Rotary Machine State Monitoring via Segmented Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing state monitoring methods for rotary machines fail to accurately discriminate anomalies caused by operating conditions or noise, leading to erroneous classifications and missed small damages.
Innovation Solution
A state monitoring method that involves obtaining and segmenting measurement data, calculating feature value vectors, and using a classification boundary and anomaly discriminant threshold to determine anomaly scores and ratios, repeatedly processing to improve accuracy and reduce error rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a normal model of physical quantity is prepared and deviation is calculated to monitor state, then anomaly discrimination is achieved, but physical quantity variation in normal state due to operating condition or noise causes erroneous discrimination or missed anomaly detection
Solution Approach 1:
The patent segments the continuous measurement data into multiple time-series data segments, each representing a specific operating condition. By analyzing multiple segments rather than relying on a single normal model, the system can distinguish between normal variations due to operating conditions and actual anomalies, thereby improving reliability without sacrificing measurement precision.
Solution Approach 2:
The patent changes the parameter representation from raw physical quantities to anomaly scores derived from multiple data segments. By transforming the data through statistical analysis across segments and comparing against learned patterns, the system maintains measurement precision while enhancing anomaly discrimination accuracy through parameter transformation.
2Device complexity
If a single normal model is used for anomaly detection, then the monitoring process is simple, but small damages are missed and erroneous discrimination occurs due to operating condition variations or noise
Solution Approach 1:
The patent divides the monitoring approach into multiple data segments representing different operating conditions. This segmentation allows the system to learn normal variations across conditions while detecting anomalies, improving reliability without requiring excessively complex system architecture.
Solution Approach 2:
The patent performs preliminary analysis by collecting and analyzing multiple time-series data segments during normal operation to establish baseline patterns. This preliminary action enables the system to distinguish normal variations from anomalies more reliably, improving detection reliability while maintaining manageable system complexity through advance data preparation.
3Productivity
If measurement data is processed without segmentation, then processing is faster and simpler, but anomaly scores cannot be accurately calculated to distinguish normal variation from actual anomalies
Solution Approach 1:
The patent segments measurement data into multiple time-series segments for parallel processing. This segmentation enables efficient computation of anomaly scores by allowing independent analysis of each segment, maintaining processing productivity while achieving the precision needed to distinguish normal variation from actual anomalies through comparative analysis.
Solution Approach 2:
The patent processes multiple data segments partially (focusing on key features and statistics) rather than performing exhaustive analysis on raw data. This partial action approach maintains processing speed while generating sufficiently accurate anomaly scores for reliable anomaly detection.
Data Source
AI summary
A state monitoring method and a state monitoring apparatus which achieve a lower error rate are provided. Rather than collectively processing whole measurement data, the state monitoring apparatus divides the whole measurement data into a plurality of segments and processes each segment. From the measurement data, training data and test data to which an anomaly detection approach is to be applied is randomly selected. By repeatedly calculating an anomaly ratio and averaging the anomaly ratios, a result of discrimination is stabilized.