Workshop manufacturing service reliability evaluation method based on convolutional graph neural network

Abstract

The invention discloses a workshop manufacturing service reliability evaluation method based on a convolutional graph neural network, and aims to help enterprises to improve the reliability of manufacturing services. Comprising the following steps: collecting key influence factors of manufacturing service reliability, and analyzing the internal relation between the factors in combination with workshop characteristics; and a new connotation of manufacturing service reliability is given from the perspective of a workshop. A workshop manufacturing service reliability evaluation model is constructed based on the key influence factors and the relation between the internal relation of the key influence factors and the manufacturing service reliability. Aiming at the characteristics of the model, a new convolutional graph neural network framework is provided, a graph neural network is used for fitting the relationship between influence factors and reliability, a neural network model is trained through historical data, and the reliability of the manufacturing service is predicted by using the trained model. According to the method, the problem that the workshop manufacturing service reliability is difficult to evaluate is effectively solved, the stability of the workshop manufacturing service is improved, and the product quality is guaranteed.

Description

Technical Field

[0001] The present application relates to the field of workshop manufacturing services and deep learning, and in particular to a workshop manufacturing service reliability assessment method based on convolutional graph neural network. Technical Background

[0002] In the context of mass customization, some manufacturing companies encapsulate workshop manufacturing resources into manufacturing services and complete manufacturing tasks through service collaboration. Encapsulating manufacturing resources into manufacturing services can effectively avoid the difficulties caused by the heterogeneity and complexity of manufacturing resources in the resource allocation process, and improve the sharing level and production efficiency of workshop manufacturing resources. As users' personalized requirements continue to increase, the degree of personalization of manufacturing tasks also increases. Manufacturing services need to frequently adjust and configure their resources to cope with complex personalized orders. Frequent resource reorganization makes the reliability assessment of manufacturing services complex and challenging. The reliability of manufacturing services is directly related to the final quality of the product. Ensuring the reliability of manufacturing services is crucial to improving product quality.

[0003] Khamdi Mubarok et al. (Mubarok K, Xu X, Ye X, et al. Manufacturing service reliability assessment in cloud manufacturing [J]. Procedia CIRP, 2018, 72.) proposed a hierarchical manufacturing service reliability assessment model to solve the reliability assessment problem of manufacturing resources in cloud manufacturing environments. The model comprehensively considers logistics reliability and cloud service performance by evaluating the reliability of four levels: components, machines, systems, and cloud systems, and ensures the service quality of the entire cloud manufacturing system. However, this method is targeted at cloud manufacturing environments and is not applicable to the workshop level.

[0004] At the workshop level, manufacturing services correspond to the manufacturing process of workshop production, and their reliability will be challenged by a variety of uncertain factors. Employees' operating errors, unexpected equipment failures, and quality defects of raw materials will directly affect the reliability of manufacturing services. In addition, indirect factors such as production environment, processing methods, and measurement methods will also affect the reliability of services. What's more complicated is that these factors do not exist in isolation. They are intertwined in the production process, forming an intricate interactive relationship. This dynamic interdependence makes it very difficult to evaluate the reliability of manufacturing services. Therefore, manufacturing companies need a set of scientific and effective workshop manufacturing service reliability evaluation methods to ensure that services can still maintain the predetermined performance and production quality in complex production environments and variable resource combinations. Summary of the invention

[0005] The purpose of the present invention is to provide a workshop manufacturing service reliability assessment method based on convolutional graph neural network to address the above problems.

[0006] The technical solution to achieve the purpose of the present invention is: a workshop manufacturing service reliability assessment method based on convolutional graph neural network, the method comprising the following steps:

[0007] Conduct failure mode and effect analysis on manufacturing services and extract factors affecting the reliability of manufacturing services;

[0008] Provide new connotations of manufacturing service reliability from the perspective of the workshop and quantify the service reliability value;

[0009] Based on the influencing factors, key influencing factors of manufacturing service reliability are screened out;

[0010] Based on the key influencing factors, explore the internal relationships between the key influencing factors;

[0011] Based on the relationship between the key factors and their internal relationships and service reliability, a manufacturing service reliability evaluation model is constructed;

[0012] According to the evaluation model, a key influencing factor relationship diagram is constructed;

[0013] According to the characteristics of the key influencing factor relationship graph, the graph convolutional neural network algorithm is improved;

[0014] Build a solution using an improved convolutional graph neural network algorithm;

[0015] According to the solution, a deep learning algorithm is used to obtain a predicted value of manufacturing service reliability.

[0016] Furthermore, the step of performing failure mode and impact analysis on manufacturing services and extracting factors affecting the reliability of manufacturing services includes:

[0017] Manufacturing service is the smallest service unit completed by the collaborative efforts of production factors, and the manufacturing service is mapped to a certain production link in the workshop;

[0018] Use the 5M1E method (Man, Machine, Material, Method, Measurement, Environment) to fully describe the physical failure modes of manufacturing services;

[0019] According to the physical failure forms of manufacturing services, the causes of failure are analyzed and the factors affecting the reliability of manufacturing services are extracted.

