Production Facility Log Analysis for Unexpected Abnormality Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing production facility monitoring technologies using sensor data cannot detect unmeasurable items, and those using log data can only detect abnormalities in pre-defined items, missing other potential issues.
Innovation Solution
A production facility monitoring system that determines abnormality levels based on log data by extracting text feature amounts from log data and creating a classification rule using machine learning, allowing for accurate detection of abnormalities without pre-defining target character strings or regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensor data is used for facility monitoring, then measurable parameters can be monitored, but items that cannot be measured by sensors cannot be monitored
Solution Approach 1:
The patent combines sensor data and log data into a unified monitoring system. The log data processing unit extracts features from log files while the sensor data processing unit processes sensor measurements, and both are integrated in the abnormality detection unit to achieve comprehensive monitoring that covers both measurable parameters and unmeasurable items through text analysis
2Reliability
If log data monitoring is used with pre-defined items, then specific abnormalities can be detected, but abnormalities in other items cannot be detected
Solution Approach 1:
The patent implements dynamic abnormality detection by learning normal log patterns during a learning period and then detecting deviations from these patterns. The detection criteria are not fixed but adapt based on learned normal behavior, allowing the system to detect abnormalities in unexpected items while maintaining high accuracy for known issues
Solution Approach 2:
The system uses feedback from learned normal log patterns to continuously improve abnormality detection. The learning unit processes normal log data to establish baseline patterns, and the abnormality detection unit uses these patterns as reference to identify deviations, creating a feedback loop that enhances both detection accuracy and scope
3Measurement precision
If text log analysis is performed without main component extraction, then all log data can be analyzed, but processing complexity increases
Solution Approach 1:
The patent segments log data into main components and other components using main component extraction. This segmentation separates the most significant features from less important details, allowing the system to focus processing on key elements while reducing overall processing complexity without sacrificing detection accuracy
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In order to determine an abnormal degree of a production facility on the basis of log data, a learning unit 44 learns the classification rule that classifies whether the production facility is normal or abnormal from a text feature amount on the basis of the text feature amount (for example, first main component and second main component) obtained from a number of texts included in a plurality of pieces of log data obtained in a predetermined process of the production facility and production history information of the production facility, an extraction unit 46 extracts a text feature amount of the log data to be monitored obtained in the predetermined process of the production facility, and a determination unit 48 determines whether the production facility is normal or abnormal when the log data to be monitored is obtained on the basis of the extracted text feature amount and a classification rule.