Product assembly abnormity process parameter tracing method and device

Abstract

The invention discloses a product assembly abnormal process parameter tracing method and device, and relates to the technical field of abnormal parameter tracing, the method takes a knowledge graph as a core technical means, and adopts a knowledge-driven ammunition assembly process quality control model to guide ammunition assembly process quality control. According to the method, the Bayesian network model and the knowledge traceability strategy are fused to help deeply identify and understand the root cause of the assembly quality problem, and powerful support is provided for systematic solution of the problem. The method can comprehensively evaluate the influence capability of process technological parameters on quality parameters, intelligently evaluate the quality defect category and predict the assembly quality consistency, and realizes the quality control of the ammunition assembly process.

Description

Technical Field

[0001] This invention relates to the field of abnormal parameter tracing technology, and in particular to a method and apparatus for tracing abnormal process parameters in product assembly. Background Technology

[0002] The assembly quality of ammunition products directly affects product performance, reliability, safety and other technical and tactical indicators. Because the ammunition assembly process involves multiple processes, multiple parameters, and quality transfer across multiple workshops, it becomes complicated to trace abnormal parameters when quality problems occur.

[0003] The extensive application of automation and intelligent technologies in ammunition production has made it possible to monitor production process parameters in real time. However, it still cannot solve the complex situation of quality anomalies caused by multiple types of parameters. Therefore, traditional methods do not consider the correlation between process parameters, resulting in problems such as low accuracy of assembly process parameter quality positioning.

[0004] It is evident that traditional process parameter tracing is based on locating abnormal process parameters by identifying quality problems. This involves manually inferring which process steps or parameters might have caused the quality issues, and then reviewing the abnormal process parameters that could have led to the quality problems. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method and apparatus for tracing the source of abnormal process parameters in product assembly to overcome or at least partially solve the above problems.

[0006] This invention provides the following solution: A method for tracing the source of abnormal process parameters in product assembly includes: An assembly quality knowledge graph is constructed using the collected product assembly quality-related knowledge; the product assembly quality-related knowledge includes quality parameters, process parameters, and entity relationship data. The nodes and relationships in the assembly quality knowledge graph are mapped to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are the leaf nodes of the Bayesian network model. The nodes in the Bayesian network model are assigned prior probabilities and conditional probabilities based on fuzzy probability theory; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. Based on the Bayesian network model, a reinforcement learning strategy is used to dynamically infer and trace the abnormal process parameters of product assembly, and a traceability report is generated.

[0007] Preferably: a triangular fuzzy number is used to represent the fuzzy probability subset of the root node experiencing quality problems; the fuzzy probability subset of the probability of the root node assembly quality having problems in the Bayesian network model is represented by the following formula: In the formula: This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node appearing.

[0008] Preferably, the OR gate algorithm in the fault tree is used to obtain the fuzzy probabilities of intermediate nodes and leaf nodes as shown in the following formula: In the formula: Indicates the number of nodes. Represents the knowledge graph of the first A fuzzy probability subset of root nodes experiencing quality problems This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node appearing.

[0009] Preferably, the fuzzy probability is used as the posterior probability of the Bayesian network. Given the occurrence of a leaf node T, the probability of the root node occurring is expressed by the following formula: In the formula: It is the prior probability of the root node. When the root node is known The probability of leaf node T occurring when it occurs.

[0010] Preferably, the dynamic reasoning and tracing further includes a key parameter identification step: Calculate the root node For leaf nodes The probability and critical importance of being in an abnormal state are determined to quantify the impact of each root node on the abnormal result and identify the key process parameters that lead to assembly abnormalities.

[0011] Preferably: root node The state is The probability formula is as follows: in, It is the root node The conditional probability, The value is 0 when there are no assembly quality issues. The value is 1 when there are assembly quality issues; leaf nodes In state The probabilities are as follows: Where: leaf nodes do not occur When a leaf node occurs, the value is 0. =1, This represents the nth root node in the set of root nodes; When the root node The state is When, make the leaf node T state is The conditional probability is given by the following formula: When the root node The state is When, the leaf node T is in only state The probabilistic importance is given by the following formula: For the root node The root node is obtained by averaging the incremental contributions across all states. For leaf nodes The overall probabilistic importance is given by the following formula: In the formula, K represents the number of assembly quality problems removed from the root node; When the root node In In the state, leaf node T alone is The critical importance of a state is as follows: root node For the state is The critical importance of the leaf node T is as follows: .