[0020] Furthermore, the step of giving a new connotation of manufacturing service reliability from the perspective of the workshop and quantifying the service reliability value includes:

[0021] From the perspective of the workshop, manufacturing services correspond to the production process of the workshop;

[0022] The performance of the production process is usually reflected by product quality and execution efficiency. The indicators considered include product quality consistency, product qualification rate, effective operation rate, and average load rate;

[0023] The CRITIC method is used to determine the weights of the indicators, and the weighted sum of the indicators is used as the reliability label.

[0024] Furthermore, the step of screening out key influencing factors of manufacturing service reliability according to the influencing factors includes:

[0025] According to the influencing factors, calculate the importance of the factors and the information contribution rate of the factors;

[0026] The risk priority number (RPN) is used to evaluate the weight of the impact factor on service reliability. The importance of the i-th impact factor is RPN = S × O × D, where S is severity, which measures the severity of the impact that a failure may have on the system, user or environment. O is frequency, which measures the probability or frequency of a failure. D is detection, which measures the ability of existing control measures to detect failures before they occur. (RPN) i Represents the sum of the risk order numbers of the potential failure modes that can be caused by the i-th influencing factor;

[0027] According to the importance calculation method of the influencing factors, removing the influencing factors with lower importance;

[0028] Quantify the remaining influencing factors after removing the influencing factors with lower importance;

[0029] According to the quantified influencing factors, the coefficient of variation (CV) is used to evaluate the information contribution rate of the influencing factors to the reliability assessment. The coefficient of variation of the i-th influencing factor is Among them, σ i is the standard deviation of the i-th influencing factor, μ i is the average value of the i-th influencing factor;

[0030] According to the information contribution rate calculation method, the influencing factors with low information contribution rate are removed to obtain the key influencing factors of service reliability.

[0031] Furthermore, the step of mining the internal relationship between the key influencing factors according to the key influencing factors includes:

[0032] There is a promoting or inhibiting relationship between the key influencing factors, which is the main relationship between two factors;

[0033] According to the production factors directly affected by the key influencing factors, the relationship between the corresponding production factors belongs to the secondary relationship between the key influencing factors;

[0034] The internal connections between different key influencing factors include primary relationships and secondary relationships.

[0035] Furthermore, the step of constructing a manufacturing service reliability evaluation model based on the key factors and their internal relationships with service reliability includes:

[0036] There is a function f between the key factors and their relationships and service reliability: Relationship,The key influencing factors K and their relationships S are mapped to the manufacturing service reliability R, and ω is the parameter set to be learned.

[0037] Furthermore, the step of constructing a key influencing factor relationship diagram according to the evaluation model includes:

[0038] Each of the key influencing factors is set as a node in a key influencing factor relationship graph, and the internal connections between different key influencing factors are set as edges in the key influencing factor relationship graph.

[0039] Furthermore, the step of improving the graph convolutional neural network algorithm according to the characteristics of the key influencing factor relationship graph includes:

[0040] The spectral graph convolution is improved, and the feature vectors of the nodes and the edges are aggregated using Chebyshev convolution ChebConv, and the convolution kernel of the ChebConv is: Where Λ=2Λ / λ max -I,λ max is the largest eigenvalue of L, T i represents the i-th Chebyshev polynomial, K is the order of the Chebyshev polynomial, θ i are learnable parameters.

[0041] The convolution formula of the node's feature vector x and the Chebyshev convolution kernel is: T i (L)=2LT i-1 (L)-T i-2 (L) Where, T0(L) = 1, T1(L) = L. L = 2L / λ max-I is the modified graph Laplacian matrix.

[0042] ChebConv can recursively obtain high-order neighborhood information, improve local information perception, and be more stable when disturbed. However, it does not directly use the features in the edge;

[0043] According to the ChebConv method, feature fusion ChebConv is used to solve the problem that ChebConv does not directly use the features in the edge;

[0044] According to the feature fusion ChebConv method, an edge graph-feature fusion ChebConv with an edge graph is used to solve the problem that the traditional graph neural network structure cannot comprehensively consider the interference between the edge information in the graph.

[0045] Furthermore, the step of constructing a solution using an improved graph convolutional neural network algorithm includes:

[0046] The attribute graph G a ={X,E f ,A,Y} is input into the improved graph convolutional neural network. Where X is the attribute matrix of key influencing factors of manufacturing service reliability, E f is the relationship matrix between influencing factors, and the adjacency matrix A represents the connection relationship between nodes, which is generated by judging whether the influencing factors have a promoting or inhibiting relationship. Y is the corresponding manufacturing service reliability label;

[0047] The G a Use the feature extraction layer to extract features and get Node feature matrix d v Represents the dimension of the output node features; edge feature matrix d e Represents the dimension of the output edge feature; n and m represent the number of nodes and edges in the attribute graph respectively;

[0048] The G b Use the graph convolution layer to perform convolution operations on relevant node features to obtain Represents the updated node feature matrix;

[0049] The updated node representation is input into the graph pooling layer to obtain a feature vector As the global embedding vector of the graph;

[0050] Select the global maximum pooling layer, and the input is the residual of the node feature matrix

[0051] The feature vector Input to the graph regression layer to obtain a scalar value that measures the reliability of the task;

[0052] The regression layer is implemented by a fully connected neural network.

[0053] Furthermore, the step of using feature fusion ChebConv to solve the problem that ChebConv does not directly use the features in the edge includes:

[0053] Before the convolution operation, the edge features and neighbor node features are fused to obtain the node features containing edge information, thus realizing the direct use of edge features.

