Signal Processing System for Oscillation Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing signal processing systems for oscillation signals, such as those used in wire rope flaw detection, do not effectively account for temporal variations in amplitude, leading to suboptimal performance in anomaly detection.
Innovation Solution
The proposed method separates the signal into an oscillation signal with constant amplitude and a signal with temporal variation in amplitude, applying dimensionality reduction and compression techniques to extract feature values, and then restores the signals using inverse processing, allowing for more accurate machine learning-based anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a predetermined frequency component of the sensor signal is used as input for machine learning, then the processing is simplified, but temporal variation in amplitude is not taken into consideration, leading to reduced detection accuracy
Solution Approach 1:
The patent segments the signal processing into distinct components: extracting both frequency components and temporal amplitude variations separately, then combining them as multi-dimensional inputs for the machine learning model. This segmentation allows each feature to be processed optimally while maintaining overall detection accuracy.
Solution Approach 2:
The patent adds temporal amplitude variation as an additional dimension to the traditional frequency-component-only input. By incorporating time-series amplitude information alongside frequency features, the input space is expanded from one dimension to multiple dimensions, enabling the machine learning model to capture both spectral and temporal characteristics of anomalies.
2Productivity
If only frequency components are extracted for anomaly detection, then the processing is faster, but information about temporal variation in amplitude is lost
Solution Approach 1:
The patent extracts temporal amplitude variation information separately from the frequency components using dedicated processing steps. By taking out this specific feature and preserving it as a distinct input element for the machine learning model, the system prevents information loss while maintaining efficient processing through specialized extraction methods.
Solution Approach 2:
The patent performs preliminary extraction and preparation of temporal amplitude features before feeding them into the machine learning model. This preliminary action ensures that amplitude information is captured and formatted appropriately in advance, preventing information loss during the main detection process while maintaining processing efficiency.
3Loss of information
If the entire signal is used for machine learning input, then all information is preserved, but the processing complexity and computational load increase significantly
Solution Approach 1:
The patent segments the raw signal into distinct feature components (frequency components and temporal amplitude variations) before inputting to the machine learning model. This segmentation preserves all relevant information while organizing it into structured, processed features that reduce computational complexity during training and inference.
Solution Approach 2:
The patent transforms the raw signal into different parameter representations (frequency domain features and temporal amplitude features) that are more suitable for machine learning processing. This parameter transformation preserves the essential information content while reducing the dimensional complexity and improving computational efficiency.
Data Source
AI summary
A highly accurate feature extraction is performed on a signal with temporal variation in amplitude, and this signal is restored to detect a state of a transmission source (output source) of this signal to be normal or abnormal. A signal processing method includes: separating a signal X into an oscillation signal with a constant amplitude X1 and a signal with temporal variation in amplitude X2, the separating performed by a signal separator; performing processing of dimensionality reduction, compression, or the like, on the oscillation signal X1 so as to extract a feature value (information) included in the oscillation signal X1; and outputting a restored signal X1′ that is restored from the oscillation signal X1 by performing processing inverse to the processing of dimensionality reduction, compression, or the like, based on the extracted feature value, performing the processing, the inverse processing, and the outputting performed by a signal X1 restorer.