most products are characterized by multiple varieties and small batches. Due to limited personal experience and different design levels, the process documents designed vary from person to person, with problems such as poor consistency, difficult optimization, and weak standardization, and it is also difficult to inherit process design experience. Along with the transformation and upgrading of the aerospace manufacturing industry, new requirements have been put forward for the preparation of process specifications for complex products such as modern aircraft products. Therefore, how to push relevant knowledge matching the process specification preparation to achieve the rapid and intelligent preparation of part process specifications is of great significance. Summary of the Invention

[0004] The present invention proposes a method and system for reasoning about aviation part process specifications based on templates. The present invention quickly and intelligently generates the required part process specifications through the part features of the part to be machined and the part features and processing data of existing historical machined parts, greatly improving the design efficiency of the process design link in the production process, improving the efficiency and intelligence of the production process, and further improving the quality and reliability of part process design.

[0005] To achieve the above-mentioned invention object, the technical solution of the present invention is as follows:

[0006] On the one hand, the present invention discloses a method for reasoning about aviation part process specifications based on templates, including the following steps:

[0007] Step S1. Extract the three-dimensional digital model information and features of the part to be machined, extract the process features and geometric structure features of the part to be machined and store them;

[0008] Step S2. Encode the feature data of the part to be machined, the feature data of historical machined parts, the process specification template data, and the process knowledge data to be used in the One-hot manner;

[0009] Step S3. According to the process characteristics and geometric structure characteristics of the part to be processed, obtain the matching process specification template through a feature-based template matching algorithm; wherein, the feature-based template matching algorithm includes:

[0010] Use the Pearson correlation coefficient and mutual information to measure the correlation between part features and template identification values respectively, and then combine the two measurement results by weighting, and select the feature attributes with the highest correlation to establish a set;

[0011] Define categories based on the template identification value data set, calculate the expected information required for sample classification, and divide the data set through feature attributes, and calculate the entropy or expected information of the subset;

[0012] Convert the initial decision tree into a rule set of condition-action pairs, and perform attribute pruning on the rules;

[0013] Use the pruned rule set to classify and predict the data samples of the part to be processed, and output the matching process specification template;

[0014] Step S4. According to the process characteristics and geometric structure characteristics of the part to be processed, obtain the typical process parameters involved in the preparation of the part process specification based on the process specification template through a typical process parameter reasoning and decision-making algorithm;

[0015] Step S5. Integrate the matching process specification template with the typical process parameters obtained by the typical process parameter reasoning and decision-making algorithm to form a part process specification;

[0016] Step S6. Edit the part process specification to finally obtain a part process specification that can be put into production use.

[0017] Preferably, in step S2, the feature attributes of the part feature library and the template identification values used to uniquely identify the template in the process specification template library are encoded in a One-hot manner, and each element is associated with a unique integer index and converted into a binary vector of the size of the vocabulary length, so as to obtain the corresponding feature attribute data set and template identification value data set , where , X i represents the i th part feature in the part instance; x ij represents the i th class feature attribute corresponding to the j th class of part instances, y j represents the number of part instances corresponding to the j th class of template identification value data set.

[0018] Preferably, in step S3, a matching process specification template is obtained through a feature-based template matching algorithm, including:

[0019] Step S31. The Pearson correlation coefficient is used to measure the correlation between the feature attributes of the part feature library and the process specification template library, and the calculation expression is as follows:

[0020] Equation (1);

[0021] Wherein, represents the Pearson correlation coefficient calculation expression; and respectively represent the i th part feature and the template identification value represents the i th covariance between the part feature and the template identification value; represents the variance of the variable; respectively represent the i th average value of the part feature and the template identification value; represents the expected value of the variable; represents the covariance obtained between the i th part feature and the template identification value; represents the variance obtained between the i th part feature and the template identification value;

[0022] Step S32. The mutual information is combined to measure the correlation between the feature attributes of the part feature library and the process specification template library, and the calculation expression is as follows:

[0023] Equation (2);

[0024] Wherein, represents the mutual information correlation calculation expression; represents the i th information entropy of the part feature; represents the template identification value Y under the condition of the i th information entropy of the part feature; represents the entropy value calculation; represents the calculation of the conditional entropy;

[0025] Step S33. The Pearson correlation analysis result and the mutual information correlation analysis result are combined using a weighted method, and the calculation expression is as follows:

[0026] Equation (3);

[0027] Wherein, For measuring the i th part feature Under the given information, for the template identification value The amount of information provided; Indicates normalizing the calculated value of mutual information; α And β Are weights, ;

[0028] Step S34. Take the K c feature attributes with the highest correlation in step S33 to establish a set ;

[0029] Among them, Indicates the corresponding relevant t part features, and there exists ;

[0030] Step S35. Assume that the template identification value data set Has m different values, define m different classes ; Assume Is the number of samples in class C l Calculate the expected information required for classifying a given sample using formula (4);

[0031] (4);

[0032] Among them Is the probability that any sample belongs to C l And estimate it using , Indicates the identification value corresponding to the actual sample; Indicates calculating the expected information required for classifying a given sample, Indicates a discrete variable;

[0033] Step S36. Let the feature attribute X i Have v different values , Use the feature attribute X i To divide D c Into v subsets ; Let d lh Be the number of samples in class D h In class C lThe number of samples, according to X i shown by the calculation expression of the entropy or expected information for dividing subsets:

[0034] Equation (5);

[0035] Among them, acts as the weight of the h th subset, and the h th subset represents the divided subset ; represents the weighted value of the characteristic attributes set by the part design principle; represents the expected information for dividing subsets according to ; represents the normalization factor;

[0036] Step S37. Convert the initial decision tree into a set of rules, where each rule is represented as a condition-action pair, where the condition is a series of feature tests and the action is the corresponding class label; the generated rule set is , r i represents the i th rule, represents the condition part of rule r i , represents the action part of rule r i ;

[0037] Step S38. Perform attribute pruning on the generated rules, and the pruned rule set is , where is the pruned rule, is the condition part of rule , is the action part of rule ;

[0038] Step S39. For the new sample of the part to be processed, use the pruned rule set to perform classification prediction, and finally output the process planning template corresponding to the part to be processed.