[0055] Furthermore, the ChebConv with edge graph-feature fusion added to the edge graph is used to solve the problem that the traditional graph neural network structure cannot comprehensively consider the interference between the edge information in the graph. This step includes:

[0056] Reasons for adding edge graph module:

[0057] According to the internal relationships between the key influencing factors, the interference between different internal relationships should also be considered. For example, when the interaction between humans and machines shows a positive or negative trend, this will indirectly affect other interactions related to humans or machines;

[0058] The limitation of graph neural networks is that their edges can only reflect the direct connection between two nodes. When the information of an edge changes, it cannot have a direct impact on other edges in the network.

[0059] Specific steps:

[0060] Constructing an edge graph of the key influencing factor relationship graph according to the key influencing factor relationship graph;

[0061] Among them, the node information of the edge graph is the edge information in the original graph, and the edge represents the interference relationship between the edge information of the original graph;

[0062] The edge graph is used to extract the features of the edges in the original image, and then it is fused with the edge features of the original image. After adding the edge graph features, any change in the information of any edge in the original image will affect other edges.

[0063] The beneficial effects of the present invention are as follows: the present invention proposes a workshop manufacturing service reliability evaluation method based on a convolutional graph neural network, which comprehensively considers the influencing factors of workshop manufacturing service reliability and the internal connection between the influencing factors, thereby improving the efficiency and accuracy of workshop manufacturing service reliability evaluation. Considering the continuous improvement of user personalized requirements, workshop manufacturing services need to frequently adjust and configure their resources to cope with complex personalized orders, making the reliability evaluation of manufacturing services complex and challenging, so a workshop-oriented manufacturing service reliability evaluation method is needed to improve the reliability of workshop manufacturing services. The present invention models the relationship between the influencing factors of manufacturing service reliability and reliability, constructs a key influencing factor relationship graph, defines the uncertain relationship between the key influencing factors in the key influencing factor relationship graph, and gives a new connotation of manufacturing service reliability from the perspective of the workshop, and quantifies the service reliability value, and inputs the feature matrix and reliability label value formed by the key influencing factor relationship graph into the improved graph neural network for training, and obtains a scalar value for measuring task reliability through a feature extraction layer, a graph convolution layer, a pooling layer, and a regression layer. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flow chart of a workshop manufacturing service reliability assessment method based on a convolutional graph neural network provided in this application.

[0065] Figure 2 Schematic diagram of the method for extracting factors affecting manufacturing service reliability in this application.

[0066] Figure 3 A schematic diagram of the construction process of the network attribute graph in this application.

[0067] Figure 4 This is a schematic diagram of the source of the attribute graph data in this application.

[0068] Figure 5 It is a schematic diagram of the structure of the manufacturing service reliability evaluation network in the present application.

[0069] Figure 6 This is a schematic diagram of the combination of the edge graph network and the spectral graph convolutional network model for feature fusion in the application content. Specific implementation methods

[0070] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The influencing factors of the embodiments of the present application described and shown in the drawings here can be analyzed and extracted in various different workshops. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but only represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0071] In a first aspect, the present invention provides a workshop manufacturing service reliability assessment method based on convolutional graph neural network.

[0072] The present invention discloses a method for evaluating the reliability of workshop manufacturing services based on convolutional graph neural networks. The method of graph neural network (GNN) can solve nonlinear problems well, especially convolutional graph neural network (ConvGNN), which can process graph data that retains more information. Utilizing the latest development of ConvGNN, the embodiment of the present invention proposes a method for evaluating the reliability of manufacturing services that combines ConvGNN to fit the uncertain relationship between the factors affecting the reliability of workshop manufacturing services and their relationship and service reliability. Specifically, it is proposed to convert the key influencing factor relationship graph into a graph model, input the feature matrix and reliability label values ​​formed by the key influencing factor relationship graph into an improved ConvGNN for training, and obtain a scalar value for measuring task reliability through a feature extraction layer, a graph convolution layer, a pooling layer, and a regression layer. Specifically, the method includes but is not limited to all of the following steps.

[0073] The embodiment of the present invention discloses a workshop manufacturing service reliability evaluation method based on a convolutional graph neural network. Specifically, referring to Figure 1 , Figure 2 The method includes:

[0074] Conduct failure mode and effect analysis on manufacturing services and extract factors affecting the reliability of manufacturing services;

[0075] Provide new connotations of manufacturing service reliability from the perspective of the workshop and quantify the service reliability value;

[0076] Based on the influencing factors, key influencing factors of manufacturing service reliability are screened out;

[0077] Based on the key influencing factors, explore the internal relationships between the key influencing factors;

[0078] Based on the relationship between the key factors and their internal relationships and service reliability, a manufacturing service reliability evaluation model is constructed;

[0079] According to the evaluation model, a key influencing factor relationship diagram is constructed;

[0080] According to the characteristics of the key influencing factor relationship graph, the graph convolutional neural network algorithm is improved;

[0081] Build a solution using an improved graph convolutional neural network algorithm;

[0082] According to the solution, a deep learning algorithm is used to obtain a predicted value of manufacturing service reliability.

