Process State Monitoring via Signal Separation and Anomaly Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current process state monitoring methods, such as model-based and database approaches, struggle to accurately identify abnormal states in manufacturing processes due to varying equipment conditions and numerous operation patterns, leading to limitations in detecting equipment deterioration, sudden failures, and noise-related issues.
Innovation Solution
A process state monitoring device and method that calculates deviation indexes from time-series data, separates components using filtering techniques, computes characteristic values, and uses machine learning to classify states, enabling detailed identification of process conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of patterns is increased to cover all operation patterns in manufacturing processes with vast variations, then the detection coverage is improved, but the system complexity and computational burden increase significantly
Solution Approach 1:
The patent segments the vast number of operation patterns into a limited set of representative patterns based on clustering analysis. Instead of creating separate patterns for every possible operation condition, the system divides patterns into clusters and selects representative patterns from each cluster, thereby reducing the total number of patterns while maintaining comprehensive detection coverage across all operation variations.
Solution Approach 2:
The patent changes the parameter of pattern quantity from a large fixed number to a dynamically determined optimal number based on clustering results. By using clustering algorithms to analyze operation data and determine the appropriate number of clusters, the system adapts the pattern quantity to match the actual operational diversity, avoiding both over-proliferation and under-coverage of patterns.
2Ease of operation
If traditional monitoring methods are used without considering occurrence forms of deviation indexes, then the monitoring process is simple, but the ability to identify specific abnormal states and distinguish between different types of troubles is insufficient
Solution Approach 1:
The patent introduces dynamic analysis by examining the occurrence form and temporal characteristics of deviation indexes. Instead of treating all deviations uniformly, the system dynamically adapts its diagnostic approach based on how the deviation index changes over time, allowing it to distinguish between gradual deterioration, sudden failures, and noise-related issues while maintaining operational simplicity through automated dynamic classification.
Data Source
Figure 1
Figure 2~2(d)
AI summary
A process state monitoring device according to the present invention includes: a component separation computation unit that separates two or more components from time series data of a value indicating a state of a process; a characteristic value computation unit that computes a characteristic value from each component separated by the component separation unit; and a state determination unit that classifies the state of the process on the basis of the characteristic values computed by the characteristic value computation unit. With this structure, the state of the process can be identified in detail. The process state monitoring device may include a similar case search unit that searches a past database for an operation case similar to the process on the basis of the characteristic values computed by the characteristic value computation unit.