[0039] Preferably, in step S4, the typical process parameter inference decision algorithm includes a process flow decision inference algorithm and a typical process parameter decision inference algorithm.

[0040] Preferably, the process flow decision inference algorithm is specifically as follows:

[0041] Step a. Index code each process flow method, and then obtain the vector of each process flow through an embedding layer with an embedding size of r, which is m dimensional; represent all process attributes using character-level one-hot vectors, and represent each character with a n dimensional one-hot vector to form a matrix of ( N ) × n dimensional;

[0042] Step b. Adopt a method of complex vector feature fusion by superimposing eigenvectors. After encoding the pre-order process flow sequence through a two-layer DNN-LSTM layer, perform vector superposition on the process flow features at each time step and the current input process attribute features;

[0043] Step c. Use a two-layer DNN-LSTM to maintain the sequential relationship between processes under process attributes, and generate subsequent process flows by decoding the vector obtained by fusing process attributes and process flow sequences.

[0044] Preferably, step c specifically includes: first, pass the vector after fusing process attributes and process flow sequences through a two-layer DNN-LSTM layer, then take the vector output from the hidden state of the last layer of the two-layer DNN-LSTM as the decoded vector, and finally normalize the decoded vector through the softmax function of the fully connected layer to obtain the confidence distribution of the currently generated process flow. Use the greedy search algorithm to select the process flow with the highest confidence on the confidence distribution of the currently generated process flow as the generated process flow and output it.

[0045] Preferably, the typical process parameter decision inference algorithm is specifically as follows:

[0046] Step d. Define selection decision rules based on the IF-THEN rule around the process flow knowledge base, tooling selection knowledge base, and process parameter decision knowledge base, so that each rule has a formal description of the IF condition and THEN conclusion;

[0047] Step e. Construct a knowledge base containing process flow decision knowledge, process parameter decision knowledge, and tooling selection knowledge;

[0048] Step f. Search for matching rules in the knowledge base according to the process flow and workpiece material, and finally output the typical process parameters corresponding to the matching rules.

[0049] Preferably, the typical process parameters include process flow, tooling, and process parameters.

[0050] Based on the same inventive concept, on the other hand, the present invention also discloses a template-based inference system for aviation part process specifications, which is used to implement the above-mentioned template-based inference method for aviation part process specifications, including:

[0051] A part feature library, which is used to store the part process features and part geometric structure features of the parts to be processed, as well as store the data of the part process features and part geometric structure features of the historical processed parts.

[0052] A process specification template library, which is used to store process specification template data;

[0053] A process knowledge library, which is used to store the process flow knowledge for making decisions on process flow selection according to part features, store the decision-making knowledge for selecting tooling and tools according to part features, and store the decision-making knowledge for process parameter decisions according to part features;

[0054] A process specification inference module, which is used to obtain the matching process specification template through a feature-based template matching algorithm according to the process features and geometric structure features of the parts to be processed; and according to the matching process specification template, obtain the process parameters involved in compiling the part process specification based on the process specification template through a typical process parameter inference decision algorithm.

[0055] A data fusion module, which is used to fuse the matching process specification template with the typical process parameters obtained by the typical process parameter inference decision algorithm to form a part process specification;

[0056] A human-computer interaction module, which is used to output and display the part process specification obtained by the data fusion module on the process specification generation interface, and allow process designers to edit the obtained part process specification.

[0057] Preferably, the part feature library includes a part process feature library and a part geometric structure feature library; wherein, the part features include part process features and part geometric structure features; the part process feature library is used to store part process feature information; the part geometric structure feature library is used to store part geometric structure feature information.

[0058] Preferably, the process knowledge library includes a process flow knowledge library, a tooling and tool selection knowledge library, and a process parameter decision knowledge library.

[0059] Preferably, the process specification inference module includes a feature-based template matching module and a typical process parameter inference decision module.

[0060] Preferably, the typical process parameter inference decision module includes a process flow inference decision module and a process parameter inference decision module.

[0061] In another aspect, the present invention also discloses a computer device, including a memory, a processor, and a computer program stored on the memory and executable in the processor. When the processor executes the computer program, the above-mentioned template-based inference method for aviation part process specifications is implemented.

[0062] A computer-readable storage medium stores a computer program, which, when executed in a computer processor, implements the above-mentioned template-based inference method for aviation part process specifications.

