A knowledge graph-based method for predicting quality problems in aviation products

Abstract

This invention relates to the field of knowledge graph collaborative filtering algorithms, specifically a knowledge graph-based method for predicting quality problems in aviation products. The specific steps of the method are as follows: S1, collaborative filtering algorithm; S2, similarity calculation; S3, problem cause prediction based on ICF algorithm; S4, quality problem prediction based on UCF algorithm. By fusing the semantic similarity of causes and the similarity of causes based on historical information, the similarity of the causes leading to different quality problems is calculated, thereby predicting the cause of a specific quality problem. Simultaneously, by calculating the similarity of quality problems caused by the causes of the problems, correlation analysis of quality problems is achieved, meaning that the occurrence of one quality problem may lead to the occurrence of another quality problem. This enables rapid control and early prevention of quality problems.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph collaborative filtering algorithms, specifically a knowledge graph-based method for predicting quality problems in aviation products. Background Technology

[0002] Numerous quality issues persist in the current research and production of aerospace products, some of which are recurring. This is because enterprises' current quality management models have not yet shifted from a "firefighting" to a "prevention" approach. In actual production, quality problems often have multiple causes; the same cause can lead to different quality issues, and these issues are often interconnected. This method introduces a collaborative filtering algorithm for quality problem correlation analysis and cause prediction, using cause fusion similarity calculations to infer potential causes of quality problems. Similarly, by analyzing the similarity of quality problems, early prevention of quality issues is achieved.

[0003] Currently, the level of decision-making regarding the quality of aviation products is low. The new production model has shifted from small-scale research and production to mass production. The sharp increase in workload has led to more stringent quality requirements and placed higher demands on the level of decision-making regarding quality issues. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a knowledge graph-based method for predicting quality issues in aviation products.

[0005] A knowledge graph-based method for predicting quality problems in aviation products comprises the following steps:

[0006] S1, Collaborative Filtering Algorithm:

[0007] S11. Collaborative filtering algorithms are used to implement prediction and recommendation functions. Collaborative filtering recommendation algorithms are divided into two types: starting from users' historical behavior data, the algorithm finds users' preferences by mining and makes recommendations.

[0008] S12. Collaborative filtering algorithms require similarity analysis and calculation;

[0009] S2, Fusion Similarity Calculation:

[0010] S21. Vectorized Representation of Knowledge Graph: Using the TransE model, a vector representation model for knowledge graphs, the triples in the knowledge graph are represented as (h, r, t), where h represents the head entity, r represents the relation, and t represents the tail entity. Each entity and relation in the product quality knowledge graph is represented as a vector, and the TransE model is represented as: h + r ≈ t.

[0011] S22. Semantic Similarity of Causes: Based on the vectorized representation of knowledge graphs, the cause entities in the product quality knowledge graph are represented as low-dimensional vectors. When the dimension space is d, the cause can be vectorized as: P i =(E 1i E 2i ,…,E di ) T E ki For reason P i The value in the k-th dimension;

[0012] S23. Similarity of causes in historical information: By introducing the TF-IDF statistical algorithm, for a given problem, the causal keyword v i The weight calculation formula is as follows:

[0013]

[0014] Where e is the tail entity in the triple (Reason, lead-to, Question), F(e, v) i ) is the tail entity and v i The number of times they appear together in the dataset, where n is the total number of triples;

[0015] S24. The weight of the cause is selected as the basis for the cause recommendation, and the result is denoted as R. m×n The pattern uses the law of cosines to calculate the two cause vectors I. i ,I j The similarity is calculated using the following formula:

[0016]

[0017] S25. Similarity of Fusion Causes: The algorithm for obtaining the similarity of fusion causes is as follows: Combining the semantic similarity of causes and the similarity of causes in historical information.

