Large-scale customized production abnormity regulation and control method based on knowledge graph

Abstract

The invention provides a large-scale customized production abnormity regulation and control method based on a knowledge graph in order to solve the technical problems that an existing production abnormity regulation and control technology mainly focuses on recognition and diagnosis of a single abnormity type, only local accurate analysis can be achieved, and large-scale customized production multi-source abnormity cooperative regulation and control cannot be achieved. By constructing the knowledge graph in the large-scale customized production scene fused with multi-source data and combining the established anomaly classification system and the abnormal event influence quantitative grading model, the anomaly is accurately recognized and graded, and different grades of abnormal events in large-scale customized production can be solved by effectively utilizing experience knowledge.

Description

Technical Field

[0001] This invention relates to the field of mass customization manufacturing systems, and more particularly to a method for controlling production anomalies in mass customization manufacturing systems. Background Technology

[0002] Mass customization, while meeting individual consumer demands, also pursues economies of scale. By building flexible manufacturing systems that combine personalization and standardization, it enables dynamic and rapid responses to user needs. However, this also directly increases the complexity of the manufacturing system, leading to frequent production anomalies such as material supply fluctuations, frequent process route changes, equipment compatibility conflicts, and scheduling disruptions. To ensure delivery efficiency and quality stability, it is essential to build an intelligent control system for production anomalies that can quickly identify, accurately trace, and dynamically adjust, in order to resolve anomalies in real time and maintain efficient and stable system operation.

[0003] Existing production anomaly control technologies mostly employ data-driven methods (such as the AI-based fault monitoring disclosed in invention patents CN116610974A and CN114357185A), focusing on the identification and diagnosis of single anomaly types (such as the equipment fault analysis disclosed in invention patents CN116975639B and CN113094512B). These methods can only achieve localized, precise analysis and cannot coordinate the control of multi-source anomalies in large-scale customized production. Furthermore, existing production anomaly control technologies lack a knowledge-based closed-loop control framework, resulting in the ineffective reuse of accumulated experience from historical production processes. This makes it difficult to support real-time handling of anomalies triggered by frequent system changes in large-scale customized production models. Summary of the Invention

[0004] To overcome the limitations of existing production anomaly control technologies, which mostly focus on the identification and diagnosis of single anomaly types and can only achieve localized precise analysis, failing to coordinate and control multi-source anomalies in large-scale customized production, this invention proposes a knowledge graph-based method for anomaly control in large-scale customized production. By constructing a knowledge graph that integrates multi-source data in a large-scale customized production scenario, and combining it with an established anomaly classification system and an anomaly event impact quantification and grading model, anomalies can be accurately identified and graded for handling. This method can effectively utilize experiential knowledge to solve anomalies of different levels in large-scale customized production.

[0005] The core solution of this invention comprises the following three parts:

[0006] (I) Constructing a knowledge graph of anomalies in large-scale customized production

[0007] First, establish a data collection framework covering three core data categories: product data, abnormal event information, and historical handling information.

[0008] Subsequently, a large-scale customized production anomaly ontology model was constructed, which includes seven types of ontology concepts such as products, anomaly objects, and anomaly events, and the corresponding seven types of ontology relationships.

[0009] Finally, an adaptive knowledge extraction method for structured data, unstructured text, and multimodal data in large-scale customized production scenarios is proposed.

[0010] (ii) Anomaly identification in large-scale customized production

[0011] First, an abnormal event perception method, including active perception and / or indirect perception, was established;

[0012] Subsequently, an anomaly classification system containing 25 typical abnormal events was defined from two dimensions: production process and production factors.

[0013] Finally, a quantitative classification model for the impact of abnormal events was established from five dimensions.

[0014] (III) Knowledge Graph-Based Classified Handling of Large-Scale Customized Production Anomalies

[0015] A tiered handling strategy has been established for three levels of anomalies: minor, important, and critical. The basic process includes anomaly location, handling plan delivery, and anomaly handling execution.

[0016] The technical solution of this invention is:

[0017] The knowledge graph-based method for controlling anomalies in large-scale customized production is unique in that it includes the following steps:

[0018] Step 1: Construct a knowledge graph of production anomalies for large-scale customization;

[0019] The knowledge graph includes historical abnormal events, abnormal objects, abnormal related products, abnormal handling plans, abnormal handling results, and knowledge of the impact of abnormalities.

[0020] Step 2: Anomaly detection;

[0021] Step 2.1: Directly and / or indirectly perceive the characteristic information of abnormal phenomena and abnormal objects;

[0022] Step 2.2: Identify the abnormal events corresponding to the perception results in Step 2.1 according to the preset abnormal classification system; the abnormal classification system is a two-dimensional classification system composed of the production process dimension and the production element dimension; the production process dimension includes R&D design, procurement execution, production manufacturing, warehousing and logistics, and quality management; the production element dimension includes personnel, machines, materials, methods, and environment; the elements in the production process dimension and the production element dimension are combined in pairs to form the two-dimensional classification system;

[0023] Step 2.3: Based on the preset abnormal event impact quantification and grading model, perform multi-dimensional quantification scoring on the abnormal events identified in Step 2.2, and classify the abnormal events into various abnormal grading types by combining the multi-dimensional quantification scoring results; the multi-dimensional factors include the impact on product quality, the frequency of abnormal occurrence, the scope of abnormal impact, the time of abnormal handling, and the impact of abnormal handling on costs;

[0024] Step 3: Classify and handle perceived anomalies;

[0025] Based on the anomalies and their classifications identified in step 2, and combined with the knowledge graph established in step 1, different handling strategies are adopted to respond to and resolve anomalies of different levels.