[0063] Advantages of the present invention:

[0064] The present invention analyzes the traditional industrial design process in the current production and manufacturing process, and proposes a method and system for template-based inference of aviation part process specifications to improve the efficiency and intelligence of part process specification compilation. The template-based inference method for aviation part process specifications can quickly and intelligently generate the required part process specifications through the part features of the parts to be processed and the part features and processing data of the existing historical processed parts, greatly improving the design efficiency of the process design link in the production process, improving the efficiency and intelligence of the production process, and further improving the quality and reliability of part process design. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The foregoing and following specific descriptions of the present invention become clearer when read in conjunction with the following drawings, in which:

[0066] Figure 1 is a flowchart of the method of the present invention;

[0067] Figure 2 is a schematic diagram of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0068] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the following will further illustrate the technical solutions for achieving the objectives of the present invention through specific embodiments. It should be noted that the technical solutions claimed by the present invention include, but are not limited to, the following embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

[0069] Due to the continuous development of the current aviation aircraft manufacturing industry, a relatively high degree of informatization and automation has been achieved in both the product design link and the product manufacturing link in the entire process. For traditional process design, there are significant differences depending on enterprise resources and process habits. Under the same resources and constraints, different process designers may formulate different process specifications. This is a decision-making process with strong empiricism and many influencing factors. Therefore, traditional industrial design usually has the following defects:

[0070] (1) It is not convenient for standardization and has poor consistency. Due to limited personal experience and different design levels, the process documents designed vary from person to person, with poor consistency, difficult optimization, weak standardization, and also difficult inheritance of process design experience.

[0071] (2) The design cycle is long and it cannot quickly respond to product changes. When the product is replaced, even if some of its components are similar to the original ones, new process specifications must be redesigned, resulting in a large amount of repetitive labor and making it difficult to adapt to changes in products and manufacturing environments.

[0072] (3) It is overly dependent on the personal experience and level of process designers. The rationality, operability, and preparation time of process documents mainly depend on the experience and proficiency of process personnel.

[0073] (4) It is difficult to ensure data accuracy. The process needs to process a large amount of graphic information, data information, and text information, and generate a large number of process documents and process data through process design. The traditional process design method for processing data is very backward and cumbersome, and is very prone to errors.

[0074] (5) It is not convenient to centrally utilize the experience and knowledge of process experts.

[0075] (6) Process personnel mainly carry out repetitive and cumbersome work, lacking research on innovative process work.

[0076] The internal structure of an aircraft is complex, with variously shaped and numerous parts. Its development process is very complicated, and most of its products are characterized by multiple varieties and small batches. Along with the transformation and upgrading of the aerospace manufacturing industry, new requirements are put forward for the preparation of process specifications for complex products such as modern aircraft products. Therefore, it is of great significance to realize the rapid and intelligent preparation of part process specifications.

[0077] Based on this, the embodiments of the present invention propose a template-based reasoning method and system for aircraft part process specifications. The present invention first introduces and explains the reasoning method for process specifications. Figure 1 For the flowchart of the method of the present invention, refer to the attached Figure 1 , the template-based reasoning method for aircraft part process specifications specifically includes the following steps:

[0078] Step S1. Extract the three-dimensional digital model information and features of the part to be processed, extract the process features and geometric structure features of the part to be processed, and store them.

[0079] In the embodiments described in the present invention, a feature refers to the set of all feature attributes of a certain part. Each feature attribute is a sub - feature of a certain part. For example, a certain part has three feature attributes: part type, material name, and material specification. However, these feature attributes are collectively referred to as the features of the part. The process features and geometric structure features of the part constitute the part features.

[0080] Step S2. Encode the feature data of the parts to be processed, the feature data of the historically processed parts, the process specification template data, and the process knowledge data using the One - hot method.

[0081] In the embodiments described in the present invention, the specific steps of step S2 are as follows:

[0082] Encode the feature attributes of the part feature library (part type, material name, material specification, material grade, material state, technical requirements, geometric features, etc.) and the template identification values used to uniquely identify the templates in the process specification template library using the One - hot method. Each element is associated with a unique integer index and is converted into a binary vector of the size of the vocabulary length, thereby obtaining the corresponding feature attribute data set and the template identification value data set , where, , X i represents the feature of the i - th part in the part instance, that is, the feature attribute set of the i - th part; i th part; x ij represents the i - th i type of feature attribute corresponding to the i - th j type of part instance, y j represents the number of part instances corresponding to the i - th j type of template identification value data set.

[0083] Step S3. According to the process features and geometric structure features of the parts to be processed, obtain the matching process specification template through a feature - based template matching algorithm.

[0084] In the embodiments described in the present invention, step S3 specifically includes the following steps:

[0085] Step S31. In order to measure the correlation between the feature attributes of the part feature library and the process specification template library, first use the Pearson correlation coefficient to measure the correlation between the feature attributes of the part feature library and the process specification template library. The calculation expression is as follows:

[0086] Equation (1);

[0087] Where, Represents the Pearson correlation coefficient calculation expression; and respectively represent the i th part feature and the template identification value represents the covariance between the i th part feature and the template identification value; represents the variance of the variable; respectively represent the i th part feature and the average value of the template identification value; represents the expected value of the variable; represents the covariance between the i th part feature and the template identification value obtained; represents the variance between the i th part feature and the template identification value obtained;

[0088] Step S32. Further, the mutual information is combined to measure the correlation between the feature attributes of the part feature library and the process specification template library. The calculation expression is as follows:

[0089] Equation (2);

[0090] wherein, represents the mutual information correlation calculation expression; represents the information entropy of the i th part feature; represents the information entropy of the Y th part feature under the condition of the template identification value i ; represents the entropy value calculation, which is used to measure the uncertainty of the template identification value ; represents the calculation of the conditional entropy. Given the i th part feature , there is still uncertainty;

[0091] Step S33. In order to make up for the limitations of different correlation measures and enable full utilization of multiple measures in correlation analysis, this embodiment uses a weighted method to combine the Pearson correlation analysis result and the mutual information correlation analysis result. Through the mutual complementarity of the two measures, the reliability of the measure is improved. The specific calculation expression is as follows:

[0092] Equation (3);