[0018] s(I i ,I j )=αs k (I i ,I j )+βs u (I i ,I j )

[0019] Where α and β represent similarity fusion factors, α∈[0,1], β∈[0,1], α+β=1;

[0020] S26. Rating Prediction: By calculating the similarity of the fusion causes, weighted scores are assigned to the reasons why quality issues were not found. The predicted causes are then sorted in descending order, generating a Top-N recommendation list for the issue. The formula for predicting the weighted scores of the causes is as follows:

[0021]

[0022] Among them, R i,k S is the score for all causes of problem i. k,j The sum of the product of the similarities of causes in cause j;

[0023] Similarly, the similarity calculation and score prediction for quality issues are performed using the same method;

[0024] S3. Problem Cause Prediction Based on ICF Algorithm: When a product has a quality problem, the ICF algorithm is applied to find the most similar problem cause by calculating the similarity of different problem causes, thereby achieving problem cause prediction.

[0025] S4. Quality Problem Prediction Based on UCF Algorithm: Based on the UCF algorithm, quality problem correlation analysis is performed to calculate the similarity of different quality problems and rank them. The principle and process are the same as the cause prediction of quality problems.

[0026] The two types of step S11 are user-based collaborative filtering algorithm (UCF) and item-based collaborative filtering algorithm (ICF).

[0027] Step S11 starts with user historical behavior data, when the user set U = (u1, u2, u3, ..., u m The product set I = (i1, i2, i3, ..., i...) n If ), then the rating matrix can be represented as R. m×n The expression is as follows:

[0028]

[0029] The similarity calculation methods in step S12 include cosine similarity, Euclidean distance, and Pearson correlation coefficient.

[0030] In step S21, Euclidean distance is used to measure cause I. i ,I j Similarities between them:

[0031]

[0032] The specific process of step S3 is as follows: the number table of quality problems and their causes is used as input to the problem cause prediction model based on ICF, then the similarity of the problem causes is calculated, a similarity matrix is ​​generated, the interest of quality problems in the problem causes is calculated, and the most likely problem cause is predicted by ranking.

[0033] The beneficial effects of this invention are: by fusing semantic similarity of causes and historical information similarity of causes, the similarity of the causes leading to different quality problems is calculated, thereby enabling the prediction of the causes of a specific quality problem; simultaneously, by calculating the similarity of quality problems caused by the causes of the problems, correlation analysis of quality problems is achieved, that is, when one quality problem occurs, it will cause another quality problem to occur. This leads to rapid control and early prevention of quality problems. Attached Figure Description

[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0035] Figure 1 This is a flowchart of the cause prediction algorithm of the present invention;

[0036] Figure 2 This invention provides an overall framework for predicting quality problems based on a knowledge graph collaborative filtering algorithm. Detailed Implementation

[0037] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below.

[0038] like Figure 1 and Figure 2 As shown, a knowledge graph-based method for predicting quality problems in aviation products comprises the following steps:

[0039] S1, Collaborative Filtering Algorithm:

[0040] S11. Collaborative filtering algorithms are used to implement prediction and recommendation functions. Collaborative filtering recommendation algorithms are divided into two types: starting from users' historical behavior data, the algorithm finds users' preferences by mining and makes recommendations.

[0041] S12. Collaborative filtering algorithms require similarity analysis and calculation;

[0042] S2, Fusion Similarity Calculation:

[0043] S21. Vectorized Representation of Knowledge Graphs: When using knowledge graphs to analyze and predict product quality issues, consider the vectorized representation of knowledge graphs. Use the classic knowledge graph vector representation model, TransE model. The triples in the knowledge graph are represented as (h, r, t), where h represents the head entity, r represents the relation, and t represents the tail entity. Represent each entity and relation in the product quality knowledge graph as a vector. The TransE model is represented as: h + r ≈ t.

