Detecting DC Arc Faults in PV Systems: FFT and Wavelet Guide

Introduction#

Series DC arc faults in photovoltaic (PV) systems are among the most challenging safety hazards to detect. Unlike AC arcs, DC arcs do not exhibit natural zero crossings, so they can sustain themselves indefinitely once established. They often occur at loose connectors, corroded terminals, or damaged cable insulation, and they can reach temperatures high enough to ignite surrounding materials. The difficulty is compounded by the fact that the arc's electrical signature is weak and easily buried under the switching noise generated by the inverter's power electronics.

Traditional protection devices such as fuses and circuit breakers respond to overcurrent or ground faults, but a series arc may draw a current lower than the normal operating current, rendering these devices ineffective. This is why arc fault circuit interrupters (AFCIs) and dedicated arc fault detectors have become a focus of PV system design. This guide explains the physical characteristics of DC arcs, the principles of FFT- and wavelet-based detection algorithms, and practical methods to improve detection accuracy and immunity to inverter noise.

Electrical and Spectral Characteristics of DC Arcs#

A series DC arc exhibits a highly non-linear voltage-current relationship. As the arc forms, the voltage across the gap rises while the current may drop slightly, creating a negative differential resistance region. In the time domain, this manifests as a random, broadband high-frequency signal superimposed on the DC operating point. The arc's current waveform contains rapid step changes and bursts of high-frequency energy that are not present in normal PV operation.

In the frequency domain, the arc noise spreads across a wide spectrum, typically from a few kilohertz up to several hundred kilohertz. The inverter, on the other hand, produces switching harmonics at its PWM frequency and multiples thereof, along with ringing from parasitic elements. These inverter-generated components can overlap with the arc's spectral footprint, making simple threshold detection unreliable. Understanding both the arc's broadband, stochastic nature and the inverter's deterministic, periodic noise is essential for designing an effective detector.

FFT-Based Detection: Principles and Limitations#

The Fast Fourier Transform (FFT) converts a time-domain signal into its frequency-domain representation, revealing the amplitude and phase of each frequency component. For arc detection, the current signal is sampled at a rate high enough to capture the arc's high-frequency content, typically 100 kS/s or more. A window function such as Hann or Hamming is applied to reduce spectral leakage, and the FFT is computed over a sliding window.

The detector then examines specific frequency bands where arc noise is prominent but inverter noise is relatively low. A common approach is to compare the energy in these bands against a baseline captured during normal operation. If the energy exceeds a threshold for a sustained period, an arc fault is declared. However, FFT alone has limitations: it assumes the signal is stationary within the window, but arcs are transient and non-stationary. Short arcs may be missed, and inverter noise can cause false alarms if the threshold is set too low.

Wavelet Analysis for Non-Stationary Arc Signals#

Wavelet transform addresses the non-stationary nature of arc faults by providing time-frequency localization. Unlike the FFT, which uses fixed-width basis functions, wavelets use variable-width functions that can capture both slow trends and fast transients. The discrete wavelet transform (DWT) decomposes the current signal into approximation and detail coefficients across multiple scales, effectively separating low-frequency operating current from high-frequency arc noise.

Arc detection using wavelets typically involves computing the energy or entropy of the detail coefficients at selected scales. A sudden increase in high-frequency detail energy indicates an arc event. Because wavelets are localized in time, they can detect short-duration arcs that FFT might miss. Moreover, the multi-resolution property allows the detector to distinguish between the broadband, random arc signature and the narrowband, periodic inverter switching noise, which tends to concentrate at specific scales.

Improving Accuracy and Noise Immunity#

To enhance detection accuracy, a hybrid approach combining FFT and wavelet analysis is often recommended. The FFT can identify the dominant noise frequencies from the inverter, while the wavelet transform isolates transient arc events. Adaptive filtering can then suppress the known inverter noise before the detection stage. For example, a notch filter or an adaptive least-mean-squares (LMS) filter can track and cancel the switching harmonics, improving the signal-to-noise ratio of the arc signature.

Another effective strategy is to use a reference channel. By monitoring the inverter's output current or a dedicated noise sensor, the detector can subtract correlated noise from the PV string current. This common-mode noise cancellation significantly reduces false positives. Additionally, machine learning classifiers trained on features extracted from both FFT and wavelet domains can improve robustness across different inverter models and operating conditions.

Proper sensor placement and shielding also play a critical role. Current transformers or Rogowski coils should be placed close to the PV strings, and signal cables should be twisted and shielded to minimize electromagnetic interference. Implementing a voting scheme with multiple detection algorithms further reduces the likelihood of false trips while maintaining high sensitivity to real arcs.

Integration and Practical Considerations#

Modern PV safety systems increasingly rely on digital signal processing platforms that can run FFT and wavelet algorithms in real time. These platforms often include communication interfaces for remote monitoring and diagnostics. For instance, EJER Tech provides environmental control and storage solutions for sensitive electronics, and its open API supports integration with enterprise ERP systems, enabling centralized management of equipment, data, and digital workflows. While EJER's focus is on dry cabinets, N2 cabinets, and drying ovens for moisture-sensitive devices, the principle of centralized data integration is equally valuable in PV monitoring, where arc fault detectors can feed data into a broader asset management system.

When deploying arc fault detection, consider the sampling rate, processing latency, and memory constraints of the embedded controller. The algorithm should be validated under various conditions, including different irradiance levels, inverter topologies, and cable lengths. Field testing with artificial arc generators is essential to tune thresholds and confirm reliable operation without nuisance tripping.

Conclusion#

Detecting series DC arc faults in PV systems requires a deep understanding of both the arc's electrical behavior and the inverter's noise characteristics. FFT provides a frequency-domain view that is useful for identifying arc noise bands, while wavelet analysis offers superior time-frequency resolution for transient events. By combining these techniques with adaptive noise cancellation, reference-channel subtraction, and robust sensor design, engineers can achieve high detection accuracy and strong immunity to inverter switching noise. As PV systems continue to grow in scale and complexity, advanced arc fault detection will remain a critical component of safe and reliable solar power generation.