[0083] It can be understood that the main purposes of the present invention are as follows:

[0084] Design a workshop-oriented manufacturing service reliability evaluation scheme based on convolutional graph neural network. Starting from the resources encapsulated by the manufacturing service, analyze the factors affecting service reliability, and predict the reliability of the manufacturing service based on the resources of the manufacturing service. According to the evaluation results, select high-reliability manufacturing services to complete the manufacturing tasks and improve the quality of the products.

[0085] This example uses a PCB production workshop as the application background to explain the invention in detail.

[0086] This example specifically discloses the step of performing failure mode and impact analysis on manufacturing services and extracting factors affecting the reliability of manufacturing services, including:

[0087] Manufacturing service is the smallest service unit completed by the collaborative efforts of production factors, and the manufacturing service is mapped to a certain production link in the workshop;

[0088] Use the 5M1E method (Man, Machine, Material, Method, Measurement, Environment) to fully describe the physical failure modes of manufacturing services;

[0089] According to the physical failure forms of manufacturing services, the causes of failure are analyzed and the factors affecting the reliability of manufacturing services are extracted.

[0090] It can be understood that manufacturing service is the smallest service unit completed by the cooperation of production factors. The reliability mechanism of workshop manufacturing service can be realized by analyzing the failure mechanism of production factors. By analyzing the production process of PCB workshop and using 5MIE analysis method, the physical failure form of manufacturing service is described and analyzed. The specific process is as follows: Figure 2As shown. It can be subdivided into six categories. They are: People: refers to situations where product quality problems are caused by insufficient technical ability of employees, operating errors or failure to strictly follow operating procedures; Machine: refers to situations where the machine fails or the precision does not meet the standard during operation, resulting in product quality problems; Material: refers to situations where the product cannot meet the processing requirements due to material reasons, causing product quality problems; Method: refers to situations where the product fails to meet the predetermined processing requirements due to improper processing technology design or improper selection of processing equipment, resulting in unqualified product quality; Environment: refers to situations where product quality is unqualified due to poor processing environment; Test: refers to situations where the problem product fails to be found in the detection link, resulting in the problem product flowing into the next process, resulting in unqualified product quality. According to the physical failure form of manufacturing services, its influencing factors are analyzed to obtain the influencing factors of manufacturing service reliability.

[0091] This example specifically discloses the steps of giving a new connotation of manufacturing service reliability from the perspective of the workshop and quantifying the service reliability value, including:

[0092] From the perspective of the workshop, manufacturing services correspond to the production process of the workshop;

[0093] The performance of the production process is usually reflected by product quality and execution efficiency. The indicators considered include product quality consistency, product qualification rate, effective operation rate, and average load rate;

[0094] The CRITIC method is used to determine the weights of the indicators, and the weighted sum of the indicators is used as the reliability label.

[0095] It is understandable that in the field of manufacturing service composition optimization, manufacturing service reliability is a basic attribute of service quality (QoS), which is described by the ratio of successful execution of service requests, reflecting the ratio of the number of times the service successfully completes tasks to the total number of requests within a certain period of time. This concept is applicable to Web services or cloud manufacturing services. However, in the workshop environment, facing the production needs of pursuing high efficiency and stability, this concept is not applicable. The traditional concept of reliability cannot distinguish between the execution efficiency and stability of manufacturing services, and it is difficult for manufacturing companies to evaluate the performance of the workshop production link in advance. In addition, the execution results of manufacturing services correspond to the products produced in the workshop, and reliable manufacturing services should also ensure product quality. Manufacturing services correspond to a certain link in the production process, so the reliability of manufacturing services can be analyzed from the perspective of the production link.

[0096] It is understandable that in the problem of comprehensive evaluation of multiple indicators, the indicator weight plays a decisive role in the evaluation results. As an objective weight method for dealing with multiple standards, the basic idea of ​​the CRITIC method is to determine the weight of the indicator by calculating the information content of the indicator through data. The information content is based on contrast intensity and conflict. Contrast intensity reflects the difference between indicators and is generally quantified by standard deviation. The greater the contrast intensity, the stronger the indicator's ability to distinguish and the higher the weight. Conflict describes the correlation between indicators. The smaller the conflict characteristics, the greater the coupling relationship between indicators and the smaller the weight is. This is very important for the quantification of reliability values ​​because there is a correlation between the selected indicators.

[0097] This example specifically discloses the step of screening out key influencing factors of manufacturing service reliability according to the influencing factors, including:

[0098] According to the influencing factors, calculate the importance of the factors and the information contribution rate of the factors;

[0099] The risk priority number (RPN) is used to evaluate the weight of the impact factor on service reliability. The importance of the i-th impact factor is RPN = S × O × D, where S is severity, which measures the severity of the impact that a failure may have on the system, user or environment. O is frequency, which measures the probability or frequency of a failure. D is detection, which measures the ability of existing control measures to detect failures before they occur. (RPN) i Represents the sum of the risk order numbers of the potential failure modes that can be caused by the i-th influencing factor;

[0100] According to the importance calculation method of the influencing factors, removing the influencing factors with lower importance;

[0101] Quantify the remaining influencing factors after removing the influencing factors with lower importance;

[0102] According to the quantified influencing factors, the coefficient of variation (CV) is used to evaluate the information contribution rate of the influencing factors to the reliability assessment. The coefficient of variation of the i-th influencing factor is Among them, σ i is the standard deviation of the i-th influencing factor, μ i is the average value of the i-th influencing factor;

[0103] According to the information contribution rate calculation method, the influencing factors with low information contribution rate are removed to obtain the key influencing factors of service reliability.