[0012] Preferably, the Bayesian network model includes an improved Q-function, the equation of which is as follows: In the formula: Indicates a reward. Indicates state, Indicates the next state. Indicates the reasoning path, Representation strategy, Actions that indicate reasoning Indicates the action of the next reasoning. This represents an action-state value function.

[0013] Preferably, the Bayesian network model uses non-policy methods to find the optimal policy, and the non-policy methods include optimization policies and execution policies.

[0014] Preferably, the optimization strategy is expressed by the following formula: The execution strategy is represented by the following formula, which is the optimal tracing action that allows the Q function to reach its maximum value under the current process state s: In the formula: Represents the optimal action-state value function. This represents the state-action value function within the Bayesian framework. This represents the action space corresponding to state s.

[0015] A device for tracing abnormal process parameters in product assembly, used to execute the above-described method for tracing abnormal process parameters in product assembly, the device comprising: The knowledge graph construction unit is used to construct an assembly quality knowledge graph using the collected product assembly quality-related knowledge; the product assembly quality-related knowledge includes quality parameters, process parameters, and entity relationship data. A Bayesian network mapping unit is used to map the nodes and relationships in the assembly quality knowledge graph to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are used as the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are used as the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are used as the leaf nodes of the Bayesian network model. The probability assignment unit is used to assign prior probabilities and conditional probabilities based on fuzzy probability theory to the nodes in the Bayesian network model; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. The traceability report generation unit is used to perform dynamic reasoning and traceability of abnormal process parameters in product assembly based on the Bayesian network model and combined with reinforcement learning strategies, and to generate a traceability report.

[0016] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This application provides a method and apparatus for tracing the source of abnormal process parameters in product assembly. This method uses knowledge graphs as its core technology and employs a knowledge-driven quality control model for ammunition assembly processes to guide quality control during assembly. The method integrates Bayesian network models with knowledge tracing strategies to help deeply identify and understand the root causes of assembly quality problems, providing strong support for systematic problem-solving. It can comprehensively assess the impact of process parameters on quality parameters, intelligently evaluate the types of quality defects, and predict assembly quality consistency, thereby achieving quality control over the ammunition assembly process.

[0017] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 This is a flowchart of a method for tracing the source of abnormal process parameters in product assembly, provided by an embodiment of the present invention. Figure 2 This is a schematic diagram of a product assembly anomaly process parameter tracing device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a product assembly abnormality process parameter traceability device provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0021] See Figure 1 This invention provides a method for tracing abnormal process parameters in product assembly, such as... Figure 1 As shown, the method may include: S101: Construct an assembly quality knowledge graph using the collected product assembly quality-related knowledge; the product assembly quality-related knowledge includes quality parameters, process parameters, and entity relationship data. S102: Map the nodes and relationships in the assembly quality knowledge graph to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are the leaf nodes of the Bayesian network model. S103: Assign prior probabilities and conditional probabilities based on fuzzy probability theory to the nodes in the Bayesian network model; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. S104: Based on the Bayesian network model, combined with reinforcement learning strategies, perform dynamic reasoning and source tracing of abnormal process parameters in product assembly, and generate a source tracing report.

[0022] Furthermore, embodiments of this application can provide a fuzzy probability subset representing the root node's quality problem using triangular fuzzy numbers; the fuzzy subset of the probability of the root node's assembly quality being problematic in the Bayesian network model is represented by the following formula: In the formula: This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node appearing.

[0023] Using the OR gate algorithm in the fault tree, the fuzzy probabilities of intermediate nodes and leaf nodes are obtained as shown in the following formula: In the formula: Indicates the number of nodes. Represents the knowledge graph of the first A fuzzy probability subset of root nodes (corresponding to ammunition assembly process parameters) where quality problems occur. This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node appearing.

[0024] Using the fuzzy probability as the posterior probability of the Bayesian network, the probability of the root node appearing when leaf node T appears is expressed by the following formula: In the formula: It is the prior probability of the root node. When the root node is known The probability of leaf node T occurring when it occurs.