[0093] wherein, is used to measure the i th part feature Given the information, for the template identification value the amount of information provided, and adjust the calculation by weighting; Indicates that the calculated value of mutual information is normalized; α and β are weights, ;

[0094] Step S34. Take the K c feature attributes with the highest correlation in step S33 to establish a set , which is used for the decision tree method construction of subsequent gain value weighted sum and rule generation attribute pruning. After feature correlation analysis, select a certain number of features with a relatively high degree of relevance to the research object as the calculation object for the subsequent steps. This calculation object is the feature set of the selected K c feature attributes D c ; among them, represents the corresponding relevant t part features, and there is ;

[0095] Step S35. As can be seen from step S2, D c is t a set of part instance data samples. Assume that the template identification value data set has m different values, define m different classes ; assume is the number of samples in class C l , and calculate the expected information required for classifying a given sample using formula (4);

[0096] Formula (4);

[0097] where is the probability that any sample belongs to C l , and is estimated using , represents the identification value corresponding to the actual sample; represents the expected information required for classifying a given sample, represents a discrete variable, that is, the sample classification identification value corresponding to template m;

[0098] Step S36. Assume that the feature attribute X i has v different values , using the feature attributeX i Divide D c into v subsets , where D h contains D samples in X i with values a h . If X i is selected as the test attribute (i.e., the best partitioning attribute), then these subsets correspond to the branches grown from the node containing the set D . Assume d lh is the number of samples of class D h in the subset C l . According to X i the calculation expression of the entropy or expected information for partitioning subsets:

[0099] Equation (5);

[0100] where acts as the weight of the h -th subset, and the h -th subset represents the partitioned subset , for information representation; represents the expected information for partitioning subsets according to . The smaller the value of , the better the subset partition result; represents the weighted value of the characteristic attributes set by the part design principle, used to adjust the influence of different characteristics; represents the normalization factor, which is to calculate the sum of a group of samples to ensure reasonable calculation ratios. For a given subset D h , its expected information is: , where , d h represents C l the number of samples in p lh , and D h is the probability that any one data sample in the subset C l belongs to the class

[0101] Step S37. Convert the initial decision tree into a set of rules, where each rule is represented as a condition-action pair. The condition is a series of feature tests, and the action is the corresponding class label; the generated rule set is , r i denotes the i th rule, denotes the condition part of rule r i , denotes the action part of rule r i ;

[0102] Step S38. Perform attribute pruning on the generated rules. The goal of attribute pruning is to delete unimportant features in the rules to simplify the rules and improve the generalization performance; denote the pruned rule set as , where is the pruned rule, is the condition part of rule , is the action part of rule ;

[0103] Step S39. For a new data sample of the part to be processed, use the pruned rule set to perform classification prediction. Each rule in the rule set will test the data sample of the part to be processed x according to the condition part. If the condition part is satisfied, the corresponding action part will be returned as the predicted class label, and finally the process planning template corresponding to the part to be processed will be output.

[0104] In the embodiment depicted in the present invention, the initial decision tree is constructed based on the part feature attributes and the template identification value data set. The specific process and principle are as follows:

[0105] 1. Data preparation

[0106] Input data:

[0107] Feature attributes of the part feature library (such as part type, material name, technical requirements, etc.), and the template identification values of the process planning templates;

[0108] Each feature attribute forms a binary feature vector after One-hot encoding;

[0109] The template identification value is used to define the class label, indicating the template class matched by the part.

[0110] 2. Decision tree generation algorithm

[0111] Decision trees are usually generated based on the following principles:

[0112] (1)Information gain

[0113] For each split of the decision tree, the feature with the largest information gain is selected as the split basis for the current node

[0114] The calculation expression of information gain is as follows:

[0115] ;

[0116] where is the entropy of the current sample set D is the entropy of the subset after partitioning according to a certain feature; D i is the weight of the th i subset; is the i th

[0117] subset;

[0118] During the process of generating a decision tree, the expected information entropy of each feature is calculated by the following formula:

[0119] ;

[0120] where m is the number of categories; p k is the probability that a sample belongs to the k th

[0121] category;

[0122] According to the information gain formula, select the feature that makes the entropy drop the fastest as the partitioning basis.

[0123] 3. Decision tree construction process

[0124] Root node: Select the first feature using the principle of maximizing information gain to partition the sample set D ;

[0125] Child nodes: For each subset D i , repeat the above steps and continue to select the feature with the largest information gain until the stopping condition is met.

[0126] 4. Stopping conditions

[0127] (1)All samples in the sample set belong to the same category;

[0128] (2)The remaining features cannot be further partitioned (or reach the maximum depth);

[0129] (3) The number of samples is too small to continue effective partitioning;

[0130] 5. Initial form of the decision tree

[0131] The generated initial decision tree is a hierarchical tree structure; among them,

[0132] Node: Each node represents a feature test.

[0133] Branch: The branch corresponds to the feature value (such as the binary value of one-hot encoding).

[0134] Leaf node: The leaf node corresponds to a class label, that is, the template identification value.

[0135] Further, in the embodiment depicted in the present invention, for how the initial decision tree in step S37 is converted into a set of rules, the essence of the decision tree is to divide the data set into multiple subsets through tests, and finally make each subset belong to the same class as much as possible. Therefore, each path of the decision tree represents a series of feature conditions, and its end point (leaf node) corresponds to a class label. Such a path can naturally be converted into a condition-action rule. The specific conversion process and principle are as follows:

[0136] 1. Corresponding relationship between decision tree and rules

[0137] (1) The decision tree is a tree structure, including:

[0138] Internal node: Represents a feature test (such as );

[0139] Branch: Represents the value of the feature (such as );

[0140] Leaf node: Represents the class label (action part);

[0141] (2) The rule is a condition-action pair, in the form of:

[0142] ;

[0143] : The condition part of the rule represents the combination of feature tests from the root node to the leaf node;

[0144] : The action part of the rule represents the class label (classification result) finally corresponding to this path;

[0145] Represents the i th specific rule, which is derived from the rule set .