[0104] It is understandable that there are many factors that affect reliability in manufacturing services. Considering all indicators will increase the complexity of the training phase and may reduce the prediction accuracy due to interference between indicators. Therefore, it is necessary to screen the indicators and eliminate weakly related or unimportant ones.

[0105] It is understood that the risk priority number (RPN) is a risk assessment tool used in failure mode and effect analysis (FMEA). It helps companies identify, evaluate and prioritize potential failure modes in products or processes. The higher the RPN value of the failure mode caused by the influencing factor, the greater its impact on service reliability.

[0106] It is understandable that the goal of manufacturing service reliability evaluation is to assist users in selecting high-reliability manufacturing services. Among many manufacturing services, if the quantitative value of a certain influencing factor in different services is relatively stable, then the factor has limited contribution to distinguishing the reliability of different services, and its contribution rate to helping users select high-reliability manufacturing services is also small. The coefficient of variation (CV) is a statistic that measures the degree of data dispersion. Its size reflects the degree of change in the data value and can indicate the degree of contribution of the factor to the service reliability evaluation.

[0107] This example specifically discloses the step of mining the intrinsic relationship between the key influencing factors according to the key influencing factors, including:

[0108] There is a promoting or inhibiting relationship between the key influencing factors, which is the main relationship between two factors;

[0109] According to the production factors directly affected by the key influencing factors, the relationship between the corresponding production factors belongs to the secondary relationship between the key influencing factors;

[0110] The internal connections between different key influencing factors include primary relationships and secondary relationships.

[0111] It is understandable that the main relationship between influencing factors is the interactive relationship between influencing factors I: the interactive relationship is divided into promotion relationship and inhibition relationship. In the production process, when a certain indicator of a production factor performs well or poorly, it will affect its own other indicators or indicators of other production factors. For example: when employees have a high quality awareness, they will be more serious in maintaining equipment, thereby improving the maintenance status of equipment; otherwise, the maintenance status of equipment may deteriorate.

[0112] It can be understood that the secondary relationships between influencing factors include the coupling relationship between heterogeneous resources C n:The coupling relationship between heterogeneous resources is mainly reflected in people, machines, and materials. From the perspective of the manufacturing process, it can be divided into series relationship c1, parallel relationship c2, and cycle relationship c3. The series relationship means that within the manufacturing service, employees first handle the materials and then process them through the equipment. It does not involve the processing of multiple materials or the use of multiple equipment. The parallel relationship means that within the manufacturing service, employees need to handle different materials first and then process them through different equipment. The cycle relationship means that employees need to process materials multiple times and the equipment needs to be used multiple times. We combine the coupling relationship between heterogeneous resources and the interaction relationship between influencing factors to define a four-dimensional unique hot vector, R n =[c1,c2,c3,I], where the element is 1 or 0, c1=1 indicates that there is a series relationship, and c2=0 indicates that there is no parallel relationship.

[0113] It is understandable that the secondary relationship between influencing factors also includes the cooperative relationship between production factors. n :The collaborative relationship is divided into the collaborative relationship between man and machine: the time L1 of human-machine interaction and the collaborative relationship between man and material, machine and material, method and man-machine-material, environment and man-machine-material: interaction efficiency T is the interaction time, and N is the number of qualified semi-finished products or finished products within time T. In addition, in order to describe the complex relationship between production resources in the PCB manufacturing process, the collaborative relationship between production resources also considers the interference between different collaborative relationships of production resources.

[0114] This example specifically discloses the step of constructing a manufacturing service reliability evaluation model based on the key factors and their internal relationships with service reliability, including:

[0115] There is a function f:K between the key factors and their relationships and service reliability. Relationship,The key influencing factors K and their relationships S are mapped to the manufacturing service reliability R, and ω is the parameter set to be learned.

[0116] Understandably, the reliability of manufacturing services is challenged by a variety of uncertainties. Employee operational errors, unexpected equipment failures, and quality defects in raw materials will directly affect the reliability of manufacturing services. In addition, indirect factors such as production environment, processing methods, and measurement methods will also affect the reliability of services. To make matters more complicated, these factors do not exist in isolation. They are intertwined in the production process, forming an intricate interactive relationship. These factors and their relationships will affect the reliability of services, and changes in any of these factors will affect the reliability of the final service.

[0117] This example specifically discloses the step of constructing a key influencing factor relationship diagram according to the evaluation model, including:

[0118] Each of the key influencing factors is set as a node in a key influencing factor relationship graph, and the internal connections between different key influencing factors are set as edges in the key influencing factor relationship graph.

[0119] It is understandable that the construction process of the property graph is as follows Figure 3 As shown, the graph G = {V, E} is a data structure, where V = {v i |i=1,2,...,n} represents a node, n is the number of nodes; E={e ij |v i ,v j ∈V} represents the connection edge between nodes. Undirected edges are used to represent the relationship between factors affecting manufacturing service reliability, because although the interaction relationship between the factors themselves is unidirectional, the coupling relationship between heterogeneous resources and the collaborative relationship between production factors they affect are bidirectional.