[0025] The Bayesian network model uses the saliency of probabilities and the saliency of keys to comprehensively compare and identify the main events causing ammunition assembly quality problems. In its specific implementation, the dynamic reasoning and tracing also includes a key parameter identification step. Calculate the root node For leaf nodes The probability and critical importance of being in an abnormal state are determined to quantify the impact of each root node on the abnormal result and identify the key process parameters that lead to assembly abnormalities.

[0026] root node The state is The probability formula is as follows: in, It is the root node The conditional probability, The value is 0 when there are no assembly quality issues. The value is 1 when there are assembly quality issues; leaf nodes In state The probabilities are as follows: Where: leaf nodes do not occur When a leaf node occurs, the value is 0. =1, Represents the first in the root node set One root node; When the root node The state is When, make the leaf node T state is The conditional probability is given by the following formula: When the root node The state is When, the leaf node T is in only state The probabilistic importance is given by the following formula: For the root node The root node is obtained by averaging the incremental contributions across all states. For leaf nodes The overall probabilistic importance is given by the following formula: In the formula, This indicates the number of assembly quality issues removed from the root node; When the root node In In the state, leaf node T alone is The critical importance of a state is as follows: root node For the state is The critical importance of the leaf node T is as follows: .

[0027] The Bayesian network model includes an improved Q-function, the equation of which is as follows: In the formula: Indicates a reward. Indicates state, Indicates the next state. Indicates the reasoning path, Representation strategy, Actions that indicate reasoning Indicates the action of the next reasoning. This represents an action-state value function.

[0028] The Bayesian network model uses non-policy methods to find the optimal policy, which includes optimization policies and execution policies.

[0029] The optimization strategy is expressed by the following formula: The execution strategy is represented by the following formula: .

[0030] In the formula: Represents the optimal action-state value function. This represents the state-action value function within the Bayesian framework. This represents the action space corresponding to state s.

[0031] The following is a detailed description of the product assembly anomaly process parameter tracing method provided in the embodiments of this application.

[0032] Knowledge graphs can be likened to Bayesian networks, where path decision-making between nodes can be viewed as reasoning from the root node to intermediate nodes and from intermediate nodes to leaf nodes. By mapping the causal relationship between assembly quality and process parameters in the production process, the assembly quality knowledge graph is transformed into a Bayesian network and nodes in a one-to-one correspondence knowledge graph. The specific process of mapping the knowledge graph to a Bayesian network includes three steps: First, the nodes in the knowledge graph are represented as nodes in a Bayesian network. The node representing the assembly quality problem in the knowledge graph is the root node of the Bayesian network, the relationship between assembly quality and process quality parameters is the intermediate node, and the node representing assembly process quality parameters is the leaf node. Second, the fuzzy probability of each assembly quality node in the knowledge graph is used as the prior probability of the corresponding root node in the Bayesian network. Finally, the logical relationships between the nodes in the knowledge graph are represented.

[0033] 1. Construct an assembly quality knowledge graph. The relevant definitions for the assembly quality knowledge graph are as follows: ① Entity: Refers to an independent entity or event involved in the ammunition assembly process, such as the shell numbered X, or the No. 5 propellant grain; ② Entity Class: A collection of entities with similar characteristics, such as parts, equipment, or processes; ③ Relationship: A connection between entities, used to form a triple structure of entity-relationship-entity, such as shell numbered X - pressing - No. 5 propellant grain; ④ Relationship Class: An abstract description of entities connected by a specific relationship, facilitating relationship reuse, such as the entity roles of "extraction equipment" and "extraction object" connected by the "pressing" relationship; ⑤ Attribute: Refers to the characteristics used to describe each entity or relationship, such as process parameters and quality parameters; ⑥ Attribute Value: Refers to the specific value of a defined attribute of an entity or relationship, such as the propellant grain assembly pressing force being 2000N.

[0034] Based on the above definition, the construction steps of the assembly quality knowledge graph are as follows: First, establish a data model that conforms to the characteristics of the ammunition assembly process and effectively integrate multi-source heterogeneous data generated during the assembly process; second, extract assembly process triples by combining grey relational analysis with expert experience knowledge; finally, perform entity alignment using the TransE algorithm to form a quality graph for quality control of the ammunition assembly process.