[0026] Furthermore, the nodes in the knowledge graph constructed in step 1 include products, abnormal events, abnormal objects, abnormal phenomenon characteristics, abnormal event impacts, abnormal handling solutions, and abnormal handling results. The relationships include <abnormal event → related → product>, <abnormal object → trigger → abnormal event>, <abnormal event → manifest as → abnormal phenomenon characteristics>, <abnormal event → lead to → abnormal event impact>, <abnormal handling solution → generate → abnormal handling result>, <abnormal handling result → feedback → abnormal event>, and <abnormal event → need → abnormal handling solution>.

[0027] Furthermore, the pre-defined anomaly classification system in step 2.2 is as follows:

[0028] .

[0029] Furthermore, the pre-defined anomaly classification system in step 2.2 is as follows:

[0030] .

[0031] Furthermore, the pre-defined abnormal event impact quantification and grading model in step 2.3 is as follows:

[0032] ;

[0033] In the above table:

[0034] First preset ratio > Second preset ratio > Third preset ratio > Fourth preset ratio;

[0035] Fifth preset time > Sixth preset time > Seventh preset time > Eighth preset time;

[0036] Fifth preset ratio > Sixth preset ratio > Seventh preset ratio > Eighth preset ratio;

[0037] If the total score for the abnormal event is 4-6 points, it is judged as a minor abnormality;

[0038] If the total score of the abnormal event is 7-15 points, it is judged as a significant abnormality;

[0039] If the total score of the abnormal event is 16-20 points, it is judged as a serious abnormality.

[0040] Furthermore, the pre-defined abnormal event impact quantification and grading model in step 2.3 is as follows:

[0041] ;

[0042] If the total score for the abnormal event is 4-6 points, it is judged as a minor abnormality;

[0043] If the total score of the abnormal event is 7-15 points, it is judged as a significant abnormality;

[0044] If the total score of the abnormal event is 16-20 points, it is judged as a serious abnormality.

[0045] Furthermore, the handling strategies for each level of anomalies in step 3 are as follows:

[0046] For minor abnormalities:

[0047] First, based on the abnormal phenomenon features and abnormal objects identified in step 2, the abnormal event nodes that match the current abnormal phenomenon features and abnormal objects are located using the "abnormal object → abnormal event → abnormal phenomenon" relationship chain defined in the knowledge graph constructed in step 1.

[0048] Then, based on the relationship of "abnormal event → need → abnormal handling plan" in the knowledge graph, retrieve and obtain the historical handling plan instances associated with the abnormal event node and their corresponding abnormal event context information;

[0049] Finally, referring to the information retrieved from the knowledge graph, the obtained historical handling plan is executed to complete the repair and handling of minor anomalies;

[0050] For significant anomalies:

[0051] First, based on the abnormal phenomenon features and abnormal objects identified in step 2, the initial abnormal event node is located using the "abnormal object → abnormal event → abnormal phenomenon feature" relationship chain in the knowledge graph constructed in step 1.

[0052] Then, based on the relationship of "abnormal event → cause → impact of abnormal event" in the knowledge graph constructed in step 1, the impact propagation mechanism of the current abnormal event is analyzed and determined.

[0053] Secondly, based on the information of the impact propagation mechanism, other abnormal objects and corresponding phenomena that may be affected are identified. Based on the characteristics of these inferred phenomena and objects, the relationship chain of "abnormal object → abnormal event → abnormal phenomenon" is used again to locate all relevant potential abnormal event nodes.

[0054] Finally, based on the identified potential abnormal event nodes, and further based on the "abnormal event → related → product" relationship in the knowledge graph constructed in step 1, we screen out the abnormal event nodes associated with historical products that are most similar to the current production product type and configuration, as well as their corresponding historical collaborative handling solutions.

[0055] For critical anomalies:

[0056] First, comprehensively identify and summarize the set of all observable anomalies and their corresponding sets of anomalies in the current manufacturing system;

[0057] Then, based on the "abnormal object → abnormal event → abnormal phenomenon feature" relationship chain in the knowledge graph constructed in step 1, locate all abnormal event nodes that match all currently identified abnormal phenomenon features and abnormal objects;

[0058] Next, all the located abnormal event nodes are sorted in descending order of their node degree in the knowledge graph;

[0059] Finally, based on the priority ranking from high node degree to low node degree, and combined with the basic information and handling solutions for abnormal events provided by the knowledge graph, the root causes of each high-priority abnormal event node are investigated in depth. Based on the investigation results and global impact analysis, the handling solution that best fits the current overall abnormal situation of the manufacturing system is selected or integrated, and necessary resources are scheduled for implementation to complete the global repair and recovery of the entire manufacturing system.

[0060] The beneficial effects of this invention are:

[0061] 1. Taking full account of the specific scenarios of handling anomalies in large-scale customized production, a knowledge graph of anomalies in large-scale customized production is constructed, which standardizes the experience knowledge of handling manufacturing anomalies in large-scale customized production, and provides a foundation for knowledge transfer and reuse under the condition of frequent changes in product objects and production resources;

[0062] 2. By establishing a large-scale customized production anomaly classification system and an anomaly event impact quantification and grading model, it is beneficial to conduct quantitative perception and analysis of various anomalies in the production scenario, and provide a reference for the selection of anomaly handling strategies.