[0146] (3)Each path from the root node to a leaf node represents a rule. For example:

[0147] Path: ;

[0148] Corresponding rule: .

[0149] 2. Conversion process

[0150] Step (1) Traverse the decision tree

[0151] Starting from the root node of the decision tree, traverse down along each path until reaching the leaf node; the path from the root node to the leaf node contains a series of feature tests, and each path corresponds to a rule;

[0152] Step (2) Extract the condition part

[0153] Each internal node (feature test) in the path will become the condition part of the rule, and the condition parts are combined by logical AND (∧). For example:

[0154] Internal node and .

[0155] The extracted condition part is ;

[0156] Step (3) Extract the action part

[0157] The action part is taken from the leaf node (class label) at the end of the path. For example:

[0158] The class label of the leaf node is C The action part is ;

[0159] Step (4) Generate the rule

[0160] Combine the condition part and the action part to form a rule, and the rule format is as follows:

[0161] .

[0162] Furthermore, in the embodiments described in the present invention, for the attribute pruning in step S38, the weighted information gain can be used to evaluate the importance of features, filter out key features, and delete unimportant features. This process not only optimizes the complexity of the rules but also improves the generalization ability. However, it is necessary to balance the degree of simplification and classification performance and ensure the pruning effect through the validation dataset. The specific pruning process is as follows:

[0163] 1. Data Preparation

[0164] Input:

[0165] The generated rule set R , where each rule R i contains a condition part C i and an action part A i ;

[0166] The condition part of each rule consists of several features;

[0167] Objective:

[0168] Delete unimportant features in the condition part, retain the most important features, and improve the generalization performance of the rules.

[0169] 2. Feature Importance Evaluation

[0170] Evaluate the importance of each feature using weighted information gain. The specific steps are as follows:

[0171] (1) Information gain calculation formula:

[0172] Calculate the information gain of a certain feature F j in the rule:

[0173] ;

[0174] Among them, is the entropy of the data set covered by the rule D ; D i Subset divided according to the feature F j ; is the weight of the i th subset; is the i th subset's entropy; means the number of subsets after division, that is, the data set D is divided into subsets ;

[0175] (2) Weighting process

[0176] Introduce the weight F j to the information gain of each feature w j :

[0177] ;

[0178] Weight w j Reflect features F j Global importance in the rule set;

[0179] (3)Normalization processing

[0180] Normalize the weighted information gain for feature ranking and selection. The calculation formula is as follows:

[0181] .

[0182] 3. Feature screening

[0183] Determine the importance of features based on the normalized information gain and delete unimportant features, including the following:

[0184] (1)Importance threshold screening:

[0185] Set a threshold θ and only retain features with a normalized information gain greater than θ. The threshold θ can usually be set through experiments;

[0186] (2)Retain the top k features:

[0187] If the rules are complex, you can directly retain the top k features in the information gain ranking to simplify the condition part of the rules.

[0188] 4. Update the rule set

[0189] The pruned rule set is updated to:

[0190] ;

[0191] is the pruned condition part, only retaining important features; represents the pruned rule action part; represents the rule condition part.

[0192] For example:

[0193] Suppose the rule R i condition part C i contains features and the calculated weighted information gain is:

[0194] F 1: 0.4;

[0195] F 2:0.3;

[0196] F 3:0.2;

[0197] F 4:0.1;

[0198] According to the threshold θ=0.2, retain F 1, F 2, F 3. Delete F 4 , then the pruned rules are:

[0199] .

[0200] Step S4. According to the process characteristics and geometric structure characteristics of the part to be processed, the process parameters involved in completing the part process specification compilation based on the process specification template are obtained through a typical process parameter reasoning decision algorithm.

[0201] In the embodiment described in the present invention, the typical process parameter reasoning decision algorithm includes a process flow decision reasoning algorithm and a typical process parameter decision reasoning algorithm.

[0202] Furthermore, the specific steps of the process flow decision reasoning algorithm are as follows:

[0203] Step a. Build a two-layer DNN-LSTM network (deep neural network combined with long short-term memory network), integrate the advantages of DNN in feature extraction and transformation and the characteristics of LSTM in sequence structure analysis, and effectively combine the two; on this basis, adopt a two-layer DNN-LSTM structure, the upper layer can capture deeper context dependencies, and the lower layer can extract context hidden dependencies, and finally get the output as ,in, H m and U m Respectively represent the network fully connected weight matrix and bias vector between the lower layer DNN-LSTM and the upper layer DNN-LSTM; Represents the hidden state of the DNN-LSTM unit, that is, the direct output of the DNN-LSTM; It represents the final output after linear transformation, and its function is to be used for prediction results of downstream tasks. The input of the two-layer DNN-LSTM includes two categories: the previous process sequence and process attributes;

[0204] For the process sequence, we first normalize the description of each process, use index encoding for each process method, and then embed the word rThe embedding layer obtains the vectors of each process flow, which are m dimensional; all process attributes are represented by character-level one-hot vectors, and each character is represented by a one-hot vector of d dimensions, forming a matrix of up to the maximum number of characters dimensional;

[0205] Step b. Incorporate the process attribute information into each process flow information, and use the process attribute information to better guide the generation of the next process flow, so that the generated process flow is more in line with the operations in the process under the description of the process attribute information. The specific operation is as follows: Adopt the method of complex vector feature fusion by superimposing eigenvectors. After encoding the previous process flow sequence through a double-layer DNN-LSTM layer, superimpose the process flow features at each time step with the current input process attribute features.

