Waveform Anomaly Detection Using Odd-Even Harmonic CCA
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In unsupervised learning approaches for anomaly detection in electric current waveforms, feature transformation often fails to create well-separated clusters between normal and anomalous signals, leading to unclear or poorly defined decision boundaries.
Innovation Solution
An anomaly detection apparatus and method that involves obtaining waveform data, calculating frequency spectra, separating odd and even frequency harmonics, performing canonical correlation analysis (CCA) on these harmonics, and applying linear transformation to generate a model for identifying normal and anomalous signals through magnitude quantization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If unsupervised clustering scheme based on feature transformation is used, then labels or information to estimate decision boundary are not required, but the decision boundary becomes unknown or not well-separated between normal and anomaly clusters
Solution Approach 1:
The patent introduces an intermediary component - the separation unit that specifically processes frequency harmonics to enhance cluster separation. This unit acts as a mediator between the raw frequency spectrum and the clustering algorithm, applying odd/even harmonic separation to create better-defined decision boundaries without requiring labeled data.
Solution Approach 2:
The patent changes the parameter representation by separating frequency harmonics into odd and even components. This parameter transformation creates more discriminative features that improve cluster separation, allowing the unsupervised method to achieve better decision boundary definition while maintaining its label-free advantage.
2Device complexity
If feature transformation is applied to waveform data, then dimension reduction and feature extraction are achieved, but discriminative features between normal and anomaly waveforms are not created
Solution Approach 1:
The patent segments the frequency spectrum into odd and even harmonic components. This segmentation creates distinct feature groups that are more discriminative for anomaly detection. By separating these components and processing them through CCA, the method maintains dimension reduction while improving the reliability of anomaly detection through better feature discrimination.
Solution Approach 2:
The patent transforms the feature space by introducing a new dimensional structure through odd/even harmonic separation and CCA projection. This dimensional transformation creates a more effective feature representation that improves anomaly detection reliability while managing feature dimension complexity through structured transformation rather than simple reduction.
3Measurement precision
If canonical correlation analysis and odd/even harmonic separation are applied, then well-separated clusters are created, but processing complexity increases
Solution Approach 1:
The patent applies preliminary action by performing odd/even harmonic separation before the main clustering process. This pre-processing step organizes the frequency data in a way that facilitates better cluster separation during subsequent CCA and clustering operations, achieving high measurement precision while managing overall processing complexity through structured preprocessing.
Data Source
AI summary
Provided a method comprising: obtaining waveform data sets of a periodic electric waveform signal, with a length set to one cycle time; calculating a frequency spectrum for each waveform data set; extracting and separating odd and even frequency harmonics to create odd and even frequency harmonic matrices on which a canonical correlation analysis (CCA) being applied to obtain CCA features; performing linear transformation on the CCA features to obtain linear transformed features; generating a model based on the linear transformed features; performing magnitude quantization of frequency spectrums of waveform data sets to identify normal and anomalous waveform signals.