[0063] 3. By establishing a knowledge graph-based strategy for handling large-scale customized production anomalies, it is possible to handle anomalies based on historical knowledge for different levels of anomalies, which is conducive to improving the effectiveness of anomaly handling in large-scale production scenarios. Attached Figure Description

[0064] Figure 1 This is the overall framework for a knowledge graph-based method for controlling anomalies in large-scale customized production.

[0065] Figure 2 This is a large-scale customized production anomaly ontology model established in one embodiment of the present invention. Detailed Implementation

[0066] This invention provides a knowledge graph-based method for anomaly control in large-scale customized production. By constructing a knowledge graph of anomalies in large-scale customized manufacturing, identifying and evaluating anomalies in the large-scale customized manufacturing system, and recommending anomaly control schemes, it achieves efficient identification and control of anomalies in the production process. The specific implementation steps of this invention are as follows:

[0067] Step 1: Construct a knowledge graph of large-scale customized production anomalies

[0068] By constructing a large-scale customized production anomaly knowledge graph, the structured integration and semantic association of multi-source heterogeneous data and historical experience knowledge during the production process are achieved, providing interpretable knowledge support for anomaly control. This knowledge graph includes at least event information, anomaly objects, related products, and handling solutions from historical anomaly events, and supports dynamic knowledge updates to adapt to the flexible needs of customized production.

[0069] The specific steps for constructing a knowledge graph of large-scale customized production anomalies include: data acquisition, ontology modeling, and knowledge extraction.

[0070] Step 1.1: Data Acquisition;

[0071] Data acquisition aims to comprehensively obtain multi-source heterogeneous data related to production anomalies, providing foundational data support for the construction of knowledge graphs. This ensures the accurate representation of the complex relationships between abnormal objects, products, events, and their handling solutions in large-scale customized production, thereby supporting precise and intelligent anomaly control. The data acquisition targets and methods must be clearly defined during the data acquisition process.

[0072] (1) Collection objects

[0073] The data to be collected refers to the data entities and their associated information that need to be obtained in the process of building a large-scale customized production anomaly knowledge graph, including at least: product data, historical anomaly event information, and historical anomaly handling information.

[0074] Data collection object 1: Product data;

[0075] Products in mass customization manufacturing systems are characterized by high personalization, complex and variable processes, and stringent quality requirements. Each product category may involve unique materials, process parameters, and processing paths. Product data reflects basic product information, processing methods, and historical anomalies, serving as a crucial source of knowledge for guiding new product production resource allocation and anomaly analysis and control. To support anomaly analysis and handling in mass customization production, the product data collected must at least include the product's serial number, name, and type.

[0076] Target 2: Historical abnormal event information;

[0077] Abnormal events in mass-customized manufacturing systems are diverse, complex, and dynamic. Abnormal event information reflects the type, occurrence time, scope of impact, and root cause of anomalies, serving as crucial knowledge for root cause localization, impact analysis, and control and management. To support the analysis and handling of anomalies in mass-customized production, the historical abnormal event information to be collected should include at least the following: basic information about the abnormal event, descriptive text of the historical abnormal event, information about the abnormal object, related product numbers, characteristic information of the abnormal phenomenon, and impact information of the abnormal event. Basic information includes the abnormal event number and the time of occurrence; abnormal object information includes the abnormal object number, type of abnormal object (equipment, personnel, materials, methods, or environment), and status data of the abnormal object; characteristic information of the abnormal phenomenon includes keywords and textual descriptions of the abnormal phenomenon; and impact information includes the scope of impact, degree of impact, and propagation mechanism.

[0078] Target 3: Historical anomaly handling information;

[0079] Anomaly handling in mass-produced customized manufacturing systems is diverse and complex. Due to the personalized needs of customized production, anomaly handling requires differentiated strategies for different anomaly types and scenarios. Historical anomaly handling information reflects the measures, effects, and experience gained in dealing with anomalies, serving as crucial knowledge for optimizing control strategies and improving anomaly response efficiency. To support anomaly analysis and handling in mass-produced customized production, the historical anomaly handling information to be collected should include at least historical anomaly handling plan information and historical anomaly handling result information. Historical anomaly handling plan information includes the plan number, handling measures, and applicable scenarios; historical anomaly handling result information includes the result number and result description text.

[0080] (2) Data collection method

[0081] Product data is typically stored in a company’s own ERP (Enterprise Resource Planning) system. The product data required by this invention can be extracted from the existing ERP system through cross-system API integration.

[0082] Information on historical abnormal events and their handling cannot be directly collected from existing information systems. Instead, it can be manually entered by frontline staff or abnormal event handling experts, depending on the specific circumstances of the abnormal event.

[0083] Step 1.2: Ontology modeling;

[0084] Ontology modeling aims to construct an ontology model to address the needs of anomaly control in large-scale customized production. This ontology model defines core concepts and their interrelationships within a specific domain, providing a unified structured framework for the semantic integration of multi-source heterogeneous data to support knowledge reasoning based on a knowledge graph of large-scale customized production anomalies. The ontology model consists of two main parts: ontology classes and ontology relations.

[0085] (1) Ontology class

[0086] An ontology class is a classification of abstract things in a specific domain. To support the construction and application of knowledge graphs for large-scale customized production anomalies, the ontology class in this invention includes at least: product, abnormal object, abnormal event, abnormal phenomenon characteristics, abnormal event impact, abnormal handling plan, and abnormal handling result.

