Anomaly Detection Using Regression Model Difference Components
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting anomalies in periodically operating bodies, such as bearings, face challenges in distinguishing normal from abnormal operations due to changing signal strengths, individual differences, and noise components, making it difficult to accurately detect anomalies through spectral analysis.
Innovation Solution
A detecting apparatus comprising a processor connected to a dividing unit, a model learning unit, a difference calculating unit, and an anomaly detecting unit, which divides data sequences, learns regression models, calculates difference components, and detects anomalies based on these components, effectively isolating non-periodic components indicative of anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If spectral analysis of predetermined frequency is used to detect anomalies, then the detection method is simple, but the detection accuracy deteriorates due to individual differences and changing signal characteristics
Solution Approach 1:
The system performs preliminary learning to create a model of normal operation characteristics before actual anomaly detection. The model learning unit stores the relationship between operation amounts and vibration characteristics during normal operation, which is then used as a reference for detecting anomalies. This preliminary action enables accurate detection despite individual differences and changing conditions.
Solution Approach 2:
The system changes from fixed frequency spectral analysis to a model-based approach that adapts to varying operation conditions. By using a learned model that accounts for individual differences and changing signal characteristics, the system achieves accurate anomaly detection across different operating states without being constrained to predetermined frequencies.
2Ease of operation
If noise component increase is used to detect anomalies, then the detection approach is straightforward, but the detection accuracy deteriorates because normal operation signals also contain noise components
Solution Approach 1:
The system extracts the periodic component from the vibration signal using the learned model, separating it from the noise component. By taking out the periodic component that represents normal operation, the system can then detect anomalies as deviations from this extracted periodic pattern, rather than relying on overall noise level increases that confound normal and abnormal operations.
Solution Approach 2:
The learned model acts as an intermediary that mediates between the raw vibration signal and anomaly detection. The model provides a reference of normal periodic behavior, enabling the system to distinguish between noise inherent in normal operation and noise indicating anomalies, thereby improving detection accuracy while maintaining simplicity.
3Device complexity
If the entire data sequence is analyzed at once, then the processing is simple, but the detection accuracy deteriorates due to changes in signal strength and frequency over time
Solution Approach 1:
The system segments the data sequence into multiple divided data sequences in the time direction, analyzing each segment separately with the model learning unit. This segmentation allows the system to account for changes in signal strength and frequency over time by creating time-specific models, thereby improving detection accuracy while managing processing complexity through systematic division of the analysis task.
Data Source
AI summary
Provided is a detecting apparatus for detecting one or more anomalies of an operating body. The detecting apparatus includes a processor communicatively coupled to a dividing unit, a model learning unit, a difference calculating unit, and an anomaly detecting unit. The dividing unit divides a data sequence corresponding to an operation of the operating body into a plurality of divided data sequences in a time direction. The model learning unit learns each of the plurality of divided data sequences according to a regression model in the time direction, and calculates a model component modeling each of the divided data sequences. The difference calculating unit calculates a difference component indicating a difference between each of the plurality of divided data sequences and their corresponding model component. The anomaly detecting unit detects one or more anomalies of the operating body based on the calculated difference components.