[0035] 2. Construct a Bayesian conditional model of assembly quality knowledge graph.

[0036] The probability of node failure is described by using fuzzy subsets represented by triangular fuzzy numbers. Let U represent a set of objects, and define the fuzzy number as the membership function for that set of objects, as shown below: The membership function of the triangular fuzzy number is as follows: In the Bayesian network of the ammunition assembly quality knowledge graph, the leaf nodes are... The root node is and satisfy In the 22 root nodes, the ammunition assembly quality is divided into two forms: with quality problems and without quality problems. Using fuzzy numbers 0 and 1, the Bayesian network node states are [0,1]. The fuzzy subset of the probability of assembly quality problems at the root nodes is then given by the following formula: in, yes The upper and lower bounds of confidence. It is the minimum probability of occurrence. It is the maximum probability of occurrence. It is the median probability of the root node appearing, and satisfies... .

[0037] An event will occur if one of the conditions is met. In Bayesian networks, the relationships between the root node and intermediate nodes, and between intermediate nodes and leaf nodes, can be viewed as logical relationships represented by OR gates. Therefore, the OR gate algorithm from fault trees is introduced, yielding the fuzzy probabilities of intermediate nodes and leaf nodes as follows: In the formula: Indicates the number of nodes. Representing the knowledge graph of the first A fuzzy probability subset of root nodes experiencing quality problems This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node occurring.

[0038] Bayesian network analysis uses known leaf node occurrence probabilities (prior probabilities) and the backward reasoning capability of BN networks to correct these prior probabilities, yielding the prior probabilities of intermediate and root nodes in the Bayesian network. Therefore, using these fuzzy probabilities as the posterior probabilities of the Bayesian network, the probability of the root node occurring (posterior probability) given the occurrence of leaf node T is as follows: in, It is the prior probability of the root node. When the root node is known When an event occurs, the probability of leaf node T occurring.

[0039] Simply considering the posterior probability of the root node to guide ammunition assembly quality is not entirely reliable. Therefore, the significance of the probability and the significance of the key are introduced. Through comprehensive comparison, the main events of the ammunition assembly quality problem are obtained.

[0040] root node The state is (The value is 0 when there are no assembly quality issues, and 1 when there are assembly quality issues), where the probability formula is as follows: in, It is the root node The conditional probability.

[0041] leaf nodes In state The probability (when leaf nodes do not occur) When a leaf node occurs, the value is 0. The probability of being 1 is as follows: When the root node The state is When, make the leaf node T state is The conditional probability is given by the following formula: When the root node The state is When, the leaf node T is in only state The probabilistic importance is given by the following formula: For the root node The root node is obtained by averaging the incremental contributions across all states. For leaf nodes The overall probabilistic importance is given by the following formula: In the formula, K represents the number of assembly quality problems removed from the root node.

[0042] When the root node In In the state, leaf node T alone is Key importance of state As shown in the following formula: root node For the state is The critical importance of leaf node T as follows: 3. Construct a Bayesian correlation network for key assembly quality parameters.

[0043] Bayesian correlation networks for key parameters of ammunition assembly quality are used to deeply analyze the probabilistic relationships between key parameters during ammunition manufacturing or assembly. Their main objective is to achieve continuous improvement in assembly quality, identify core influencing parameters, and understand their impact. By establishing a Bayesian correlation network, each key parameter in the assembly process is structurally modeled, considering the complex dependencies between parameters. This network considers not only direct relationships between parameters but also potential implicit relationships, thus providing more comprehensive and accurate analytical results.

[0044] Bayesian networks are used to build models using probability distributions over a set of variables. The advantage of Bayesian networks is their compact representation of complex problem domains. Furthermore, Bayesian networks offer decision-making, smoothness, consistency, and flexible applicability in complex domains. On the other hand, association rules are based on antecedent and consequent components, each with conditional and decision attributes, respectively. Each association rule has a confidence percentage value, probabilistically indicating the rule's effectiveness. Therefore, association rules and Bayesian networks can be linked due to probabilistic methods. Thus, focusing on association rules, a Bayesian association network for key assembly quality parameters is developed to explore the relationship between key process quality parameters in ammunition assembly and assembly quality.