[0087] Ontology Class 1: Product;

[0088] Product classes define the various products produced in a mass customization manufacturing system. In the mass customization production anomaly knowledge graph, nodes belonging to product classes can be created. A product class node represents a type of product object. The identification information of the product class comes from the product number in the product data and historical anomaly event information. Other data in the product data is stored in the form of product class node attributes.

[0089] Ontology class 2: Exception object;

[0090] Anomaly object classes define the attributes and states of various anomaly objects (including personnel, machines, materials, methods, and environments) in a mass-customized manufacturing system. In the mass-customized production anomaly knowledge graph, nodes belonging to anomaly object classes can be created; one anomaly object class node represents one anomaly object instance. The identification information of anomaly object classes comes from the anomaly object numbers in historical anomaly event information, while other anomaly object data is stored in the form of anomaly object class node attributes.

[0091] Ontology Class 3: Exception Events;

[0092] Exception event classes define various exceptions that occur during mass customization production. In the mass customization production exception knowledge graph, nodes belonging to exception event classes can be created; each exception event class node represents an independent exception event instance. The unique identifier of an exception event class comes from the basic exception event information in historical exception event information; other text describing the exception event is stored as attributes of the exception event class node.

[0093] Ontology Class 4: Characteristics of Abnormal Phenomena;

[0094] Anomaly feature classes define the specific characteristics exhibited when an anomaly occurs, such as reduced personnel efficiency, abnormal equipment parameters, material shortages, process anomalies, and environmental data fluctuations. In the knowledge graph of anomalies in large-scale customized production, nodes belonging to anomaly feature classes can be created. Each anomaly feature class node represents an observable anomaly feature. The identification data for anomaly feature classes comes from the anomaly feature information field in historical anomaly event information, and the textual description of this feature is stored as an attribute of the anomaly feature class node.

[0095] Ontology Class 5: Impact of Abnormal Events;

[0096] An exception impact class defines the effect of an exception event on the manufacturing system. In the knowledge graph of exceptions in mass customized production, nodes belonging to exception impact classes can be created, with each node representing an instance of an exception event's impact. The identification data for exception impact classes comes from the exception event impact information in historical exception event information. The scope, degree, and propagation mechanism of impact information in historical exception event information are stored as attributes of exception impact class nodes.

[0097] Ontology Class 6: Exception Handling Plan;

[0098] The exception handling plan class defines the handling measures for different exception events. In the large-scale customized production exception knowledge graph, nodes belonging to exception handling plan classes can be created, with each exception handling plan class node representing an instance of an exception handling plan. The identifier of the exception handling plan class comes from the handling measure field of historical exception handling information, and the textual description of the handling plan is stored in the form of exception handling plan class node attributes.

[0099] Ontology Class 7: Exception Handling Results;

[0100] The anomaly handling result class defines the implementation effect of anomaly handling measures. In the knowledge graph of anomalies in large-scale customized production, nodes belonging to the anomaly handling result class can be created. An anomaly handling result class node represents the execution result of an anomaly handling plan. The identifier of the anomaly handling result class comes from the handling result information in historical anomaly handling information, and the textual description of the handling result is stored in the form of anomaly handling result class node attributes.

[0101] (2) Ontological Relationship

[0102] Ontology relations are semantic associations between ontology classes. To support knowledge reasoning based on large-scale customized production anomaly knowledge graphs, the ontology relations in this invention include at least: related, triggering, manifesting as, causing, needing, generating, and feedback.

[0103] Ontology Relationship 1: <Abnormal Event → Related → Product>;

[0104] The "relevance" relationship indicates the specific product object associated with an anomaly. This relationship is used to locate the carrier of the anomaly or the specific product instance / model affected, and is the basis for tracing the root cause of the anomaly, assessing the scope of impact, and optimizing the production process for specific products.

[0105] Ontology Relationship 2: <Exceptional Object → Triggering → Exceptional Event>;

[0106] The "triggering" relationship indicates the abnormal events that an abnormal object may trigger. It is used to analyze how changes in the state of an abnormal object lead to production abnormalities, thereby optimizing the monitoring and intervention process for abnormal objects.

[0107] Ontology Relationship 3: <Abnormal Event → Manifestation as → Abnormal Phenomenon Characteristics>;

[0108] The "manifestation as" relationship indicates the specific characteristics exhibited by an abnormal event when it occurs. It is used to identify and diagnose abnormal events, and to quickly determine the type and severity of the abnormality by analyzing the characteristics of the abnormal phenomenon.

[0109] Ontology Relationship 4: <Abnormal Event → Causes → Impact of Abnormal Event>;

[0110] The "cause" relationship indicates the multifaceted impact of an abnormal event on the manufacturing system. It is used to assess the severity of the abnormal event and to provide a basis for classifying abnormality levels.

[0111] Ontology Relationship 5: <Abnormal Event → Need → Abnormal Handling Plan>;

[0112] The "need" relationship indicates the measures that need to be taken in response to a specific abnormal event, in order to quickly respond to the abnormal event and optimize the handling strategy.

[0113] Ontology Relationship 6: <Exception Handling Plan → Generation → Exception Handling Result>;

[0114] The "generation" relationship indicates the effect produced after the implementation of the abnormality handling plan. It is used to evaluate the effectiveness of the handling plan and to provide a basis for the selection of future abnormality control strategies.

