CN-122022777-A - Industrial data management method based on big data

CN122022777ACN 122022777 ACN122022777 ACN 122022777ACN-122022777-A

Abstract

The invention discloses an industrial data management method based on big data, which belongs to the technical field of industrial data management and specifically comprises the steps of constructing an after-sales data warehouse based on historical after-sales maintenance records, extracting part association relation and restoring a finished product part level to the after-sales data warehouse, constructing a coupling association topology, screening abnormal samples with deviation from an expected failure mode, extracting corresponding supplier production process parameters and batch information to form a process parameter set, deconstructing the coupling association topology, combining the process parameter set to establish an attribution mapping parameter set, constructing a causal inference model between the attribution mapping parameter set and the failure mode to obtain attribution weight of each supplier batch to the failure mode, and marking the supplier batch with attribution weight exceeding a preset threshold as a target tracing batch and outputting the supplier batch. The invention realizes accurate attribution from after-sale fault data to the batch of suppliers and provides quantitative basis for quality control of a supply chain.

Inventors

SHI ZHIJIAO
Ye Huawu
YE RUOYI
SU LIWEN
LI YIXUAN

Assignees

福州市数据资产运营有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (8)

1. An industrial data management method based on big data is characterized by comprising the following steps: s1, extracting a structured fault code, an unstructured maintenance text, a replacement part identifier and an equipment running log in a historical after-sale maintenance record, and constructing an after-sale data warehouse containing multi-source data; S2, extracting the association relation of the parts and restoring the hierarchy of the finished parts to the after-sales data warehouse to obtain the assembly relation of the parts and a signal transmission path, and constructing a coupling association topology comprising finished nodes, part nodes, supplier nodes and failure mode nodes; S3, screening abnormal samples with deviation from an expected failure mode in the coupling correlation topology from after-sale samples of different batches and working conditions, extracting production process parameters and batch information of suppliers corresponding to the abnormal samples, and integrating to form a process parameter set; s4, deconstructing the coupling association topology, identifying conduction paths of interaction among components in the abnormal sample, and establishing an attribution mapping parameter set by combining the process parameter set; S5, constructing a causal inference model between the attribution mapping parameter set and the fault mode, and obtaining attribution weights of all supplier batches to the fault mode based on the conduction path identification and production process parameters of all supplier batches on the path; and S6, marking the supplier batch and the production batch information thereof with attribution weight exceeding a preset threshold as a target tracing batch, and outputting the supplier code and the batch identification of the target tracing batch.
2. The industrial data management method based on big data according to claim 1, wherein in S1, the specific construction process of the after-sales data warehouse is as follows: Extracting a history maintenance work order from an after-sales management system, obtaining a structured fault code field corresponding to each work order, extracting unstructured maintenance text description from a remark text field of the maintenance work order, extracting a replacement part identifier from a replacement material detail of the maintenance work order, wherein the replacement part identifier comprises a part unique serial number and a supplier code; Performing time stamp alignment on the structured fault code, the unstructured maintenance text and the replacement part identification and the equipment operation log based on the equipment unique identification to generate an initial record set containing multi-source data; performing word segmentation and entity recognition on unstructured maintenance texts in the initial record set, extracting fault phenomenon keywords and maintenance action keywords, and establishing a mapping index with a structured fault code; The method comprises the steps of organizing the processed structured fault codes, unstructured maintenance texts, replacement part identifiers and equipment operation logs according to equipment-time dimensions, constructing wide table data taking maintenance work orders as units, and storing the wide table data in a distributed column database to form an after-sales data warehouse.
3. The industrial data management method based on big data according to claim 1, wherein in S2, the specific acquiring process of the component assembly relation and the signal transmission path is as follows: Based on the assembly records, the final product node is taken as a root, and the parts of each level are recursively unfolded according to a level relation to generate an assembly tree taking the final product node as a root node, the part node as an intermediate node and the part node as a leaf node, wherein the father-son relation corresponding to each side in the assembly tree is a part assembly relation; Extracting a plurality of groups of equipment operation logs corresponding to the same finished product from an after-sale data warehouse, wherein the equipment operation logs comprise time sequence data of each component sensor and control instruction time stamps; sequentially selecting each pair of component combinations for cross-correlation analysis on sensor time sequence data of different components under the same finished product, calculating the similarity of the time sequence of the pair of components under different time offsets, and determining the transmission direction and delay time of signals among the components according to the offset corresponding to the similarity peak value; And for the part pairs with time sequence correlation but without direct electrical connection, constructing a multistage transmission link based on the signal transmission characteristics of intermediate parts to obtain a signal transmission path set comprising a direct path and an indirect path.
