Unsupervised Transformer Signal Separation via Blind Source Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ambient vibration data collected from aerial fiber cables using Distributed Acoustic Sensing (DAS) are mixed with multiple signals, making it challenging to extract useful information such as transformer vibration patterns, health, and position without supervised training data.
Innovation Solution
The method employs unsupervised blind source separation using a deep neural network trained with an unsupervised loss computed between estimated source signals and mixtures, allowing for the separation of transformer signals from environmental noises without ground-truth labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If ambient vibration data is collected from aerial fiber cables using DAS, then vibration data can be obtained for environmental monitoring, but the data becomes a mixture of multiple signals making it difficult to extract useful information
Solution Approach 1:
The patent applies segmentation by dividing the mixed vibration signal into multiple separate source signals using blind source separation (BSS) techniques. The DAS data, which contains mixed signals from transformers, traffic, and environmental noises, is segmented into individual source components through mathematical decomposition methods such as Independent Component Analysis (ICA) or Non-negative Matrix Factorization (NMF), enabling extraction of useful transformer information from the composite signal.
Solution Approach 2:
The patent extracts useful transformer vibration patterns from the mixed ambient data by applying source separation algorithms that identify and extract specific source signals. The system extracts transformer-related signals while filtering out unwanted components like traffic vibrations and environmental noises, obtaining the desired information without physical separation of the sources.
2Measurement precision
If traditional signal processing methods are used to separate mixed signals, then some separation can be achieved, but supervised training data is required which is difficult to obtain
Solution Approach 1:
The patent implements self-service by using unsupervised blind source separation methods that do not require external training data or ground truth labels. The algorithm automatically learns to separate mixed signals by exploiting statistical properties of the sources, such as independence or non-negativity constraints, enabling the system to perform signal separation autonomously without manual supervision or pre-labeled datasets.
Solution Approach 2:
The patent changes the approach from supervised learning (requiring labeled data) to unsupervised learning by modifying the training objective. Instead of minimizing error against known labels, the system optimizes separation based on statistical independence criteria or other unsupervised metrics, fundamentally changing the parameter space from labeled accuracy to statistical separation quality.
3Loss of information
If unsupervised blind source separation is applied to separate transformer signals from environmental noises, then useful information can be extracted without ground-truth labeling, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by focusing the blind source separation on extracting only the most relevant source signals (e.g., transformer vibrations) rather than attempting to separate all possible sources in the mixture. This selective approach reduces computational complexity by limiting the separation task to essential components while still achieving the goal of extracting useful transformer information.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach effectively decomposes complex vibration patterns into individual source signals, enabling accurate analysis of transformer health, position, and power outages, even in noisy and entangled signal environments.
Implementation Method 1
separating the mixture of mixtures into a plurality of estimated source signals using a separation model, wherein the separation model is trained using an unsupervised loss computed between the estimated source signals and the at least two mixtures
Data Source
AI summary
Systems and methods include collecting vibration data along an optical fiber cable using distributed acoustic sensing (DAS). The collected vibration data is preprocessed to separate the vibration data into at least two mixtures. The at least two mixtures are combined into a mixture of mixtures. The mixture of mixtures is separated into a plurality of estimated source signals using a separation model. The separation model is trained using an unsupervised loss computed between the estimated source signals and the at least two mixtures.