Arcing Hazard Identification Using Machine Learning Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing power distribution systems lack effective high-fidelity sensing infrastructure to detect arcing faults, especially those with low current amplitudes, which often go undetected until they cause significant damage, such as wildfires.
Innovation Solution
An Identification of Arcing Hazards (IAH) system that uses machine learning classifiers to analyze real-time voltage and current measurements from phasor measurement units (PMUs) to identify arcing events, employing unsupervised and supervised algorithms to differentiate between arcing faults and noise in load current, even at low current magnitudes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-fidelity sensing infrastructure is deployed throughout power distribution systems, then arcing fault detection capability is improved, but system cost and complexity increase significantly
Solution Approach 1:
The patent enables existing PMUs to perform arcing fault detection by adding signal processing capabilities to their existing voltage and current measurement functions. The same infrastructure serves both traditional power monitoring and arcing detection purposes, eliminating the need for separate dedicated sensing equipment.
Solution Approach 2:
The invention transforms the approach by changing from direct physical detection of arcing phenomena to indirect detection through analysis of electrical parameter variations (voltage and current waveforms) that existing PMUs already capture. This parameter-based approach leverages existing measurements rather than requiring new sensors.
2Device complexity
If traditional sensing methods are used, then system cost is reduced, but ability to detect low amplitude arcing faults is insufficient
Solution Approach 1:
The system performs preliminary classification of waveform data to identify candidate arcing events before detailed analysis. This two-stage approach allows the system to focus computational resources on promising candidates while maintaining high detection sensitivity for low-amplitude faults using existing infrastructure.
Solution Approach 2:
The patent introduces machine learning classifiers as an intermediary between raw PMU measurements and arcing fault detection. These classifiers act as a bridge that extracts subtle arcing signatures from noisy electrical measurements, enabling detection of low-amplitude faults without requiring enhanced physical sensors.
3Measurement precision
If machine learning classifiers are implemented, then arcing fault identification accuracy is improved, but computational processing requirements increase
Solution Approach 1:
The patent segments the detection process into distinct stages: initial candidate identification from waveform data, classification of candidates using machine learning, and final arcing event determination. This segmentation allows computational resources to be distributed efficiently across processing stages rather than concentrated in a single high-complexity step.
Solution Approach 2:
The system applies machine learning classifiers selectively to candidate events identified through preliminary analysis, rather than processing all waveform data through full ML pipelines. This partial application of computational intensity maintains high accuracy for critical detections while reducing overall processing burden.
Data Source
AI summary
A system to enable identification of arcing hazards comprises a data storage to store a set of measurements acquired by measurement units of a power distribution system. The system further comprises at least one processor configured to identify candidate arcing events represented by the measurements by using an unsupervised machine learning process, and to train a supervised machine learning classifier for automatic real-time identification of arcing events, by using labeled training data based on the identified candidate arcing events.