CNN Arc Fault Interrupter for Series Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing arc fault detection methods in photovoltaic systems are challenged by the stochastic nature of arc faults, noise interference, and the inability of overcurrent protection devices to detect series arc faults, leading to inaccurate detection and potential fire hazards.
Innovation Solution
A lightweight convolutional neural network (CNN) model, pretrained using a knowledge distillation-based teacher-student approach, is integrated with a solid-state circuit breaker to accurately detect series arc faults by processing electrical current data cyclically, tripping the circuit only after detecting an arc fault for a predetermined number of consecutive cycles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional overcurrent protection devices are used to detect arc faults, then the system can identify high current parallel arc faults, but the devices fail to detect series arc faults due to added series impedance reducing load current
Solution Approach 1:
The patent replaces traditional electromagnetic overcurrent protection devices with a solid-state circuit breaker that uses a convolutional neural network for intelligent arc fault detection. The CNN analyzes current waveforms and identifies both series and parallel arc faults through pattern recognition, overcoming the limitations of conventional thermal-magnetic trip units that cannot detect series arcs.
Solution Approach 2:
The system changes the detection parameter from simple overcurrent threshold to complex waveform analysis using CNN. The neural network processes current signal characteristics including harmonics, transient patterns, and waveform morphology to detect arc faults across different current levels, enabling detection of both series and parallel arc conditions.
2Measurement precision
If arc fault detection sensitivity is increased to detect all arc faults, then detection accuracy improves, but false trips increase due to noise interference from inverter switching and other sources
Solution Approach 1:
The CNN model uses feedback from continuous waveform analysis to distinguish true arc faults from noise. The network learns from training data to recognize genuine arc patterns versus normal operational variations, adjusting its detection criteria dynamically to maintain high accuracy while minimizing false trips through pattern recognition feedback.
Solution Approach 2:
The system performs preliminary training of the CNN model with extensive arc and non-arc data before deployment. This preliminary action embeds knowledge of noise patterns and arc characteristics into the network, enabling it to differentiate between harmful arcs and benign operational variations during actual operation, thus reducing false trips.
3Measurement precision
If complex detection algorithms are used to improve arc fault detection accuracy, then detection precision improves, but computational complexity and processing time increase
Solution Approach 1:
The patent replaces complex manual signal processing algorithms with a trained convolutional neural network that automatically learns optimal detection features. The CNN architecture with convolutional layers, pooling operations, and fully connected layers provides high detection precision while the trained model efficiently processes waveforms without requiring complex real-time computational operations.
4Speed
If the circuit breaker trips immediately upon detecting an arc fault, then response time is minimized, but false trips cause unnecessary system interruptions
Solution Approach 1:
The system uses feedback from the CNN's confidence assessment and multi-cycle verification before tripping. When an arc pattern is detected, the network evaluates the certainty of detection and may require confirmation across multiple waveform cycles before initiating a trip, providing feedback-based verification that reduces false trips while maintaining rapid response to genuine faults.
Data Source
AI summary
An apparatus includes a solid state circuit breaker selectively configured as a closed switch or an open switch. A current sensor is coupled to a current path passing through the closed switch and configured to sense an electrical current flowing on the current path while the switch is closed. A student convolutional neural network (CNN) model is pretrained using a knowledge distillation-based teacher-student approach. The student CNN model is coupled to the solid state circuit breaker and the current sensor and configured to process data representative of the electrical current, the data being processed cyclically with a period defined by an arc fault detection cycle. According to some aspects, in response to the student CNN model detecting, in association with the electrical current, an arc fault lasting a predetermined number of consecutive arc fault detection cycles, the solid state circuit breaker is reconfigured as the open switch.