[0115] Ontology Relationship 7: <Abnormal Handling Result → Feedback → Abnormal Event>;

[0116] The "feedback" relationship represents the feedback information from the abnormal handling results to the abnormal event, which is used to form a closed-loop management of abnormal control and continuously optimize the abnormal handling strategy through feedback information.

[0117] This invention constructs a large-scale customized production anomaly ontology model based on the aforementioned ontology classes and ontology relationships using existing conventional methods, as shown below. Figure 2 As shown.

[0118] Step 1.3: Map Construction;

[0119] Knowledge graph construction refers to the process of transforming the multi-source data collected in step 1.1 into a structured knowledge graph based on the ontology model constructed in step 1.2. This invention instantiates the concepts and relationships in the ontology model into nodes and edges in the knowledge graph based on mapping rules.

[0120] The specific mapping rules are as follows:

[0121] Node mapping rule 1: <Product data → Product node>

[0122] Construct product nodes, using the product number in the product data as the product node identifier, and store the name and type as node attributes.

[0123] Node mapping rule 2: <Historical abnormal event information → Abnormal event node>

[0124] Construct abnormal event nodes, using the abnormal event number in the historical abnormal event information as the node identifier, and store the abnormal event description text and the time of occurrence of the abnormal event as node attributes.

[0125] Node mapping rule 3: <Historical abnormal event information → Abnormal object node>

[0126] Construct an exception object node, using the exception object number in the exception event information as the node identifier, and store the exception object type and exception object status data as node attributes.

[0127] Node mapping rule 4: <Historical abnormal event information → Abnormal phenomenon feature nodes>

[0128] Construct anomalous phenomenon feature nodes, using keywords from the anomalous phenomenon feature information as node identifiers, and store the anomalous event feature description text as node attributes.

[0129] Node mapping rule 5: <Historical abnormal event information → Nodes affected by abnormal events>

[0130] Construct nodes that are affected by abnormal events, and store the degree of impact and the mechanism of impact propagation as node attributes.

[0131] Node mapping rule 6: <Historical anomaly handling information → Anomaly handling plan node>

[0132] Construct an anomaly handling plan node, using the handling plan number in the historical anomaly handling information as the node identifier, and store the handling measures text and applicable scenarios as node attributes.

[0133] Node mapping rule 7: <Historical anomaly handling information → Anomaly handling result node>

[0134] Construct an exception handling result node, using the handling result number as the node identifier, and store the result description text as the node attribute.

[0135] Edge mapping rule 1:

[0136] An edge with the relationship <abnormal event→related→product> is established between the abnormal event node obtained by mapping historical abnormal event information and the product node corresponding to the relevant product number in the information.

[0137] Edge mapping rule 2:

[0138] Nodes obtained from mapping information of the same historical abnormal event are connected by four types of relationships based on the ontology class to which these nodes belong: <Abnormal object → trigger → abnormal event>, <Abnormal event → manifests as → abnormal phenomenon characteristics>, <Abnormal event → leads to → abnormal event impact>, and <Abnormal event → requires → abnormal handling plan>.

[0139] Edge mapping rule 3:

[0140] For nodes obtained by mapping the same historical anomaly handling information, establish two types of edges based on the ontology class to which these nodes belong: <anomaly handling plan → generation → anomaly handling result> and <anomaly handling result → feedback → anomaly event>.

[0141] Step 2: Identification of anomalies in large-scale customized production;

[0142] In mass customization production, anomaly detection refers to the use of intelligent algorithms to analyze massive amounts of heterogeneous production data from multiple sources in real time to identify abnormal events that deviate from predetermined standards, plans, or expected states. Its goal is to quickly pinpoint the source of anomalies, identify the type of anomaly, and define its scope of impact in a complex and ever-changing production environment, thereby providing accurate target and contextual information for subsequent anomaly handling. This process specifically includes three stages: anomaly perception, anomaly event classification, and anomaly level assessment.

[0143] Step 2.1: Perception of anomalies in mass customization production;

[0144] Anomaly perception is the starting point of the anomaly identification process. In a complex and ever-changing large-scale customized production environment, it refers to the real-time monitoring and analysis of massive amounts of multi-source heterogeneous data to instantly capture observable deviations from normal states that characterize the occurrence of abnormal events. These perceived anomaly characteristics and objects serve as the basis for subsequent anomaly classification and grading. Depending on the information acquisition method, the perception method employed in this invention includes at least one of direct and indirect perception; enterprises can choose the appropriate method based on their own production conditions.

[0145] (1) Direct perception

[0146] Direct sensing primarily targets abnormal objects in the machine and environmental categories. It utilizes sensors such as temperature, vibration, and pressure to acquire dynamic operational data, automatically detecting and proactively reporting any abnormal phenomena occurring within the device. For example, vibration signals collected by a vibration sensor mounted on a bearing housing can be used to detect abnormalities in the drive motor of a device using the LightGBM model. Those skilled in the art can configure sensors and train the sensing model based on the possible abnormal events of the abnormal object. Specific implementation methods can refer to other data-driven anomaly sensing methods (such as patent CN115204555A). The improvement of this invention lies not in the method of identifying anomalies through direct sensing, but merely in using the sensing results of this step as input information for subsequent anomaly analysis and handling.

