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

Introduction#

Series DC arc faults are among the most dangerous failure modes in photovoltaic (PV) systems. Unlike a parallel fault, a series arc develops in a current-carrying conductor that has been loosened, corroded, or mechanically damaged, so the fault current stays close to the normal operating current. Conventional overcurrent protection therefore cannot see it, and the arc can persist for hours, heating connectors and creating a genuine fire risk. Detecting these faults reliably is difficult because the arc signature is weak and easily buried in the switching noise produced by the inverter.

The detection problem is fundamentally a signal-processing problem. An arc has a distinctive voltage-current (V-I) characteristic and a broadband high-frequency noise spectrum, while the inverter contributes periodic switching components at known frequencies. Effective detectors combine time-domain features with spectral analysis, using the Fast Fourier Transform (FFT) for steady-state spectral fingerprints and wavelet analysis for transient, non-stationary events. This guide explains the underlying principles and offers practical solutions for improving detection accuracy and noise immunity.

The Arc Signature: V-I Behavior and High-Frequency Noise#

A stable series arc exhibits a characteristic V-I curve. As current rises, the arc voltage drops slightly and then flattens, producing a negative-resistance region that differs from the near-linear behavior of a healthy resistive load. In the time domain, the arc appears as a random, high-frequency fluctuation superimposed on the DC operating point, with intermittent reignition spikes as the arc restrikes each half-cycle of any ripple present.

In the frequency domain, the arc behaves like a broadband noise source. Its energy spreads across tens of kilohertz up to several megahertz, with no single dominant tone. This is the key discriminator: inverter switching noise is concentrated at the switching frequency and its harmonics, whereas arc noise is spread across the spectrum. A detector that measures broadband energy while rejecting narrowband switching lines can separate the two sources far more reliably than a simple amplitude threshold.

FFT-Based Detection: Spectral Fingerprints#

The FFT converts a windowed time-domain sample into a frequency spectrum, revealing which frequencies carry energy. For arc detection, the algorithm typically computes the spectrum of the DC current or voltage, then compares energy in arc-sensitive bands against energy at known switching frequencies. A healthy system shows sharp peaks at the switching frequency and its harmonics with low inter-harmonic noise; an arcing system shows elevated broadband energy between those peaks.

A practical FFT pipeline includes windowing to reduce spectral leakage, averaging across multiple frames to stabilize the estimate, and a band-energy ratio that normalizes arc-band energy against total energy. This ratio makes the detector less sensitive to load changes, since a rise in overall current does not automatically trigger an alarm. The main limitation is that the FFT assumes the signal is stationary within each window, so short arc transients can be smeared or missed entirely.

Wavelet Analysis: Capturing Non-Stationary Arcs#

Wavelet transforms decompose a signal into scaled and shifted versions of a mother wavelet, providing simultaneous time and frequency localization. This suits arc detection well because arcs are transient: they start abruptly, flicker, and extinguish. A discrete wavelet transform splits the signal into approximation and detail coefficients across multiple levels, and arc events appear as sharp spikes in the high-frequency detail coefficients.

Detection logic based on wavelets typically tracks the energy of detail coefficients over a sliding window and flags a fault when that energy exceeds a threshold for a sustained period. Because the method responds to sudden changes rather than average spectral content, it catches intermittent arcs that FFT-based methods may average away. Combining both approaches gives a robust detector: FFT confirms the broadband spectral signature, while wavelets confirm the transient timing and duration.

Improving Accuracy and Noise Immunity#

Several engineering measures raise detection accuracy in real installations. First, synchronize sampling with the inverter switching period so that switching harmonics fall into predictable bins that can be notched out. Second, apply adaptive thresholding based on the measured noise floor rather than fixed limits, so the detector adjusts to different inverters and operating conditions. Third, use multi-feature voting, requiring agreement between spectral, wavelet, and time-domain indicators before tripping, which sharply reduces false alarms.

Hardware choices matter as much as algorithms. High-bandwidth current sensors with good common-mode rejection, shielded twisted-pair routing, and careful grounding keep switching noise out of the measurement chain. Filtering should be applied selectively so that arc-band energy is preserved while switching lines are attenuated. Finally, validation should include nuisance sources such as motor starts, grid transients, and dust on modules, because a detector that trips on these events will be disabled by operators and provide no real protection.

Digital Integration and Data Management#

Modern arc detection is increasingly part of a connected monitoring architecture. Detector nodes stream spectral features and event logs to a central platform, where trends across strings and sites can be analyzed. In this context, EJER provides an open API that supports integration with enterprise ERP systems, enabling centralized management of devices, data, and workflows across the full chain. For PV safety teams, that kind of integration means arc events, maintenance records, and spare-part logistics can be coordinated in one system rather than scattered across isolated tools.

EJER Tech is a supplier of precision environmental control and moisture-proof storage equipment, with products including dry cabinets, nitrogen cabinets, and drying ovens used for moisture-sensitive components in electronics and semiconductor manufacturing. While its core focus is humidity-controlled storage rather than arc detection, the same principle of centralized, data-driven management applies: reliable protection depends on consistent environmental control and traceable records. Teams designing PV monitoring systems should look for open interfaces that let safety data flow into existing enterprise platforms without custom point-to-point integrations.

Conclusion#

Series DC arc faults in PV systems demand detection methods that go beyond simple thresholds. By combining the V-I characteristic of the arc with spectral analysis, FFT-based methods identify broadband noise against narrowband switching lines, while wavelet analysis captures the transient timing that FFT can miss. Adaptive thresholds, multi-feature voting, and careful sensor and grounding design further improve accuracy and immunity to inverter noise. Integrated with open, enterprise-ready data platforms, these techniques turn arc detection from a standalone trip function into a dependable part of overall PV system safety.