[0206] Step c. Use a double-layer DNN-LSTM to maintain the sequential relationship between operations under process attributes. The vector obtained by fusing the process attributes and the process flow sequence is decoded and output to generate the subsequent process flow. In this process, first pass the vector obtained by fusing the process attributes and the process flow sequence through a double-layer DNN-LSTM layer, then take the vector output from the hidden state of the last layer of the double-layer DNN-LSTM as the decoded vector, and finally normalize the decoded vector through the softmax function of the fully connected layer to obtain the confidence distribution of the currently generated process flow. Use the greedy search algorithm to select the process flow with the highest confidence on the confidence distribution of the currently generated process flow as the generated process flow and output it.

[0207] Furthermore, the specific steps of the typical process parameter decision and reasoning algorithm are as follows:

[0208] Step d. Define selection decision rules based on the IF-THEN rules around the process flow knowledge base, tooling and fixture selection knowledge base, and process parameter decision knowledge base, so that each rule has a formal description of the IF condition and THEN conclusion.

[0209] Step e. Construct a knowledge base containing process flow decision knowledge, process parameter decision knowledge, and tooling and fixture selection knowledge.

[0210] Step f. When the user provides information such as process flow and workpiece material as input, the system searches for matching rules in the knowledge base according to these conditions. Once a matching rule is found, the system will automatically push the corresponding typical process parameters to the user according to the THEN conclusion of the rule, without the user having to manually search and set them.

[0211] Step S5. Integrate the matching process specification template with the process parameters obtained by the typical process parameter reasoning and decision algorithm to form a part process specification.

[0212] Step S6. Edit the part process plan to finally obtain a part process plan that can be put into production and use.

[0213] Based on the same inventive concept, the embodiments of the present invention also disclose a template-based reasoning system for aviation part process plans. Since the principle of solving problems by this system is similar to that of the template-based reasoning method for aviation part process plans, the implementation of this system can refer to the implementation of the method, and the repeated parts will not be elaborated. As used hereinafter, the term "unit" or "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated. Figure 2 The structural schematic diagram of the template-based reasoning system for aviation part process plans provided by the embodiments of the present invention is shown as Figure 2 shown. The system can include: a storage module, a process plan reasoning module, a data fusion module, and a human-computer interaction module; wherein,

[0214] The storage module further includes a part feature library, a process plan template library, and a process knowledge library; wherein,

[0215] The part feature library is used to store the features of the parts to be processed and the feature data of the parts processed in the past, and is composed of a part process feature library and a part geometric structure feature library; the part process feature library is used to store the part process feature information; the part geometric structure feature library is used to store the part geometric structure feature information;

[0216] The process plan template library is used to store the process plan template data; the process plan template establishes different types of typical part process plans according to the feature models of different typical parts to form different templates; among them, different process plan templates include the technical requirements, process routes, process processes of the same type of typical parts, and the operation names and related attributes used should be standardized and normalized;

[0217] The process knowledge library includes a process flow knowledge library, a tooling and tool selection knowledge library, and a typical process parameter decision knowledge library; wherein,

[0218] The process flow knowledge library is used to store the process flow knowledge for making decisions on process flow selection according to part features; the process flow knowledge is the process knowledge collected, sorted, classified, and summarized from the past for making decisions on process flow selection, and after being standardized and formalized, it forms rules for making decisions on process flow selection;

[0219] The fixture and tool selection knowledge base is used to store the decision-making knowledge for fixture and tool selection based on part features; the fixture and tool selection knowledge is the process knowledge collected, sorted, classified, and summarized from the past for fixture and tool selection decision-making, and after being standardized and formalized, it forms the rules for fixture and tool selection decision-making;

[0220] The process parameter decision-making knowledge base is used to store the decision-making knowledge for process parameters based on part features; the process parameter decision-making knowledge is the process knowledge collected, sorted, classified, and summarized from the past for process parameter selection decision-making, and after being standardized and formalized, it forms the rules for fixture and tool selection decision-making; the process parameters are the process parameters required for each process in the process specification template;

[0221] The process specification reasoning module includes a feature-based template matching module and a typical process parameter reasoning and decision-making module. Further, the typical process parameter reasoning and decision-making module includes a process flow reasoning and decision-making module and a typical process parameter reasoning and decision-making module; among them,

[0222] The feature-based template matching module obtains a matching process specification template according to the process features and geometric structure features of the part to be machined through a feature-based template matching algorithm;

[0223] The process flow reasoning and decision-making module reasons and decides the process flow of the part process specification according to the matching process specification template;

[0224] The typical process parameter reasoning and decision-making module is used to reason and decide the fixture and tool, process parameter data required for compiling the part process specification; the typical process parameters are the process parameters involved in completing the part process specification compilation based on the process specification template;

[0225] The data fusion module is used to fuse the matching process specification template with typical process parameter data such as the process flow, fixture and tool, and process parameters obtained through reasoning and decision-making to form a part process specification; the process flow contained in the process specification template matched is determined by combining with the process flow data reasoned and decided by the typical process parameter reasoning and decision-making module, so as to complete the correct process specification template; the fixture and tool data required for each process in the process specification template is determined by the fixture and tool data reasoned and decided by the typical process parameter reasoning and decision-making module and filled in; the specific process parameters required to be determined for each process in the process specification template are determined by the fixture and tool data reasoned and decided by the typical process parameter reasoning and decision-making module and filled in;

[0226] The human-computer interaction module is used to output and display the part process plan obtained by the data fusion module on the process plan generation interface of the system, and allows process designers to edit the obtained part process plan. After being edited by the process designers, a part process plan that can be put into production use is obtained.