[0147] (2) Indirect perception

[0148] Indirect perception refers to the manual identification of abnormal phenomena in an object or manufacturing system through observation and analysis of production data and / or on-site conditions. This method is suitable for scenarios where the object itself lacks self-diagnostic capabilities or requires analysis of the interrelationships between various stages in a complex production environment from a global perspective. For example, a material planner can identify a critical material shortage anomaly by observing the inventory status and production consumption rate of components, indicating that the expected available inventory of critical component A will fall below the safety stock level within the next 8 hours. Specific implementation methods can refer to existing anomaly identification technologies or processes. The improvement of this invention does not lie in the method of identifying anomalies through indirect perception, but rather in using the perception results obtained in this step as input information for subsequent anomaly classification and level assessment.

[0149] Step 2.2: Classification of abnormal events in large-scale customized production;

[0150] In large-scale customized production, anomaly classification refers to defining the type of anomaly events based on the characteristic information and objects of the anomaly captured in the anomaly perception process. Its core objective is to clarify the attribution of the anomaly, enabling the manufacturing system or relevant personnel to quickly access standardized handling procedures, knowledge bases, or resources for that type of anomaly from the knowledge graph constructed in step 1. This provides accurate contextual basis for subsequent anomaly level assessment and handling decisions.

[0151] To identify anomalous events in mass customization scenarios, this invention proposes a two-dimensional anomaly classification system. This system approaches the issue from two key dimensions: production processes and production factors, defining 25 typical production anomaly events. The production process dimension covers the main business domains of mass customization production, including R&D design, procurement execution, manufacturing, warehousing and logistics, and quality management. These processes are interconnected and operate collaboratively, forming a complete production value chain. The production factor dimension focuses on the fundamental conditions supporting production operations, including personnel, machines, materials, methods, and environment. The effective allocation and integration of these factors are prerequisites for smooth process operation. Any production anomaly can be traced back to the specific process step in which it occurred and its related root causes. By pairwise combining the five production processes with the five production factors, the mass customization production anomaly classification system shown in Table 1 is constructed.

[0152] Table 1. Classification System for Anomalies in Mass Customized Production

[0153] ;

[0154] This anomaly classification system allows for the systematic and standardized categorization of diverse anomalies in mass-customized production, providing a foundation for rapid and accurate anomaly identification and handling. This embodiment defines anomaly classification systems for mass-customized production and their quantitative evaluation thresholds. Those skilled in the art can adjust the thresholds according to specific production requirements. For example, some companies produce products with strong customization characteristics, requiring frequent process adjustments and optimizations to improve product quality. Therefore, in terms of methodological elements during the manufacturing phase, the threshold for the number of product process changes can be increased to 10.

[0155] Step 2.3: Assessment of anomaly levels in large-scale customized production;

[0156] In large-scale customized production, anomaly level assessment refers to the quantitative evaluation of the impact and urgency of identified and categorized anomalous events, based on the results of anomaly perception and classification. This assessment determines the potential or actual scope and intensity of the anomaly's impact on production objectives, thereby determining the priority of its handling and the required resource allocation level. By assessing the impact level of anomalous events, it is possible to ensure that in the resource-constrained, highly complex, and volatile environment of large-scale customized production, limited resources can be precisely focused on the most impactful and urgent critical anomalies, effectively preventing local anomalies from escalating into systemic risks.

[0157] To guide anomaly response decisions in large-scale customized production, this invention proposes an anomaly classification method based on quantitative assessment. This method establishes a quantitative classification model of anomaly impact, scores five key classification indicators for each anomaly, and then combines the scores of these five key indicators to quantitatively classify the anomaly. These five key classification indicators include the anomaly's impact on product quality, frequency, scope of impact, detection and resolution time, and cost impact. This invention establishes a quantitative classification model of anomaly impact based on these key classification indicators to quantitatively score the indicators of anomalies. The scores of different indicators are added together, and the anomaly is classified into three severity levels based on the total score: minor anomaly, major anomaly, and critical anomaly. The quantitative classification model of anomaly impact is shown in Table 2.

[0158] Table 2 Quantitative Grading Model of Abnormal Event Impact

[0159] ;

[0160] Minor anomalies are scored between 4 and 6 points. These anomalies typically do not cause significant disruption to the production process and can be resolved through standard operating procedures or minor adjustments, such as minor equipment malfunctions or temporary delays in raw material supply; their long-term impact is limited.

[0161] The total score for critical anomalies ranges from 7 to 15 points. These anomalies have a significant negative impact on the production process and product quality. Although they do not immediately jeopardize the stability of the manufacturing system, specific measures must be taken immediately to handle them, such as adjusting production plans, reallocating resources, or temporarily reassigning personnel. For example, a critical component failure may require emergency repair.

[0162] The total score for serious anomalies ranges from 16 to 20 points. These anomalies pose an immediate and significant threat to personnel safety and product quality, or may lead to substantial economic losses, requiring an emergency response, including suspending production lines, equipment maintenance, and activating emergency plans. Examples include serious equipment failures or production accidents that may involve injuries or large-scale recalls.

[0163] Step 3: Classified handling of large-scale customized production anomalies based on knowledge graph;

[0164] Based on the identification of the phenomenological characteristics and classification types of abnormal events in step 2, this step implements graded handling for abnormal events of different severity levels in large-scale customized production. Specifically, according to the differences in the severity of abnormal events, three different handling strategies are adopted for proactive response and resolution: individual handling strategy for minor abnormalities, local handling strategy for important abnormalities, and global handling strategy for critical abnormalities.