[0120] This example specifically discloses the step of improving the graph convolutional neural network algorithm according to the characteristics of the key influencing factor relationship graph, including:

[0121] The spectral graph convolution is improved, and the feature vectors of the nodes and the edges are aggregated using Chebyshev convolution ChebConv, and the convolution kernel of the ChebConv is: Where Λ=2Λ / λ max -I,λ max is the largest eigenvalue of L, T i represents the i-th Chebyshev polynomial, K is the order of the Chebyshev polynomial, θ i are learnable parameters.

[0122] The convolution formula of the node's feature vector x and the Chebyshev convolution kernel is: T i (L)=2LT i-1 (L)-T i-2 (L) Where, T0(L) = 1, T1(L) = L. L = 2L / λ max -I is the modified graph Laplacian matrix.

[0123] ChebConv can recursively obtain high-order neighborhood information, improve local information perception, and be more stable when disturbed. However, it does not directly use the features in the edge;

[0124] According to the ChebConv method, feature fusion ChebConv is used to solve the problem that ChebConv does not directly use the features in the edge;

[0125] According to the feature fusion ChebConv method, an edge graph-feature fusion ChebConv with an edge graph is used to solve the problem that the traditional graph neural network structure cannot comprehensively consider the interference between the edge information in the graph.

[0126] It is understandable that ChebConv uses Chebyshev polynomials to approximate filters. It can recursively obtain high-order neighborhood information, improve local information perception, and be more stable when disturbed.

[0127] This example specifically discloses the step of building a solution using an improved graph convolutional neural network algorithm, including:

[0128] The attribute graph G a ={X,E f ,A,Y} is input into the improved graph convolutional neural network. Where X is the attribute matrix of key influencing factors of manufacturing service reliability, E f is the relationship matrix between influencing factors, and the adjacency matrix A represents the connection relationship between nodes, which is generated by judging whether the influencing factors have a promoting or inhibiting relationship. Y is the corresponding manufacturing service reliability label;

[0129] The G a Use the feature extraction layer to extract features and get Node feature matrix d v Represents the dimension of the output node features; edge feature matrix d e Represents the dimension of the output edge feature; n and m represent the number of nodes and edges in the attribute graph respectively;

[0130] The G b Use the graph convolution layer to perform convolution operations on relevant node features to obtain Represents the updated node feature matrix;

[0131] The updated node representation is input into the graph pooling layer to obtain a feature vector As the global embedding vector of the graph;

[0132] Select the global maximum pooling layer, and the input is the residual of the node feature matrix

[0133] The feature vector Input to the graph regression layer to obtain a scalar value that measures the reliability of the task;

[0134] The regression layer is implemented by a fully connected neural network.

[0135] Understandably, Figure 4 The construction process of the manufacturing service reliability evaluation model diagram data is shown: production monitoring data and product or semi-finished product quality data are used to evaluate the task reliability value Y of the manufacturing service; the influencing factor data of the manufacturing service production factors are obtained from the enterprise service platform and used to construct the manufacturing service reliability attribute matrix X; the interaction relationship (promotion or inhibition) between the influencing factors is used to construct the adjacency matrix A that describes the connection structure of the influencing factors; after determining the connection relationship between the influencing factors, combined with the actual production process, the production factors associated with the adjacent factors are examined to determine the coupling and collaboration relationship between the production factors, which is used to construct E f .

[0136] It can be understood that the reliability evaluation network (REN) consists of four modules: feature extraction module, graph convolution module, global graph pooling module and regression module. The input of REN is the attribute graph of the key factors of manufacturing service reliability, and the output is a scalar representing the reliability of manufacturing service. The detailed structure of REN is as follows Figure 5 shown.

[0137] It can be understood that the feature extraction module encodes the attribute vectors of nodes and edges into deep feature vectors, with the aim of mining the relationship between individual manufacturing service QoS attributes, controlling the dimension of data output and obtaining a richer representation. The feature extraction targets the node attribute matrix X and edge attribute matrix E of the manufacturing service reliability evaluation model. f First, we use ChebConv and Conv1D based on edge graph to extract E f ChebConv is used to mine the influence relationship between different edge information, and Conv1D is used to enrich the information features of the edge. The feature information extracted by the two is combined to obtain Then use Conv1D to extract the features of X.

[0138] It is understandable that the core idea of ​​graph convolution is to update the representation of nodes based on node features, edge features, and node connection relationships, while keeping the structure of the graph unchanged. Figure 5 As shown in the figure, the order of the Chebyshev polynomial is set to 2[]. The feature aggregation mechanism effectively combines edge features, which is conducive to extracting deep structural features from the graph. In order to improve the convergence speed and generalization ability of the model, each graph convolution layer is configured with a batch normalization layer and a ReLu activation function. In addition, to prevent gradient vanishing or gradient explosion, the graph convolution module also adopts a residual connection learning strategy, which helps to alleviate the over-smoothing effect caused by stacking multiple layers of graph convolution and improve the expressiveness of the model.

[0139] It is understandable that the graph convolution layer only learns to update the representation of nodes. For graph-level tasks, a pooling layer needs to be introduced to achieve the transition from node-level output to graph-level output. The present invention uses the global maximum pooling layer because it is simple to implement and is often used in a variety of popular GNN architectures. The input of the graph pooling module is the residual of the node feature matrix Global max pooling finds the maximum value in each dimension of the node feature and then forms a feature vector And used as the global embedding vector of the graph.