[0044] S22. Semantic Similarity of Causes: Based on the vectorized representation of knowledge graphs, the cause entities in the product quality knowledge graph are represented as low-dimensional vectors. When the dimension space is d, the cause can be vectorized as: P i =(E 1i E 2i ,…,E di ) T E ki For reason P i The value in the k-th dimension;

[0045] S23. Similarity of Historical Information Causes: The TF-IDF statistical algorithm is introduced. This algorithm uses weighted techniques to achieve information retrieval and text mining. This method is used to measure the similarity of historical information causes. The calculation result of triples-cause keyword weights represents the historical access data information of the causes. Therefore, for a specific question, the cause keyword v... i The weight calculation formula is as follows:

[0046]

[0047] Where e is the tail entity in the triple (Reason, lead-to, Question), F(e, v) i ) is the tail entity and v i The number of times they appear together in the dataset, where n is the total number of triples;

[0048] S24. The weight of the cause is selected as the basis for the cause recommendation, and the result is denoted as R. m×n The pattern uses the law of cosines to calculate the two cause vectors I. i ,I j The similarity is calculated using the following formula:

[0049]

[0050] S25. Similarity of Fusion Causes: The algorithm for obtaining the similarity of fusion causes is as follows: Combining the semantic similarity of causes and the similarity of causes in historical information.

[0051] s(I i ,I j )=αsk (I i ,I j )+βs u (I i ,I j )

[0052] Where α and β represent similarity fusion factors, α∈[0,1], β∈[0,1], α+β=1;

[0053] S26. Rating Prediction: By calculating the similarity of the fusion causes, weighted scores are assigned to the reasons why quality issues were not found. The predicted causes are then sorted in descending order, generating a Top-N recommendation list for the issue. The formula for predicting the weighted scores of the causes is as follows:

[0054]

[0055] Among them, R i,k S is the score for all causes of problem i. k,j The sum of the product of the similarities of causes in cause j;

[0056] Similarly, the similarity calculation and score prediction for quality issues are performed using the same method;

[0057] S3. Problem Cause Prediction Based on ICF Algorithm: When a product has a quality problem, there is often more than one cause. The occurrence of quality problems is often caused by many factors. In order to analyze the common causes of quality problems, we can learn from the ideas of user product recommendation on the Internet, replace the user with the quality problem name, and treat the problem cause as the purchased product. By applying the ICF algorithm, we can find the most similar problem cause by calculating the similarity of different problem causes, thereby realizing problem cause prediction.

[0058] S4. Quality Problem Prediction Based on UCF Algorithm: Quality problem correlation analysis based on UCF algorithm mainly calculates the similarity of different quality problems and ranks them to find the set of quality problems caused by the same problem. When a quality problem occurs, it finds other quality problems that are most likely to be caused by that quality problem, thereby realizing the early prevention of quality problems. Its principle and process are the same as the cause prediction of quality problems.

[0059] The two types of step S11 are user-based collaborative filtering algorithm (UCF) and item-based collaborative filtering algorithm (ICF).

[0060] Step S11 starts with user historical behavior data, when the user set U = (u1, u2, u3, ..., u m The product set I = (i1, i2, i3, ..., i...) nIf ), then the rating matrix can be represented as R. m×n The expression is as follows:

[0061]

[0062] The similarity calculation methods in step S12 include cosine similarity, Euclidean distance, and Pearson correlation coefficient.

[0063] In step S21, Euclidean distance is used to measure cause I. i ,I j Similarities between them:

[0064]

[0065] The above formula represents the similarity between causes; the higher the result, the higher the similarity. This allows the generation of a semantic similarity matrix for the causes, which can then be integrated into the collaborative filtering algorithm.

[0066] The specific process of step S3 involves using a table of quality issues and their causes as input to an ICF-based cause prediction model, calculating the similarity between the causes, generating a similarity matrix, calculating the interest of quality issues in their causes, and predicting the most likely cause through ranking. The specific algorithm flow is as follows: Figure 1 As shown.

[0067] By integrating the semantic similarity of causes and the similarity of causes in historical information, the similarity of the causes of different quality problems is calculated, thereby enabling the prediction of the causes of a specific quality problem. At the same time, by calculating the similarity of quality problems caused by the causes of the problems, the correlation analysis of quality problems is realized, that is, when one quality problem occurs, it will cause another quality problem to occur, thereby enabling rapid control and early prevention of quality problems.

