Partial Discharge Signal Separation via FFT Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for separating partial discharge and noise signals in electrical systems are inefficient, often relying on heuristic techniques that lose waveform information and fail to accurately identify damage, leading to potential equipment damage and workplace hazards.
Innovation Solution
A system and method using arithmetic coding in the time domain and magnitude distribution in the frequency domain, which compresses signals without losing waveform information, generates features through histogram binning and Fast Fourier Transform, and applies clustering algorithms to distinguish partial discharge from noise signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If heuristic techniques in time domain are used to engineer statistical features for separation, then the separation process is simple, but waveform information is lost and separation accuracy deteriorates
Solution Approach 1:
The patent transforms the signal separation problem from time domain to frequency domain by applying Fast Fourier Transform (FFT). This dimensional change allows the system to analyze magnitude distributions across different frequency components, preserving waveform information while enabling effective separation of partial discharge signals from noise through frequency-based clustering.
Solution Approach 2:
The patent changes the parameter space by using magnitude distributions in frequency domain instead of statistical features in time domain. By computing the magnitude spectrum and analyzing the distribution of magnitudes across frequency bins, the system preserves the original waveform characteristics while creating a new feature space that enables accurate signal separation through clustering algorithms.
2Power
If heuristic techniques with statistical features are used, then the processing is computationally simple, but separation accuracy deteriorates
Solution Approach 1:
The patent applies Fast Fourier Transform to convert time domain signals into frequency domain representations. This transformation enables the system to capture waveform information through magnitude distributions across frequency components, significantly improving separation accuracy while the computational cost remains manageable through efficient FFT algorithms and clustering.
3Measurement precision
If arithmetic coding in time domain and magnitude distribution in frequency domain are used, then separation accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the signal processing into distinct stages: time domain processing using arithmetic coding to compress and preserve waveform information, followed by frequency domain transformation using FFT to obtain magnitude distributions. This segmentation allows each processing stage to be optimized independently, managing overall system complexity while achieving high separation accuracy.
Solution Approach 2:
The patent employs clustering algorithms that can operate on various types of feature representations. The same clustering framework works with both time domain features and frequency domain magnitude distributions, providing a universal approach that handles the increased complexity through a unified, multi-functional processing architecture.
Data Source
Figure 1
Figure 2
Figure 3.1~3.2
AI summary
This invention relates to a system and method for separating partial discharge and noise signals from digital signals. The system comprises a data collecting module configured to recording and translating electromagnetic signals measured from a sensor to digital signals and a data processing module configured to: receive digital signals from the data collecting module; generate a feature from a time domain for each of the digital signals; generate a plurality of features from a frequency domain for each of the digital signals; apply clustering algorithm on the generated features for all the digital signals to identify a plurality of distinct clusters; and display each distinct cluster on a Phase-Resolved Partial Discharge (PRPD) chart.