4. The industrial data management method based on big data according to claim 3, wherein in S2, the specific construction process of the coupling association topology is as follows: Based on the component assembly relation, taking the finished node as a root, adding the component nodes in the assembly tree into the topology according to the hierarchy, and adding assembly relation edges between the nodes with the assembly relation; adding a signal transfer relation edge between component nodes with signal transfer relation based on the signal transfer path set, wherein the signal transfer relation edge comprises a transfer direction attribute; Extracting fault mode codes corresponding to all components from an after-sales data warehouse, adding the fault mode codes serving as fault mode nodes into a topology, and adding fault association relation edges between the component nodes and the corresponding fault mode nodes; extracting a supplier code from the replacement part identifier, adding the supplier code as a supplier node into the topology, and adding a supply relation edge between the part node and a corresponding supplier node; and fusing the assembly relation edge, the signal transmission relation edge, the fault association relation edge and the supply relation edge to generate a coupling association topology comprising a finished product node, a component node, a provider node and a fault mode node.
5. The industrial data management method based on big data according to claim 1, wherein in S3, the specific generation process of the process parameter set is as follows: Extracting a finished product model, a part identifier, a supplier code, production batch information and a fault mode code corresponding to each maintenance work order from an after-sale data warehouse; counting the total frequency of occurrence of each fault mode code under the same finished product model by taking the finished product model as a classification dimension, calculating the ratio of the frequency of occurrence of each fault mode code to the total frequency of occurrence of all fault mode codes under the finished product model, and taking the ratio as the reference distribution probability of each fault mode under the finished product model; Combining the component nodes appearing in the same maintenance work order according to the supplier codes and the production batch information based on the coupling association topology to form a component combination identifier, wherein the component combination identifier comprises a plurality of supplier batch information appearing in the same finished product at the same time; counting the occurrence frequency of each fault mode code under the same component combination identifier, calculating the ratio of the occurrence frequency of each fault mode code under the component combination identifier to the number of all maintenance workers under the component combination identifier, and taking the ratio as the actual fault probability under the component combination identifier; calculating the difference value between the actual fault probability of each fault mode under the component combination identifier and the reference distribution probability according to each component combination identifier and the corresponding actual fault probability, and taking the difference value as a deviation value; The maintenance work order corresponding to the component combination identifier with the deviation value exceeding the preset threshold is marked as an abnormal sample, the supplier code and the production batch information are extracted from the replacement component identifier corresponding to the abnormal sample, and the production process parameters corresponding to the production batch are acquired from a supplier management system; And carrying out association integration on the supplier codes, the production batch information and the production process parameters to form a process parameter set corresponding to the abnormal sample.
6. The industrial data management method based on big data according to claim 1, wherein in S4, the specific process of establishing the attribution mapping parameter set is: extracting a finished product model, a fault mode code and a replacement part identifier from a maintenance work sheet corresponding to the abnormal sample; Based on the coupling association topology, performing reverse traversal along the signal transmission relation side and the assembly relation side by taking a fault mode node corresponding to the fault mode code as a starting point to acquire all conduction paths from the fault mode node to each component node, wherein the conduction paths comprise intermediate component nodes which sequentially pass through and signal transmission directions on the paths; extracting supplier codes and production batch information of each component node from the replacement component identifiers corresponding to the abnormal samples for the component nodes on each conducting path, and matching production process parameters corresponding to the production batch information from the process parameter set; And carrying out association combination on the fault mode code, the conduction path identifier, the provider code of each component node on the path, the production batch information and the production process parameters to form an attribution mapping parameter set taking the fault mode-conduction path as an index.
7. The industrial data management method based on big data according to claim 1, wherein in S5, the specific construction process of the causal inference model is as follows: extracting each fault mode code from the attribution mapping parameter set, and conducting path identifiers corresponding to each fault mode code and production process parameters of each supplier batch on the path as an initial sample set; Taking the fault mode codes in the initial sample set as result variables and taking the conducting path identifiers and the production process parameters of all the supplier batches on the path as reason variables; Counting distribution frequencies of different supplier batch combinations under each conducting path identifier, and corresponding fault mode code distribution frequencies of each supplier batch combination; For each conduction path identifier, constructing a structural equation model taking production process parameters of each supplier batch on the path as independent variables and taking a fault mode code corresponding to the path as the dependent variables; Solving path coefficients in the structural equation model through maximum likelihood estimation to obtain causal effect values of production process parameters of each supplier batch on fault mode codes; And summarizing the causal effect values of all the supplier batches under each conducting path identifier to obtain a causal inference model.
8. The industrial data management method based on big data according to claim 1, wherein in S5, the specific process of obtaining the attribution weight of each supplier batch to the failure mode is: extracting causal effect values of each supplier batch to the fault mode under each conducting path identification from the causal inference model; normalizing the causal effect value for each supplier batch under the same conducting path identifier to obtain an initial attribution weight of each supplier batch under the conducting path; counting the occurrence frequency of each supplier batch combination in the abnormal sample under each conduction path identifier and the total frequency of all abnormal samples under the conduction path identifier, and taking the ratio of the occurrence frequency to the total frequency as a confidence coefficient of the supplier batch combination; The product of the initial attribution weight and the confidence coefficient is taken as the final attribution weight of each supplier batch under the conduction path.