[0068] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely prisms of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph-based method for predicting quality problems in aviation products, characterized in that: The specific steps are as follows: S1, Collaborative Filtering Algorithm: S11. Collaborative filtering algorithms are used to implement prediction and recommendation functions. Collaborative filtering recommendation algorithms are divided into two types: starting from users' historical behavior data, the algorithm finds users' preferences by mining and makes recommendations. S12. Collaborative filtering algorithms require similarity analysis and calculation; S2, Fusion Similarity Calculation: S21. Vectorized Representation of Knowledge Graphs: Using the TransE model, a vector representation model for knowledge graphs, the triples in the knowledge graph are represented as (h, r, t), where h represents the head entity, r represents the relation, and t represents the tail entity. Each entity and relation in the product quality knowledge graph is represented as a vector in the TransE model as follows: ; S22. Semantic Similarity of Causes: Based on the vectorized representation of knowledge graphs, the cause entities in the product quality knowledge graph are represented as low-dimensional vectors. When the dimension space is d, the cause can be vectorized as follows: ,in, For the reason The value in the k-th dimension; S23. Similarity of causes in historical information: By introducing the TF-IDF statistical algorithm, for a specific problem, the key causes... The weight calculation formula is as follows: ; in, It is the tail entity in the triple (Reason, lead-to, Question). Is the tail entity and The number of times they appear together in the dataset It is the total number of triples; S24. The weight of the cause is selected as the basis for the cause recommendation, and the result is denoted as... The pattern uses the law of cosines to calculate the two cause vectors. The similarity is calculated using the following formula: ; S25. Similarity of Fusion Causes: The algorithm for obtaining the similarity of fusion causes is as follows: Combining the semantic similarity of causes and the similarity of causes in historical information. ; in, Represents the similarity fusion factor. ; S26. Rating Prediction: By calculating the similarity of the fusion causes, weighted scores are assigned to the reasons why quality issues were not found. The predicted causes are then sorted in descending order, generating a Top-N recommendation list for the issue. The formula for predicting the weighted scores of the causes is as follows: ; in, Rate all causes for problem i. The sum of the product of the similarities of causes in cause j; Similarly, the similarity calculation and score prediction for quality issues are performed using the same method; S3. Problem Cause Prediction Based on ICF Algorithm: When a product has a quality problem, the ICF algorithm is applied to find the most similar problem cause by calculating the similarity of different problem causes, thereby achieving problem cause prediction. S4. Quality Problem Prediction Based on UCF Algorithm: Based on the UCF algorithm, quality problem correlation analysis is performed to calculate the similarity of different quality problems and rank them. The principle and process are the same as the cause prediction of quality problems.

2. The method for predicting quality problems in aviation products based on knowledge graphs according to claim 1, characterized in that: The two types of step S11 are user-based collaborative filtering algorithms and item-based collaborative filtering algorithms.

3. The method for predicting aviation product quality problems based on knowledge graphs according to claim 1, characterized in that: Step S11, starting from user historical behavior data, when the user set Product collection Then the rating matrix can be expressed as The expression is as follows: .

4. The method for predicting aviation product quality problems based on knowledge graphs according to claim 1, characterized in that: The similarity calculation methods in step S12 include cosine similarity, Euclidean distance, and Pearson correlation coefficient.

5. The method for predicting quality problems in aviation products based on knowledge graphs according to claim 1, characterized in that: In step S21, the cause is measured using Euclidean distance. Similarities between them: 。 6. The method for predicting aviation product quality problems based on knowledge graphs according to claim 1, characterized in that: The specific process of step S3 is as follows: the number table of quality problems and their causes is used as input to the problem cause prediction model based on ICF, then the similarity of the problem causes is calculated, a similarity matrix is ​​generated, the interest of quality problems in the problem causes is calculated, and the most likely problem cause is predicted by ranking.

Images