Process State Monitoring via Component Separation and Case Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current process state monitoring methods, such as the model base and database approaches, struggle to accurately diagnose abnormal states in manufacturing processes like steel production due to varying equipment conditions and numerous operation patterns, leading to incomplete identification of trouble forms, especially when equipment deteriorates or sudden failures occur.
Innovation Solution
A process state monitoring device that separates components from time series data using filtering, computes characteristic values such as statistics and ratios, and employs machine learning for state classification, along with a similar case search to identify process states by analyzing deviation indexes and their occurrence forms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the database approach uses past trouble cases for diagnosis, then the diagnosis method can handle known abnormal patterns, but it cannot diagnose unprecedented troubles or sudden failures that differ from historical data
Solution Approach 1:
The system performs preliminary action by continuously monitoring and detecting unusual situations during normal operation before they develop into actual troubles. The unusual situation detection unit identifies deviations from normal patterns in real-time, enabling early intervention and prevention of future failures rather than relying solely on historical trouble data.
Solution Approach 2:
The diagnosis system is segmented into multiple functional units: unusual situation detection unit, deviation index calculation unit, component separation computation unit, and state determination unit. This segmentation allows each unit to specialize in specific aspects of monitoring and analysis, improving both the detection of novel abnormalities and the overall reliability of the diagnosis system.
2Measurement precision
If the system monitors all deviation indexes thoroughly, then all types of troubles can be detected, but the complexity of analysis increases when different trouble forms produce different deviation patterns
Solution Approach 1:
The deviation index is segmented into multiple frequency components using the component separation computation unit, which applies fast Fourier transform to divide the composite deviation signal into individual frequency components. This segmentation enables precise identification of specific trouble types by analyzing each component's characteristics without being overwhelmed by the complexity of the overall signal.
Solution Approach 2:
Instead of analyzing all possible deviation patterns simultaneously, the system focuses on detecting unusual situations by comparing current operation against normal patterns. It applies partial action by concentrating computational resources on identifying deviations from normality rather than exhaustively analyzing every possible trouble mode, thereby reducing analysis complexity while maintaining detection precision.
3Measurement precision
If the system stores and analyzes detailed operation data for every batch process, then accurate state classification is achieved, but the data storage and processing burden increases significantly
Solution Approach 1:
The system extracts only the essential characteristics needed for state classification by calculating deviation indexes that represent the difference between actual operation and normal patterns. Instead of storing and analyzing all raw operation data, it extracts and stores only the computed deviation indexes and their frequency components, significantly reducing data volume while preserving the information necessary for accurate state classification.
Solution Approach 2:
The system transforms raw operation data into different parameter representations through computational processing. It converts time-series operation data into deviation indexes, then further transforms these into frequency domain components using Fourier transform. These parameter changes enable accurate state classification with compact data representations, reducing the storage burden while maintaining classification accuracy.
Data Source
AI summary
A process state monitoring device incudes a processor including hardware, the processor being configured to: separate two or more components from time series data of a value indicating a state of a process; compute a characteristic value from each of the separated components; and classify the state of the process on the basis of the computed characteristic values.