Process Model Abnormal Event Detection From Manufacturing Logs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting abnormal events in manufacturing facilities are costly, difficult to implement in real-time, and struggle to identify errors or abnormal patterns in the entire manufacturing process, often leading to incorrect determinations of normal or abnormal states.
Innovation Solution
A method and apparatus that analyze log data to generate a process model, identifying sequentially executed activities and their relationships, and output results in real-time to detect abnormal events by calculating transition ratios, removing unconnected nodes and edges, and setting threshold ranges for activity executions and residence times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If statistical prediction techniques using sensor data and performance-related data are used to predict failure of manufacturing facilities, then detection accuracy is improved, but device complexity and cost increase due to additional sensors and infrastructure investment
Solution Approach 1:
The patent creates a virtual copy of the manufacturing process by building a process model that replicates normal activity sequences and transition ratios from historical log data. This virtual model serves as a reference for detecting abnormalities without requiring additional physical sensors, thus achieving detection accuracy while avoiding infrastructure complexity
Solution Approach 2:
The patent replaces the mechanical sensor-based detection system with an information-processing system that analyzes existing log data. Instead of using physical sensors to collect data, the system substitutes computational analysis of textual/log data to achieve abnormal event detection, eliminating the need for additional hardware infrastructure
2Reliability
If conventional statistical prediction techniques are used to detect abnormal events, then detection capability is improved, but real-time detection is difficult because analysis is performed after failure or abnormal event occurs
Solution Approach 1:
The patent performs preliminary analysis by building a process model from historical log data that captures normal activity sequences and transition ratios before actual operation. This pre-established model enables real-time comparison with current log data, allowing immediate detection of deviations from normal patterns without waiting for failures to occur
Solution Approach 2:
The patent implements a dynamic detection system that continuously compares real-time log data against the established process model. The system dynamically identifies abnormalities by detecting deviations in activity sequences and transition ratios as they occur, enabling real-time response rather than post-failure analysis
3Measurement precision
If conventional methods analyze specific equipment hardware to determine abnormality, then equipment-level detection is improved, but process-level error identification is difficult
Solution Approach 1:
The patent segments the manufacturing process into discrete activities represented as nodes in a process model, with edges representing transition relationships. This segmentation allows the system to identify not only which equipment is abnormal but also the specific process stage and activity where the error occurs, preserving process-level information that would be lost in hardware-only approaches
Solution Approach 2:
The patent introduces a process model as an intermediary layer between raw log data and abnormal event detection. This model acts as a mediator that translates equipment-level log entries into process-level activity sequences, enabling the system to identify both equipment abnormalities and their contextual location within the overall manufacturing process
Data Source
AI summary
A method for detecting an abnormal event performed by a computing device according to an embodiment of the present disclosure includes analyzing log data to identify sequentially executed activities and generating a process model comprising a node indicating each of the activities and an edge indicating an execution predecessor relationship between the activities, and outputting a result of analyzing log data generated in real time based on the process model.