Control-State Feature Extraction for Low-Latency Abnormality Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional predictive maintenance systems for equipment face delays in detecting abnormalities due to long data collection cycles and processing delays in control apparatuses, which hinder timely detection of equipment issues.
Innovation Solution
A control apparatus that acquires and updates state values for each control period, calculates feature quantities from these values, and uses an abnormality detection unit to generate detection results, ensuring consistent processing from data input to output, and allowing for arbitrary unit section settings and learning data-based abnormality evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data aggregation is performed in a host data logger system, then data integration is achieved, but detection delay increases
Solution Approach 1:
The patent segments the abnormality detection function from the central data logger and distributes it to individual control apparatuses at the source of data generation. Each control apparatus independently performs feature quantity extraction and abnormality detection on its own time series data, eliminating the need to wait for centralized data aggregation and thereby reducing detection delay while maintaining detection accuracy.
Solution Approach 2:
The patent implements preliminary abnormality detection processing directly at the control apparatus before data is sent to the data logger. By extracting feature quantities and detecting abnormalities in advance at the source, the system eliminates the time lag associated with centralized post-processing, achieving timely abnormality detection without sacrificing accuracy.
2Reliability
If abnormality detection is performed using data from multiple control cycles, then detection accuracy improves, but processing delay increases
Solution Approach 1:
The patent maintains continuous abnormality detection processing at each control apparatus by continuously extracting feature quantities from incoming time series data and continuously comparing them against learned patterns. This continuous local processing eliminates idle waiting periods between control cycles while maintaining the use of multi-cycle data for accurate detection.
Solution Approach 2:
The system performs preliminary feature quantity extraction and abnormality assessment at each control cycle before data leaves the control apparatus. This preliminary action allows the system to use data from multiple control cycles for accurate detection without incurring additional processing delays, as the detection work is already done locally and continuously.
3Reliability
If centralized data aggregation is used, then data integration is achieved, but real-time processing capability is reduced
Solution Approach 1:
The patent segments the detection function across multiple independent control apparatuses, each capable of real-time processing of its own data. This segmentation allows parallel processing at the source, dramatically increasing overall processing speed while maintaining detection consistency through standardized feature extraction and comparison methods implemented at each node.
Solution Approach 2:
Each control apparatus performs self-service abnormality detection on its own time series data without requiring centralized processing. This self-service capability enables real-time processing at the source, as each apparatus independently extracts features, compares them against learned patterns, and generates detection results immediately, eliminating the bottleneck of centralized aggregation while maintaining consistent detection standards.
Data Source
AI summary
A control apparatus includes: a management unit that updates internal state values for each control period with a state value generated in a subject to be controlled; a feature quantity extraction unit that calculates a feature quantity from a change of a second internal state value for each unit section determined according to a value of a first internal state value; and an abnormality detection unit that generates detection results indicating whether any abnormality has occurred in a subject to be monitored included in the subject to be controlled, on the basis of the feature quantity. The feature quantity extraction unit and the abnormality detection unit output the values of the first internal state value used to determine the unit section, in correspondence with the corresponding unit section.