[0140] It can be understood that after the graph pooling module, the regression module is designed to evaluate the reliability of the target manufacturing service. The input of this module is the graph embedding vector of the manufacturing service reliability evaluation model The output is a scalar value that measures the reliability of the task. This module is implemented by a fully connected neural network with 2 hidden layers, corresponding to 128 and 64 neurons respectively. Batch normalization strategy is adopted in each layer, and Dropout=0.2 is set to prevent overfitting of the model. ReLu is used as the activation function to ensure that the output is not negative.

[0141] This example specifically discloses the step of using feature fusion ChebConv to solve the problem that ChebConv does not directly use the features in the edge, including:

[0142] Before the convolution operation, the edge features and neighbor node features are fused to obtain the node features containing edge information, thus realizing the direct use of edge features.

[0143] It is understandable that the edge information and node information need to be converted in dimension after feature extraction, and the dimension of the converted edge feature needs to be the same as the dimension of the node feature.

[0144] This example specifically discloses the step of using ChebConv with edge graph-feature fusion to solve the problem that the traditional graph neural network structure cannot comprehensively consider the interference between the edge information in the graph, including:

[0145] Reasons for adding edge graph module:

[0146] According to the internal relationships between the key influencing factors, the interference between different internal relationships should also be considered. For example, when the interaction between humans and machines shows a positive or negative trend, this will indirectly affect other interactions related to humans or machines;

[0147] The limitation of graph neural networks is that their edges can only reflect the direct connection between two nodes. When the information of an edge changes, it cannot have a direct impact on other edges in the network.

[0148] Specific steps:

[0149] Constructing an edge graph of the key influencing factor relationship graph according to the key influencing factor relationship graph;

[0150] Among them, the node information of the edge graph is the edge information in the original graph, and the edge represents the interference relationship between the edge information of the original graph;

[0151] The edge graph is used to extract the edge features in the original image, and then it is fused with the edge features of the original image.

[0152] It is understandable that the construction process of the edge graph is as follows Figure 5 As shown in the figure, the edge graph aggregates the information of each edge in the original graph, which can be used to extract the features of the edges in the original graph and explore the relationship between the edges.

[0153] It can be understood that the edge graph is used to extract the edge features in the original image to explore the influence relationship between the edges, and then it is fused with the edge features of the original image. The specific process is as follows Figure 6 As shown in the figure. After adding the edge graph feature, the edges in the original graph not only contain the information of the original edges, but also cover the relationship between the edges. Any change in the information in any edge will affect other edges.

[0154] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements and corrections to this specification. Such modifications, improvements and corrections are suggested in this specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of this specification.

[0155] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A workshop manufacturing service reliability assessment method based on convolutional graph neural network, characterized in that: include: Conduct failure mode and effect analysis on manufacturing services and extract factors affecting the reliability of manufacturing services; Provide new connotations of manufacturing service reliability from the perspective of the workshop and quantify the service reliability value; Based on the influencing factors, key influencing factors of manufacturing service reliability are screened out; Based on the key influencing factors, explore the internal relationships between the key influencing factors; Based on the relationship between the key factors and their internal relationships and service reliability, a manufacturing service reliability evaluation model is constructed; According to the evaluation model, a key influencing factor relationship diagram is constructed; According to the characteristics of the key influencing factor relationship graph, the graph convolutional neural network algorithm is improved; Build a solution using an improved graph convolutional neural network algorithm; According to the solution, a deep learning algorithm is used to obtain a predicted value of manufacturing service reliability.

2. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: The step of performing failure mode and effect analysis on manufacturing services and extracting factors affecting the reliability of manufacturing services includes: Manufacturing service is the smallest service unit completed by the collaboration of production factors, and the manufacturing service corresponds to a certain production link in the workshop; Use the 5M1E method (Man, Machine, Material, Method, Measurement, Environment) to fully describe the physical failure modes of manufacturing services; According to the physical failure forms of manufacturing services, the causes of failure are analyzed and the factors affecting the reliability of manufacturing services are extracted.

3. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: The step of giving a new connotation to manufacturing service reliability from the perspective of the workshop and quantifying the service reliability value includes: From the perspective of the workshop, manufacturing services correspond to the production process of the workshop; The performance of the production process is usually reflected by product quality and execution efficiency. The indicators considered include product quality consistency, product qualification rate, effective operation rate, and average load rate. The CRITIC method is used to determine the weights of the indicators, and the weighted sum of the indicators is used as the reliability label.

4. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: Based on the influencing factors, the step of screening out the key influencing factors of manufacturing service reliability includes: According to the influencing factors, calculate the importance of the factors and the information contribution rate of the factors; The risk priority number (RPN) is used to evaluate the weight of the impact factor on service reliability. The importance of the i-th impact factor is RPN = S × O × D, where S is severity, which measures the severity of the impact that a failure may have on the system, user or environment; O is frequency, which measures the probability or frequency of a failure; and D is detection, which measures the ability of existing control measures to detect failures before they occur. (RPN) i Represents the sum of the risk order numbers of the potential failure modes that can be caused by the i-th influencing factor; According to the importance calculation method of the influencing factors, removing the influencing factors with lower importance; Quantify the remaining influencing factors after removing the influencing factors with lower importance; According to the quantified influencing factors, the coefficient of variation (CV) is used to evaluate the information contribution rate of the influencing factors to the reliability evaluation. The coefficient of variation of the i-th influencing factor is Among them, σ i is the standard deviation of the i-th influencing factor, μ i is the average value of the i-th influencing factor; According to the information contribution rate calculation method, the influencing factors with low information contribution rate are removed to obtain the key influencing factors of service reliability.

5. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: Based on the key influencing factors, the step of exploring the internal relationship between the key influencing factors includes: There is a promoting or inhibiting relationship between the key influencing factors, which is the main relationship between two factors; According to the production factors directly affected by the key influencing factors, the relationship between the corresponding production factors belongs to the secondary relationship between the key influencing factors; The internal connections between different key influencing factors include primary relationships and secondary relationships.

6. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: According to the relationship between the key factors and their internal relationships and service reliability, the step of constructing a manufacturing service reliability evaluation model includes: There is a function between the key factors and their relationship and service reliability Relationship,The key influencing factors K and their relationships S are mapped to the manufacturing service reliability R, and ω is the parameter set to be learned.

7. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: According to the evaluation model, the step of constructing a key influencing factor relationship diagram includes: Each of the key influencing factors is set as a node in a key influencing factor relationship graph, and the internal connections between different key influencing factors are set as edges in the key influencing factor relationship graph.

8. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: According to the characteristics of the key influencing factor relationship graph, the step of improving the graph convolutional neural network algorithm includes: The spectral graph convolution is improved, and the feature vectors of the nodes and the edges are aggregated using Chebyshev convolution ChebConv, and the convolution kernel of the ChebConv is: Where Λ=2Λ / λ max -I,λ max is the largest eigenvalue of L, T i represents the i-th Chebyshev polynomial, K is the order of the Chebyshev polynomial, θ i is a learnable parameter; The convolution formula of the node's feature vector x and the Chebyshev convolution kernel is: T i (L)=2LT i-1 (L)-T i-2 (L) Where, T0(L) = 1, T1(L) = L. L = 2L / λ max -I is the modified graph Laplacian matrix; ChebConv can recursively obtain high-order neighborhood information, improve local information perception, and be more stable when disturbed. However, it does not directly use the features in the edge; According to the ChebConv method, feature fusion ChebConv is used to solve the problem that ChebConv does not directly use the features in the edge; According to the feature fusion ChebConv method, an edge graph-feature fusion ChebConv with an edge graph is used to solve the problem that the traditional graph neural network structure cannot comprehensively consider the interference between the edge information in the graph.

9. According to claim 1, a workshop manufacturing service reliability assessment method based on convolutional graph neural network is characterized in that: The step of building a solution using an improved graph convolutional neural network algorithm includes: The attribute graph G a ={X,E f ,A,Y} is input into the improved graph convolutional neural network, where X is the attribute matrix of key influencing factors of manufacturing service reliability, E f is the relationship matrix between influencing factors, the adjacency matrix A represents the connection relationship between nodes, and is generated by judging whether the influencing factors have a promoting or inhibiting relationship. Y is the corresponding manufacturing service reliability label; The G a Use the feature extraction layer to extract features and get Node feature matrix d v Represents the dimension of the output node features; edge feature matrix d e Represents the dimension of the output edge feature; n and m represent the number of nodes and edges in the attribute graph respectively; The G b Use the graph convolution layer to perform convolution operations on relevant node features to obtain Represents the updated node feature matrix; The updated node representation is input into the graph pooling layer to obtain a feature vector As the global embedding vector of the graph; Select the global maximum pooling layer, and the input is the residual of the node feature matrix The feature vector Input to the graph regression layer to obtain a scalar value that measures the reliability of the task; The regression layer is implemented by a fully connected neural network.

10. The method for evaluating workshop manufacturing service reliability based on convolutional graph neural network according to claim 8, characterized in that: The step of using feature fusion ChebConv to solve the problem that ChebConv does not directly use the features in the edge includes: Before the convolution operation, the edge features and neighbor node features are fused to obtain the node features containing edge information, thus realizing the direct use of edge features.

11. The method for evaluating workshop manufacturing service reliability based on convolutional graph neural network according to claim 8, characterized in that: We use ChebConv with edge graph-feature fusion to solve the problem that the traditional graph neural network structure cannot comprehensively consider the interference between the edge information in the graph. This step includes: Reasons for adding edge graph module: According to the internal relationships between the key influencing factors, the interference between different internal relationships should also be considered. For example, when the interaction between humans and machines shows a positive or negative trend, this will indirectly affect other interactions related to humans or machines. The limitation of graph neural networks is that their edges can only reflect the direct connection between two nodes. When the information of an edge changes, it cannot have a direct impact on other edges in the network. Specific steps: Constructing an edge graph of the key influencing factor relationship graph according to the key influencing factor relationship graph; Among them, the node information of the edge graph is the edge information in the original graph, and the edge represents the interference relationship between the edge information of the original graph; The edge graph is used to extract the features of the edges in the original image, and then it is fused with the edge features of the original image. After adding the edge graph features, any change in the information of any edge in the original image will affect other edges.

Images