[0165] Strategy 1: Handling minor abnormalities

[0166] This strategy is suitable for scenarios where a single abnormal object malfunctions, affecting only its own production capacity, can be quickly restored through simple repairs or resource replacement, and does not disrupt the continuity of the overall production process. The execution process of this strategy includes:

[0167] 1) Event node location:

[0168] Based on the abnormal phenomenon features and abnormal objects identified in step 2, the abnormal event nodes that match the current abnormal phenomenon features and abnormal objects are located using the "abnormal object → abnormal event → abnormal phenomenon" relationship chain defined in the large-scale customized production abnormality knowledge graph constructed in step 1.

[0169] 2) Search for treatment options:

[0170] Based on the relationship of "abnormal event → need → abnormal handling plan" in the knowledge graph, retrieve and obtain the historical handling plan instances associated with the abnormal event node and their corresponding abnormal event context information.

[0171] 3) Solution Implementation and Closed-Loop Repair:

[0172] Referring to the information retrieved from the knowledge graph, the acquired historical handling plan is executed to quickly complete the repair and handling of individuals with minor anomalies, and verify that the abnormal object has returned to normal production status, thus achieving closed-loop processing.

[0173] Strategy 2: Localized handling of important anomalies

[0174] This strategy is applicable to scenarios where abnormal events not only affect the source of the abnormality but also pose a direct threat to the stable operation of a local production unit. The handling requires a holistic consideration of the source of the abnormality and its potential affected entities, with coordinated local intervention while ensuring the core objectives of the overall system are met. The execution process of this strategy includes:

[0175] 1) Source event node location:

[0176] Based on the abnormal phenomena features and abnormal objects identified in step 2, the initial abnormal event node is located using the "abnormal object → abnormal event → abnormal phenomenon feature" relationship chain in the large-scale customized production abnormality knowledge graph constructed in step 1.

[0177] 2) Analysis of the influencing mechanism:

[0178] Based on the "abnormal event → cause → impact of abnormal event" relationship in the large-scale customized production anomaly knowledge graph constructed in step 1, the impact propagation mechanism of the current abnormal event is analyzed and determined.

[0179] 3) Affects object identification and related event localization:

[0180] Based on the information about the impact propagation mechanism, other potentially affected anomalous objects and corresponding phenomena are identified. Based on these inferred characteristics and objects, the relationship chain of "anomalous object → anomalous event → anomalous phenomenon" is used again to locate all relevant potential anomalous event nodes.

[0181] 4) Screening and coordinated handling of similar historical solutions:

[0182] Based on the identified potential abnormal event nodes, and further based on the "abnormal event → related → product" relationship in the large-scale customized production abnormality knowledge graph constructed in step 1, abnormal event nodes and their corresponding historical collaborative handling solutions that are most similar to the current production product type and configuration are selected.

[0183] Strategy 3: Global Handling of Critical Anomalies

[0184] This strategy is applicable to scenarios where multiple interrelated abnormal objects and phenomena occur simultaneously within a manufacturing system, representing systemic failures or major bottlenecks. Handling this requires a holistic perspective, coordinating and scheduling various resources within the manufacturing system for comprehensive intervention to restore overall system functionality. The execution process of this strategy includes:

[0185] 1) Global Abnormal State Scan:

[0186] Comprehensively identify and summarize the set of all observable anomalies and their corresponding sets of anomalies in the current manufacturing system.

[0187] 2) Locating the entire set of related event nodes:

[0188] Based on the "abnormal object → abnormal event → abnormal phenomenon feature" relationship chain in the large-scale customized production anomaly knowledge graph constructed in step 1, locate all abnormal event nodes that match all currently identified abnormal phenomenon features and abnormal objects.

[0189] 3) Event node priority ranking:

[0190] All located anomalous event nodes are sorted in descending order of their node degree (i.e., the number of connections between the node and other nodes) within the mass-customized production anomaly knowledge graph. A higher node degree indicates a more central or fundamental position of the anomalous event in the knowledge graph, suggesting a potentially wider impact.

[0191] 4) Prioritization and overall solution implementation:

[0192] Based on the ranking priority (from high node degree to low node degree), and combined with the basic information and handling plans for abnormal events provided by the knowledge graph, the root causes of each high-priority abnormal event node are investigated in depth. Based on the investigation results and global impact analysis, the handling plan that best fits the current overall abnormal situation of the manufacturing system is selected or integrated, and necessary resources are allocated for implementation to complete the global repair and recovery of the entire manufacturing system.

Claims

1. A method for controlling anomalies in large-scale customized production based on knowledge graphs, characterized in that, Includes the following steps: Step 1: Construct a knowledge graph of production anomalies for large-scale customization; The knowledge graph includes historical abnormal events, abnormal objects, abnormal related products, abnormal handling plans, abnormal handling results, and knowledge of the impact of abnormalities. Step 2: Anomaly detection; Step 2.1: Directly and / or indirectly perceive the characteristic information of abnormal phenomena and abnormal objects; Step 2.2: Identify the abnormal events corresponding to the perception results in Step 2.1 according to the preset abnormal classification system; the abnormal classification system is a two-dimensional classification system composed of the production process dimension and the production element dimension; the production process dimension includes R&D design, procurement execution, production manufacturing, warehousing and logistics, and quality management; the production element dimension includes personnel, machines, materials, methods, and environment; the elements in the production process dimension and the production element dimension are combined in pairs to form the two-dimensional classification system; Step 2.3: Based on the preset abnormal event impact quantification and grading model, perform multi-dimensional quantification scoring on the abnormal events identified in Step 2.2, and classify the abnormal events into various abnormal grading types by combining the multi-dimensional quantification scoring results; the multi-dimensional factors include the impact on product quality, the frequency of abnormal occurrence, the scope of abnormal impact, the time of abnormal handling, and the impact of abnormal handling on costs; Step 3: Classify and handle perceived anomalies; Based on the anomalies and their classifications identified in step 2, and combined with the knowledge graph established in step 1, different handling strategies are adopted to respond to and resolve anomalies of different levels.