[0045] The Bayesian correlation network of ammunition assembly quality is just a set of variables. It is also a causal probability network, which compactly represents the joint probability distribution over these variables. This representation consists of a set of local conditional probability distributions, combined with a set of conditionally independent assertions, allowing the construction of a global joint distribution from the local distributions. The decomposition is based on a probability-based chain rule, as shown in the following formula: For each ammunition's key process quality parameters ,let make and A set of conditionally independent variables, namely: Given these sets, the Bayesian association network can be described as a directed acyclic graph such that each ammunition assembly critical process quality parameter... Corresponding to a node in the graph, and The parent node of the corresponding node corresponds to The variables in.

[0046] With each node The associated conditional probability distribution is , Each instance has a distribution. By comparing the above formula, we can find that... Any Bayesian network uniquely determines the joint probability distribution of these variables, that is: When K=5, the above formula becomes: It can be seen that using Bayesian association networks greatly simplifies the calculation of joint probabilities of variables. All variables are Boolean variables, the number of variables is K, and the required number of independent probabilities is... For example, when K equals 5, the number of independent probabilities is... However, the number of independent probabilities calculated using the above formula is 10. Clearly, the more the K value increases, the more the required probability decreases, and the easier the problem becomes to solve. This is very helpful in solving multiple key process quality parameters for ammunition assembly. Integrating the above knowledge, we can obtain a Bayesian association network for key ammunition assembly quality parameters and a Bayesian conditional model for an ammunition assembly quality knowledge graph. Combining these two methods enables the tracing of assembly quality knowledge through Bayesian association reasoning.

[0047] 4. Origin tracing of assembly quality problems based on Bayesian association reasoning.

[0048] Assembly quality knowledge tracing based on Bayesian association reasoning is a highly intelligent method used to deeply analyze the essential causes of assembly quality problems, tracing the problems back to the key parameters, conditions, or events that led to their occurrence. The core idea of ​​this method is to integrate Bayesian network models with knowledge tracing strategies to help deeply identify and understand the root causes of assembly quality problems, providing strong support for the systematic solution of these problems.

[0049] Describing knowledge graph tracing as a tuple Among them, the tuples Representing a state, also known as a knowledge triple. , , For the source entity, Indicate the query target; Indicates the current entity Output the set of edges; and the transition The state transition probability is a probability distribution used to identify the next state, and it is defined as follows: For trajectory Every step inside Set the reward function to ,in yes The terminal state. In other words, if the agent reaches the correct target entity, it will receive a reward of 1, otherwise 0.

[0050] As mentioned earlier, a single distribution may be insufficient to simulate the uncertainty of entities and relations in a knowledge graph. Therefore, a Gaussian distribution is used to represent an entity and relation: in, The average vector, Let be the covariance matrix (currently, diagonal covariance is computationally efficient). Clearly, a state or action defined above also follows the joint distribution function as follows: In the formula: Representing state The joint probability distribution, Representing the current entity Source entity Query target The probability distribution, Representing the current entity Integral differential element, source entity Integral differential element, query target Integral and differential elements are used to calculate the state. The joint distribution Indicates action The joint probability distribution, Let represent the probability distributions of the entities involved in the action and the relationships involved in the action, respectively. Representing entities in the action Integral and differential elements, relationships in action The integral and differential elements are used to calculate the joint distribution of actions.

[0051] As an iterative interaction with the environment, the agent, following an unknown state-action distribution, extends its path from the source entity to the target entity. With training, the posterior values ​​of these Gaussian distributions will begin to converge, uncertainty may decrease, and thus the agent will become more deterministic. Therefore, the inference path under policy π can be... The rationale is described as follows: in, It obeys assumption 1 and its length is less than or equal to , This is a typical Bayesian inference about Markov chains, relying on It can be observed that The uncertainty will be passed on to This leads to uncertainty in the predictive reasoning path. Indicating the reasoning path initial state The probability distribution.

[0052] Using reinforcement learning to train a behavior similar to The agent is evaluated by a state-value function, the Q-function, for each observation. This represents the Q-function. Therefore, the corresponding Bellman equation in a strategy... Below, from the state Begin, and in the action space China takes action The formula is as follows: In the formula: Representation Strategy The expected operator below, Representing state The corresponding reward function value, This represents the conversion factor, which ranges from [0,1].