[0227] It should be noted that the systems, devices, models or units described in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. For the convenience of description, in this specification, when describing the above systems, various units are described separately according to their functions. Of course, when implementing the present invention, the functions of each unit can be implemented in one or more software and / or hardware.

[0228] In addition, in this specification, adjectives such as first and second can only be used to distinguish one element or action, and do not necessarily imply or suggest any actual such relationship or order.

[0229] Furthermore, in another aspect of this embodiment, a computer device is also provided. The computer device includes a processor, an input device, an output device, and a memory, and the processor, input device, output device, and memory are interconnected; wherein, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the steps in the above embodiments.

[0230] Even further, in yet another aspect of this embodiment, a computer-readable storage medium is also provided, which is characterized in that: the computer-readable storage medium stores a computer program, the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to execute the steps in the above embodiments.

[0231] In this embodiment, the processor can be a central processing unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. chips, or combinations of the above types of chips.

[0232] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the corresponding program units in the above method embodiments of the present invention. By running the non-transitory software programs, instructions, and modules stored in the memory, the processor can execute various functional applications of the processor and process the work data, that is, implement the methods in the above method embodiments.

[0233] The memory may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created by the processor and the like. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0234] The one or more units are stored in the memory and, when executed by the processor, execute the methods in the above embodiments.

[0235] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0236] The above are only the preferred embodiments of the present invention, and do not impose any formal limitations on the present invention. Any simple modifications and equivalent changes made to the above embodiments based on the technical essence of the present invention all fall within the protection scope of the present invention.

Claims

1. A template-based inference method for the process planning of aviation parts, characterized in that, Including the following steps: Step S1. Extract the three-dimensional digital model information and features of the part to be processed, extract the process features and geometric structure features of the part to be processed and store them; Step S2. Encode the feature data of the part to be processed, the feature data of the historical processed parts, the process specification template data, and the process knowledge data that need to be used in the One-hot manner; Step S3. According to the process features and geometric structure features of the part to be processed, obtain the matching process specification template through the feature-based template matching algorithm; wherein, the feature-based template matching algorithm includes: Use the Pearson correlation coefficient and mutual information to measure the correlation between the part features and the template identification values respectively, and then combine the two measurement results in a weighted manner to select the feature attributes with the highest correlation to establish a set; Define categories based on the template identification value dataset, calculate the expected information required for sample classification, and divide the dataset through feature attributes to calculate the entropy or expected information of the subset; Convert the initial decision tree into a set of condition-action pair rules and prune the attributes of the rules; Use the pruned rule set to classify and predict the data samples of the part to be processed, and output the matching process specification template; Step S4. According to the process features and geometric structure features of the part to be processed, obtain the typical process parameters involved in the preparation of the part process specification based on the process specification template through the typical process parameter reasoning and decision-making algorithm; Step S5. Integrate the matching process specification template with the typical process parameters obtained by the typical process parameter reasoning and decision-making algorithm to form the part process specification; Step S6. Edit the part process specification to finally obtain the part process specification that can be put into production use.

2. The template-based inference method for the process planning of aviation parts according to claim 1, wherein In step S2, the feature attributes of the part feature library and the template identification values used to uniquely identify the templates in the process planning template library are encoded in a One-hot manner. Each element is associated with a unique integer index and converted into a binary vector of the size of the vocabulary length, so as to obtain the corresponding feature attribute data set and the template identification value data set , where , X i represents the i th part feature in the part instance; x ij represents the i th type of feature attribute corresponding to the j th type of part instance in the y j represents the number of part instances corresponding to the j th type of template identification value data set.

3. The template-based inference method for the process planning of aviation parts according to claim 1, characterized in that In step S3, obtain the matching process specification template through the feature-based template matching algorithm, including: Step S31. Use the Pearson correlation coefficient to measure the correlation between the feature attributes of the part feature library and the process specification template library, and the calculation expression is as follows: Formula (1); Among them, represents the Pearson correlation coefficient calculation expression; and respectively represent the i th part feature and the template identification value; represents the covariance between the i th part feature and the template identification value; represents the variance of the variable; respectively represent the i th part feature and the average value of the template identification value; represents the expected value of the variable; represents the variance between the i th part feature and the template identification value obtained; represents the covariance between the i th part feature and the template identification value obtained; Step S32. Combine the mutual information to measure the correlation between the feature attributes of the part feature library and the process specification template library, and the calculation expression is as follows: Formula (2); Among them, represents the mutual information correlation calculation expression; represents the i information entropy of the th part feature; Y represents the i information entropy of the th part feature under the condition of the template identification value represents the calculation of conditional entropy; Step S33. Use a weighted method to combine the Pearson correlation analysis result and the mutual information correlation analysis result, and the calculation expression is as follows: Formula (3); Among them, used to measure the i th part feature under the given information, the amount of information provided by the template identification value ; indicating the normalization of the mutual information calculation value; α and β are weights, ; Step S34. Take the K c feature attributes with the highest relevance in Step S33 to establish a set ; Among them, represents the corresponding t part features, and there exists ; Step S35. Assume the template identification value data set has m different values, define m different classes ; Assume is the number of samples in class C l , and calculate the expected information required for classifying a given sample using formula (4); Formula (4); where is the probability that any sample belongs to C l and is estimated using ; represents the identification value corresponding to the actual sample; represents the expected information required to calculate the classification of the given sample, represents a discrete variable; Step S36. Set characteristic attributes X i has v different values , and use the characteristic attribute X i to D c divide it into v subsets ; Let d lh be the number of samples of class D h in the subset C l . According to X i , the calculation expressions for the entropy or expected information of dividing subsets are as follows: Formula (5); Among them, serves as the weight of the h th subset, and the h th subset represents the divided subset ; represents the weighted value of the characteristic attributes set by the part design principle; represents the expected information for dividing subsets according to ; represents the normalization factor; Step S37. Convert the initial decision tree into a set of rules, where each rule is represented as a condition-action pair, with the condition being a series of feature tests and the action being the corresponding class label; the generated set of rules is , r i denotes the i th rule, denotes the r i condition part of rule denotes the r i action part of rule Step S38. Perform attribute pruning on the generated rules, and the pruned rule set is , where is the pruned rule, is the condition part of rule , and is the action part of rule . Step S39. For the new data sample of the part to be machined, use the trimmed rule set to perform classification prediction and finally output the process planning template corresponding to the part to be machined.