2. The method for controlling anomalies in large-scale customized production based on knowledge graphs according to claim 1, characterized in that, The nodes in the knowledge graph constructed in Step 1 include products, abnormal events, abnormal objects, characteristics of abnormal phenomena, impact of abnormal events, handling solutions, and results of abnormal handling. The relationships include <abnormal event → related → product>, <abnormal object → trigger → abnormal event>, <abnormal event → manifest as → characteristics of abnormal phenomena>, <abnormal event → leads to → impact of abnormal events>, <abnormal handling solution → generate → result of abnormal handling>, <abnormal handling result → feedback → abnormal event>, and <abnormal event → need → handling solution>.

3. The method for controlling anomalies in large-scale customized production based on knowledge graphs according to claim 1 or 2, characterized in that, The pre-defined anomaly classification system in step 2.2 is as follows: 。 4. The method for controlling anomalies in large-scale customized production based on knowledge graphs according to claim 3, characterized in that, The pre-defined anomaly classification system in step 2.2 is as follows: 。 5. The method for controlling anomalies in large-scale customized production based on knowledge graphs according to claim 3, characterized in that, The pre-defined abnormal event impact quantification and grading model in step 2.3 is as follows: ； In the above table: First preset ratio > Second preset ratio > Third preset ratio > Fourth preset ratio; Fifth preset time > Sixth preset time > Seventh preset time > Eighth preset time; Fifth preset ratio > Sixth preset ratio > Seventh preset ratio > Eighth preset ratio; If the total score for the abnormal event is 4-6 points, it is judged as a minor abnormality; If the total score of the abnormal event is 7-15 points, it is judged as a significant abnormality; If the total score of the abnormal event is 16-20 points, it is judged as a serious abnormality.

6. The method for controlling anomalies in large-scale customized production based on knowledge graphs according to claim 5, characterized in that, The pre-defined abnormal event impact quantification and grading model in step 2.3 is as follows: ； If the total score for the abnormal event is 4-6 points, it is judged as a minor abnormality; If the total score of the abnormal event is 7-15 points, it is judged as a significant abnormality; If the total score of the abnormal event is 16-20 points, it is judged as a serious abnormality.

7. The method for controlling anomalies in large-scale customized production based on knowledge graphs according to claim 5, characterized in that, The handling strategies for each level of anomalies in step 3 are as follows: For minor abnormalities: First, based on the abnormal phenomenon features and abnormal objects identified in step 2, the abnormal event nodes that match the current abnormal phenomenon features and abnormal objects are located using the "abnormal object → abnormal event → abnormal phenomenon" relationship chain defined in the knowledge graph constructed in step 1. Then, based on the relationship of "abnormal event → need → abnormal handling plan" in the knowledge graph, retrieve and obtain the historical handling plan instances associated with the abnormal event node and their corresponding abnormal event context information; Finally, referring to the information retrieved from the knowledge graph, the obtained historical handling plan is executed to complete the repair and handling of minor anomalies; For significant anomalies: First, based on the abnormal phenomenon features and abnormal objects identified in step 2, the initial abnormal event node is located using the "abnormal object → abnormal event → abnormal phenomenon feature" relationship chain in the knowledge graph constructed in step 1. Then, based on the "abnormal event → cause → impact of abnormal event" relationship in the knowledge graph constructed in step 1, the impact propagation mechanism of the current abnormal event is analyzed and determined. Secondly, based on the information of the impact propagation mechanism, other abnormal objects and corresponding phenomena that may be affected are identified. Based on the characteristics of these inferred phenomena and objects, the relationship chain of "abnormal object → abnormal event → abnormal phenomenon" is used again to locate all relevant potential abnormal event nodes. Finally, based on the identified potential abnormal event nodes, and further based on the "abnormal event → related → product" relationship in the knowledge graph constructed in step 1, we screen out the abnormal event nodes associated with historical products that are most similar to the current production product type and configuration, as well as their corresponding historical collaborative handling solutions. For critical anomalies: First, comprehensively identify and summarize the set of all observable anomalies and their corresponding sets of anomalies in the current manufacturing system; Then, based on the "abnormal object → abnormal event → abnormal phenomenon feature" relationship chain in the knowledge graph constructed in step 1, locate all abnormal event nodes that match all currently identified abnormal phenomenon features and abnormal objects; Next, all the located abnormal event nodes are sorted in descending order of their node degree in the knowledge graph; Finally, based on the priority ranking from high node degree to low node degree, and combined with the basic information and handling solutions for abnormal events provided by the knowledge graph, the root causes of each high-priority abnormal event node are investigated in depth. Based on the investigation results and global impact analysis, the handling solution that best fits the current overall abnormal situation of the manufacturing system is selected or integrated, and necessary resources are scheduled for implementation to complete the global repair and recovery of the entire manufacturing system.

Images