Description

Industrial data management method based on big data Technical Field The invention relates to the technical field of industrial data management, in particular to an industrial data management method based on big data. Background With the rapid development of intelligent manufacturing and internet of things technology, the product structure in modern supply chain systems is increasingly complex, and a finished product is often assembled by a plurality of parts from different suppliers. After-sales maintenance data is used as an important information source for connecting the user experience of the terminal and the quality of an upstream supply chain, and contains a large number of fault phenomena, maintenance records and equipment operation logs. How to accurately locate the root of quality problems from these after-market data, identify the specific suppliers and production lots that are causing the failure, becomes a key requirement for enterprises to promote product quality and optimize supply chain management. In the prior art, fault statistics are mainly performed by relying on the identification of replacement parts in after-sales maintenance records, and quality performance of suppliers is evaluated by analyzing the replacement frequency of each part. Such methods typically employ "change-to-change" statistical logic, i.e., a direct determination is made that the supplier of the component is the responsible party for the failure based on the component being replaced in the repair worksheet. However, in practical applications, the failure of the finished product is often caused by interaction of parts of multiple suppliers, for example, the motor burnout may be caused by a process defect of the motor itself, or may be caused by abnormal current output by the controller, so that interaction influence between conductivity of the failure and the parts is ignored. On the other hand, there is a cross-domain gap between the failure phenomena in the after-market repair records and the upstream supplier production process parameters. Repair records typically exist in the form of fault codes or text descriptions, while quality issues for suppliers are manifested in deviations in process parameters for a particular production lot. The prior art lacks an effective means for correlating the after-sales failure phenomenon with the production process parameters of suppliers, so that quality tracing stays at a component level, the production batch and the process parameter dimension are difficult to go deep, and accurate attribution from after-sales data to a supply chain source is difficult to realize. Disclosure of Invention The invention aims to provide an industrial data management method based on big data, which solves the following technical problems: The existing after-sales fault attribution method attributes faults to single suppliers according to replacement part statistics, and coupling correlation of fault modes and multi-supplier batch process parameters is not established, so that fault attribution and supplier process parameter traceable cutting is conducted, faults caused by multi-supplier batch interaction are wrongly attributed to the single suppliers, and quality problem batches cannot be accurately positioned and supply chain quality control fails. The aim of the invention can be achieved by the following technical scheme: an industrial data management method based on big data comprises the following steps: s1, extracting a structured fault code, an unstructured maintenance text, a replacement part identifier and an equipment running log in a historical after-sale maintenance record, and constructing an after-sale data warehouse containing multi-source data; S2, extracting the association relation of the parts and restoring the hierarchy of the finished parts to the after-sales data warehouse to obtain the assembly relation of the parts and a signal transmission path, and constructing a coupling association topology comprising finished nodes, part nodes, supplier nodes and failure mode nodes; S3, screening abnormal samples with deviation from an expected failure mode in the coupling correlation topology from after-sale samples of different batches and working conditions, extracting production process parameters and batch information of suppliers corresponding to the abnormal samples, and integrating to form a process parameter set; s4, deconstructing the coupling association topology, identifying conduction paths of interaction among components in the abnormal sample, and establishing an attribution mapping parameter set by combining the process parameter set; S5, constructing a causal inference model between the attribution mapping parameter set and the fault mode, and obtaining attribution weights of all supplier batches to the fault mode based on the conduction path identification and production process parameters of all supplier batches on the path; and S6, marking the supplier batch and the pr