[0053] Given the current state and action, the environment has a probability of Next, randomly enter the subsequent state. and provide rewards Given the policy π and the Markov assumptions of the model, the equation for the Q function can be rewritten as follows: Due to state-action pairs in a knowledge graph environment The combination space is large, making it difficult to directly derive the Q function from the equation. To approximate the Q function, a Bayesian neural network is used for modification, where the weights follow a prior distribution. First, the current state is processed using a Bayesian LSTM. Encode as a latent vector The announcement is as follows: Bayesian LSTM is trainable for random variables, allowing input probability distributions and outputting probability distributions. Then, a Bayesian linear regression layer is used. To learn the Q value for each action: in, , This indicates the relationship corresponding to action 'a'. This indicates the next entity to which action a is transferred after it is performed.

[0054] Existing RL-based methods typically use reinforcement to optimize the objective function. However, reinforcement encourages extending the next state with high rewards, thus biasing towards searching for the optimal policy and leading to unstable training with high variance. To address this issue, a non-policy is employed to find the optimal policy. The non-policy is divided into an optimization policy and an execution policy. The definition of the optimization policy is: Due to any... Greediness is optimal, and the optimization strategy is a greedy strategy that guarantees exploitation. Then, the optimal state-action value function of an action can be written using the optimal values ​​of its subsequent states as: In the formula: Represents the optimal action-state value function. This represents the state-action value function within the Bayesian framework. This represents the action space corresponding to state s.

[0055] It is worth noting that the conversion core was not known in advance. Therefore, an execution strategy is used to determine the next action for the next state. At each step, weights are drawn from the posterior distribution. Then the agent executes the action with the maximum reward: It guarantees uncertainty through posterior sampling and allows agents with high uncertainty to explore efficiently. To estimate the posterior distribution... Applying Bayesian variational learning to find distributions of The distribution exhibits the smallest Coulomb-Leibler divergence over the parameters: In the formula, for The prior distribution of D. Minimizing this is called variational free energy, which is equivalent to maximizing the log-likelihood of the given training data D. However, it is subject to KL penalty, which acts as a regularizer. Indicates in the parameter Minimize the variational distribution within the range With prior distribution The divergence; Indicates in the parameter Minimize the log-likelihood within the range ; Indicates parameters Follows variational distribution When, the expected value of the log-likelihood of the data.

[0056] Because the prior distribution of the weights is uncertain and follows a Gaussian distribution, therefore It can be decomposed into the sum of the average value and Gaussian noise: in, It is zero-mean Gaussian noise. Let... To predict the distribution. Therefore, the goal of step t is: By approximating the objective using Bayesian methods, the probability distribution on the weights of the neural network is learned, as shown in the following formula: Among them, subscript Indicates the first One sample, Represents variational distribution With prior distribution KL divergence; Indicates parameters Follows variational distribution The expectation operator at time; This represents the actual observed value of the Q function; Indicates the state ,action Model parameters Below, the target value predicted by the Q function can be estimated. By repeating from the variational posterior Monte Carlo samples extracted from [the sample].

[0057] First, all Gaussian priors for entities and relations are randomly initialized according to the prior distribution pre-trained by the probability model. These are parameters of the network architecture. These are parameters of the neural network architecture. An embedding layer representing entities and relationships. Because These are different types of geometric objects, so the average value should not be allowed to grow too large. Therefore, when estimating... When applying the following hard constraints: In the formula: Representing entity relationships in an assembly quality knowledge graph The embedded mean vector.

[0058] To ensure that the covariance matrix is ​​positive definite and of a reasonable size, the diagonal covariance matrix is ​​constrained to a hypercube. Within, that is: .

[0059] In the formula: This indicates that for all entity relations , Represents the set of entities in a knowledge graph. Represents the set of relations in a knowledge graph. This represents the lower bound constraint of the covariance matrix. Representing entity relations in a knowledge graph The embedding covariance matrix, This represents the upper bound constraint of the covariance matrix.

