Abnormality predictor diagnosis system and abnormality predictor diagnosis method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality prediction systems in mechanical facilities, such as chemical and pharmaceutical plants, face challenges in accurately diagnosing rapidly varying and gently varying waveforms, leading to inadequate diagnostic accuracy, and neural network-based systems face computational inefficiencies when reflecting time-series waveforms in normal patterns.
Innovation Solution
An abnormality predictor diagnosis system that learns a normal waveform model using sensor data and identifies abnormality predictors by converting detection values and linear function values into two-dimensional feature vectors, allowing for precise clustering and diagnosis based on predetermined times, thereby distinguishing between waveform types and improving diagnostic accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clusters are collectively learned in each time slot based on detection values only, then the learning process is simple, but rapidly varying and gently varying waveforms cannot be distinguished leading to low diagnostic accuracy
Solution Approach 1:
The patent transforms the one-dimensional detection value data into two-dimensional feature vectors by incorporating time information as a second dimension. Specifically, it extracts both the detection value and the rate of change (time derivative) to form feature vectors (detection value, rate of change), thereby adding temporal dimensionality to distinguish between different waveform characteristics while maintaining cluster-based simplicity
Solution Approach 2:
The patent changes the parameters used for clustering from simple detection values to composite features including detection values and their rates of change. By transforming the feature space to include temporal derivatives, the system can differentiate between rapidly varying and gently varying waveforms that would otherwise be indistinguishable using detection values alone
2Measurement precision
If image data is obtained with 15-minute intervals for learning, then data acquisition is simple, but the temperature distribution variations every moment cannot be accurately reflected leading to low diagnostic accuracy
Solution Approach 1:
The patent extracts only the essential temporal characteristics (rate of change) from the time-series data rather than processing the entire waveform. By focusing on the derivative information and key feature points, it captures the essential dynamics of temperature changes without requiring computation on every momentary variation, thus maintaining accuracy while improving efficiency
Solution Approach 2:
The patent uses partial action by selecting specific critical features (detection values and rates of change at key time points) rather than processing complete continuous waveforms. This partial extraction of temporal information is sufficient for accurate diagnosis while significantly reducing computational burden compared to full waveform analysis
3Measurement precision
If the waveform of time-series data is not distinguished, then the diagnosis process is simple, but important abnormality predictors are missed leading to low diagnostic accuracy
Solution Approach 1:
The patent segments the waveform analysis by dividing it into discrete feature extractions at specific time points. Instead of analyzing the continuous waveform holistically, it extracts features (detection values and rates of change) at segmented intervals and processes them as discrete feature vectors, simplifying the complexity while capturing essential waveform characteristics for accurate diagnosis
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An abnormality predictor diagnosis system includes: a sensor data acquisition means that acquires sensor data; a learning means that identifies a detection value of a sensor when a predetermined time has passed since start of an operation process, identifies a value of a predetermined function when the predetermined time has passed since the start of the operation process using the predetermined function that outputs different values for respective times elapsed as time passes, and learns a normal model of the waveform based on the identified detection value and the value of the function; and a diagnosis means that, in a time-series waveform of sensor data as a diagnosis target, diagnoses the mechanical facility for presence of an abnormality predictor based on comparison of the detection value of the sensor and the value of the function when the predetermined time has passed since the start of the operation process, with the normal model. The abnormality predictor diagnosis system can diagnose the mechanical facility for the presence of an abnormality predictor with high accuracy.