4. The method for reasoning the process plan of aviation parts based on templates according to claim 1, characterized in that, In step S4, the typical process parameter reasoning and decision-making algorithm includes a process flow decision-making reasoning algorithm and a typical process parameter decision-making reasoning algorithm.

5. The template-based inference method for the process planning of aviation parts according to claim 4, wherein The specific process of the process flow decision-making reasoning algorithm is as follows: Step a. Index code each process flow method, and then obtain the vector of each process flow through an embedding layer with an embedding size of r as m dimension; Represent all process attributes using character-level one-hot vectors, and represent each character with n -dimensional one-hot vectors to form a matrix with a maximum number of characters N × n dimension; Step b. Adopt the complex vector feature fusion method of superimposing feature vectors. After the pre-order process flow sequence is encoded by the double-layer DNN-LSTM layer, the process flow features at each time step are vectorially superimposed with the current input process attribute features; Step c. Use the double-layer DNN-LSTM to maintain the sequential relationship between processes under the process attributes, and the vector obtained by fusing the process attributes and the process flow sequence is decoded and output to generate the subsequent process flow.

6. The template-based reasoning method for the process planning of aviation parts according to claim 5, characterized in that Step c specifically includes: First, pass the vector after fusing the process attributes and process flow sequence through a two-layer DNN-LSTM layer. Then, take the vector output from the hidden state of the last layer of the two-layer DNN-LSTM as the decoded vector. Finally, normalize the decoded vector through the softmax function of the fully connected layer to obtain the confidence distribution of the currently generated process flow. Use the greedy search algorithm to select the process flow with the highest confidence on the confidence distribution of the currently generated process flow as the generated process flow and output it.

7. The template-based reasoning method for the process planning of aviation parts according to claim 4, characterized in that The specific typical process parameter decision and reasoning algorithm is as follows: Step d. Define selection decision rules based on the IF-THEN rules around the process flow knowledge base, tooling selection knowledge base, and process parameter decision knowledge base, so that each rule has a formal description in the form of an IF condition and a THEN conclusion. Step e. Construct a knowledge base containing process flow decision knowledge, process parameter decision knowledge, and tooling selection knowledge. Step f. Search for matching rules in the knowledge base according to the process flow and workpiece material, and finally output the typical process parameters corresponding to the matching rules.

8. The template-based reasoning method for aviation part process planning according to claim 7, wherein The typical process parameters include process flow, tooling, and process parameters.

9. A template-based process planning inference system for aviation parts, the system being used to implement the template-based process planning inference method for aviation parts according to any one of the above claims 1-8, characterized in that, It includes: A part feature library for storing the part process features and part geometric structure features of the parts to be machined, as well as storing the part process feature and part geometric structure feature data of the historically machined parts. A process specification template library for storing process specification template data. A process knowledge base for storing process flow knowledge for making process flow selection decisions based on part features, storing tooling selection decision knowledge based on part features, and storing process parameter decision knowledge based on part features. A process specification reasoning module for obtaining a matching process specification template through a feature-based template matching algorithm according to the process features and geometric structure features of the parts to be machined. And according to the matching process specification template, obtain the typical process parameters involved in compiling the part process specification based on the process specification template through the typical process parameter reasoning and decision algorithm. A data fusion module for fusing the matching process specification template with the typical process parameters obtained by the typical process parameter reasoning and decision algorithm to form a part process specification. A human-computer interaction module for outputting and displaying the part process specification obtained by the data fusion module on the process specification generation interface, and allowing process designers to edit the obtained part process specification.

10. The template-based inference system for aviation part process planning according to claim 9, wherein, The part feature library includes a part process feature library and a part geometric structure feature library; among them, the part features include part process features and part geometric structure features; the part process feature library is used to store part process feature information; the part geometric structure feature library is used to store part geometric structure feature information.

11. The template-based process planning inference system for aviation parts according to claim 9, characterized in that, The process knowledge base includes a process flow knowledge base, a tooling selection knowledge base, and a process parameter decision knowledge base.

12. The template-based process planning inference system for aviation parts according to claim 9, wherein, The process specification reasoning module includes a feature-based template matching module and a typical process parameter reasoning and decision module.

13. The template-based inference system for aviation part process planning according to claim 12, wherein The typical process parameter reasoning and decision module includes a process flow reasoning and decision module and a process parameter reasoning and decision module.

Images