[0060] In summary, the product assembly anomaly process parameter tracing method provided in this application utilizes knowledge graphs as its core technology and employs a knowledge-driven ammunition assembly process quality control model to guide ammunition assembly process quality control. This method integrates Bayesian network models with knowledge tracing strategies to help deeply identify and understand the root causes of assembly quality problems, providing strong support for systematic problem-solving. It can comprehensively assess the impact of process parameters on quality parameters, intelligently evaluate the types of quality defects and predict assembly quality consistency, thereby achieving quality control over the ammunition assembly process.

[0061] See Figure 2 This application embodiment can also provide a device for tracing abnormal process parameters in product assembly, such as... Figure 2 As shown, the apparatus used to perform the above-described method for tracing the abnormal process parameters in product assembly may include: The knowledge graph construction unit 201 is used to construct an assembly quality knowledge graph using the collected product assembly quality-related knowledge; the product assembly quality-related knowledge includes quality parameters, process parameters, and entity relationship data. The Bayesian network mapping unit 202 is used to map the nodes and relationships in the assembly quality knowledge graph to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are used as the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are used as the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are used as the leaf nodes of the Bayesian network model. The probability assignment unit 203 is used to assign prior probabilities and conditional probabilities based on fuzzy probability theory to the nodes in the Bayesian network model; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. The traceability report generation unit 204 is used to perform dynamic reasoning and traceability of abnormal process parameters in product assembly based on the Bayesian network model and combined with reinforcement learning strategies, and to generate a traceability report.

[0062] This application embodiment can also provide a product assembly anomaly process parameter tracing device, the device including a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the steps of the above-described method for tracing the process parameters of abnormal product assembly according to the instructions in the program code.

[0063] like Figure 3 As shown in the figure, the product assembly anomaly process parameter traceability device provided in this application embodiment may include: a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, memory 11, and communication interface 12 all communicate with each other through the communication bus 13.

[0064] In this embodiment, the processor 10 may be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0065] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiment of the product assembly abnormal process parameter tracing method.

[0066] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions: An assembly quality knowledge graph is constructed using the collected product assembly quality-related knowledge; the product assembly quality-related knowledge includes quality parameters, process parameters, and entity relationship data. The nodes and relationships in the assembly quality knowledge graph are mapped to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are the leaf nodes of the Bayesian network model. The nodes in the Bayesian network model are assigned prior probabilities and conditional probabilities based on fuzzy probability theory; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. Based on the Bayesian network model, a reinforcement learning strategy is used to dynamically infer and trace the abnormal process parameters of product assembly, and a traceability report is generated.

[0067] In one possible implementation, the memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function (such as file creation or data read / write). The data storage area may store data created during use, such as initialization data.

[0068] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0069] Communication interface 12 can be an interface for a communication model, used to connect with other devices or systems.

[0070] Of course, it should be noted that, Figure 3 The structure shown does not constitute a limitation on the product assembly anomaly process parameter traceability device in the embodiments of this application. In practical applications, the product assembly anomaly process parameter traceability device may include more than Figure 3 More or fewer components as shown, or combinations of certain components.

[0071] This application embodiment may also provide a computer-readable storage medium for storing program code for executing the steps of the above-described method for tracing the source of abnormal process parameters in product assembly.

[0072] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0073] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0074] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for tracing the source of abnormal process parameters in product assembly, characterized in that, include: An assembly quality knowledge graph was constructed using the collected knowledge related to product assembly quality. The knowledge related to product assembly quality includes quality parameters, process parameters, and entity relationship data. The nodes and relationships in the assembly quality knowledge graph are mapped to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are the leaf nodes of the Bayesian network model. The nodes in the Bayesian network model are assigned prior probabilities and conditional probabilities based on fuzzy probability theory; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. Based on the Bayesian network model, a reinforcement learning strategy is used to dynamically infer and trace the abnormal process parameters of product assembly, and a traceability report is generated.

2. The method for tracing abnormal process parameters in product assembly according to claim 1, characterized in that, The fuzzy probability subset of root node quality problems is represented by a triangular fuzzy number; the fuzzy probability subset of the probability of root node assembly quality problems in the Bayesian network model is represented by the following formula: In the formula: This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node appearing.

