Abnormality Prediction Using Waveform Feature Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current abnormality prediction systems in mechanical facilities, such as chemical and pharmaceutical plants, face challenges in accurately distinguishing between rapidly and gently varying waveforms in time-series data, leading to diagnostic inaccuracies, and existing neural network-based methods require excessive computation to accurately reflect varying temperature distributions.
Innovation Solution
The system extracts feature points from sensor data, representing them as multi-dimensional feature vectors and clusters these vectors to learn a normal waveform model, allowing for precise diagnosis of abnormality predictors by comparing new data against learned clusters, reducing computational complexity and improving diagnostic accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clusters are collectively learned in each time slot using conventional clustering methods, then the diagnostic system can process time-series data, but it cannot distinguish between rapidly varying and gently varying waveforms, leading to diagnostic inaccuracy
Solution Approach 1:
The patent segments the time-series data by extracting feature points (peaks, valleys, inflection points) from the waveform, converting continuous data into discrete characteristic points. This segmentation allows the system to capture waveform shape information while reducing data volume, enabling accurate distinction between rapidly and gently varying waveforms without excessive computational complexity
Solution Approach 2:
The patent transforms the original time-series data into a different parameter space by extracting waveform features (amplitude, time position, curvature) and representing them as feature vectors. This parameter transformation enables the clustering algorithm to effectively distinguish between different waveform types based on their characteristic features rather than raw time-series values
2Measurement precision
If neural networks are used to learn normal patterns of temperature distribution, then diagnostic accuracy can be improved, but the amount of computation becomes huge
Solution Approach 1:
The patent extracts only the essential waveform feature points (peaks, valleys, inflection points) from the complete time-series data, discarding redundant information. This extraction approach retains the critical diagnostic information needed to distinguish abnormal waveforms while significantly reducing the computational energy required for pattern recognition compared to processing complete temperature distribution data
Solution Approach 2:
Instead of using comprehensive neural network models that process all temperature data points, the patent applies partial action by focusing only on the most diagnostically relevant feature points. This selective approach achieves sufficient diagnostic accuracy with much lower computational energy requirements
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An abnormality predictor diagnosis system 1 includes: a sensor data acquisition means 12 that acquires sensor data including a detection value of a sensor installed in a mechanical facility 2; a learning means that sets a learning target of sensor data in a period in which the mechanical facility 2 is known to be normal, and learns a time-series waveform of the sensor data as a normal model; and a diagnosis means that diagnoses the mechanical facility 2 for the presence of an abnormality predictor based on comparison between the normal model and the time-series waveform of the sensor data of a diagnosis target. The abnormality predictor diagnosis system 1 can diagnose the mechanical facility for the presence of an abnormality predictor with high accuracy.