3. The method for tracing abnormal process parameters in product assembly according to claim 2, characterized in that, Using the OR gate algorithm in the fault tree, the fuzzy probabilities of intermediate nodes and leaf nodes are obtained as shown in the following formula: In the formula: Indicates the number of nodes. Representing the knowledge graph of the first A fuzzy probability subset of root nodes experiencing quality problems This represents the minimum probability of occurrence. Indicates the maximum probability of occurrence. This represents the median probability of the root node appearing.

4. The method for tracing abnormal process parameters in product assembly according to claim 1, characterized in that, Using the fuzzy probability as the posterior probability of the Bayesian network, the probability of the root node appearing when leaf node T appears is expressed by the following formula: In the formula: It is the prior probability of the root node. When the root node is known The probability of leaf node T occurring when it occurs.

5. The method for tracing abnormal process parameters in product assembly according to claim 1, characterized in that, The dynamic reasoning and tracing also includes a key parameter identification step: Calculate the root node For leaf nodes The probability and critical importance of being in an abnormal state are determined to quantify the impact of each root node on the abnormal result and identify the key process parameters that lead to assembly abnormalities.

6. The method for tracing abnormal process parameters in product assembly according to claim 5, characterized in that, root node The state is The probability formula is as follows: in, It is the root node The conditional probability, The value is 0 when there are no assembly quality issues. The value is 1 when there are assembly quality issues; leaf nodes In state The probabilities are as follows: Where: leaf nodes do not occur When a leaf node occurs, the value is 0. =1, This represents the nth root node in the set of root nodes; When the root node The state is When, make the leaf node T state is The conditional probability is given by the following formula: When the root node The state is When, the leaf node T is in only state The probabilistic importance is given by the following formula: For the root node The root node is obtained by averaging the incremental contributions across all states. For leaf nodes The overall probabilistic importance is given by the following formula: In the formula, K represents the number of assembly quality problems removed from the root node; When the root node In In the state, leaf node T alone is The critical importance of a state is as follows: root node For the state is The critical importance of the leaf node T is as follows: 。 7. The method for tracing abnormal process parameters in product assembly according to claim 1, characterized in that, The Bayesian network model includes an improved Q-function, the equation of which is as follows: In the formula: Indicates a reward. Indicates state, Indicates the next state. Indicates the reasoning path, Representation strategy, Actions that indicate reasoning Indicates the action of the next reasoning. This represents an action-state value function.

8. The method for tracing abnormal process parameters in product assembly according to claim 1, characterized in that, The Bayesian network model uses non-policy methods to find the optimal policy, which includes optimization policies and execution policies.

9. The method for tracing abnormal process parameters in product assembly according to claim 8, characterized in that, The optimization strategy is expressed by the following formula: The execution strategy is represented by the following formula, which is the optimal tracing action that allows the Q function to reach its maximum value under the current process state s: In the formula: Represents the optimal action-state value function. This represents the state-action value function within the Bayesian framework. This represents the action space corresponding to state s.

10. A device for tracing abnormal process parameters in product assembly, characterized in that, The apparatus for performing the product assembly anomaly process parameter tracing method according to any one of claims 1-9, the apparatus comprising: The knowledge graph construction unit is used to construct an assembly quality knowledge graph using the collected product assembly quality-related knowledge; the product assembly quality-related knowledge includes quality parameters, process parameters, and entity relationship data. A Bayesian network mapping unit is used to map the nodes and relationships in the assembly quality knowledge graph to a Bayesian network model, wherein the nodes of assembly quality problems in the assembly quality knowledge graph are used as the root nodes of the Bayesian network model, the relationships between assembly quality and process quality parameters in the assembly quality knowledge graph are used as the intermediate nodes of the Bayesian network model, and the nodes of assembly process quality parameters in the assembly quality knowledge graph are used as the leaf nodes of the Bayesian network model. The probability assignment unit is used to assign prior probabilities and conditional probabilities based on fuzzy probability theory to the nodes in the Bayesian network model; the fuzzy probabilities of each assembly quality node in the assembly quality knowledge graph are used as the prior probabilities of the corresponding root nodes in the Bayesian network model. The traceability report generation unit is used to perform dynamic reasoning and traceability of abnormal process parameters in product assembly based on the Bayesian network model and combined with reinforcement learning strategies, and to generate a traceability report.

Images