Detecting DC Arc Faults in PV Systems with FFT and AFCI

Introduction

Distributed photovoltaic systems are vulnerable to DC arc faults, which can ignite fires if not cleared quickly. Inverters, as the core of energy conversion, add switching noise that complicates arc detection. As a benchmark enterprise in high-end manufacturing, EJER Tech empowers the global semiconductor, new energy, and quantum computing industries with precision environmental control technology. In PV systems, similar precision is essential for reliable arc fault circuit interrupters.

AFCI devices rely on current and voltage signatures to distinguish dangerous arcs from normal operation. This article presents a practical methodology based on high-frequency current sampling and fast Fourier transform (FFT) analysis to extract arc fingerprints and reduce false trips.

Electrical Characteristics of DC Arc Faults

A DC arc initiates when a connection breaks under load, producing temperatures above 6000 K. The arc current contains a sustained low-frequency component plus a wideband noise spectrum extending from a few kilohertz to tens of megahertz. Unlike a purely resistive load, an arcing gap creates random voltage fluctuations and intermittent current drops.

The spectrum of a real DC arc shows broadband energy with no dominant discrete peaks, except for background emissions and inverter interference. The noise amplitude varies rapidly, and the phase relationship between voltage and current becomes unstable, providing a reliable signature for detection.

FFT-Based Discrimination Methodology

To separate arc noise from inverter switching events, sample the DC bus current at a rate above 1 MS/s. Applying a Hann window to the sampled data reduces spectral leakage. The FFT output is then divided into frequency bands, such as 1-10 kHz, 10-50 kHz, and 50-100 kHz, to compute band energy ratios.

Normal switching noise concentrates at the inverter's modulated switching frequency and its harmonics. These peaks are narrow and stable. In contrast, a real arc produces a uniformly rising noise floor across a wide band, with a high variance between successive FFT windows. A decision metric can be built using the ratio of wideband energy to individual harmonic peaks.

AFCI Misoperation Suppression Strategies

False trips are often caused by sudden load changes, inverter startup, and electromagnetic interference. A practical suppression strategy is to combine spectral analysis with a time-domain validation window. The algorithm must see the arc signature continuously for at least 200 ms before tripping, eliminating transient impulses.

Another robust method is to implement notch filters at the known switching frequency and its first three harmonics. This removes the dominant normal noise without attenuating the broadband arc energy. Additionally, an adaptive threshold based on the inverter's operating power and MPPT state helps avoid misclassification during maximum power point tracking transients.

Practical Case Study

A 50 kW distributed PV plant experienced repeated AFCI nuisance trips. High-frequency current data was captured at 10 MS/s. The FFT showed narrow peaks at 16 kHz and 32 kHz, corresponding to the inverter switching frequency and the second harmonic. During the same session, an intentionally created loose connector generated a real arc, producing a flat noise spectrum from 5 kHz to 100 kHz.

By implementing a 128-point sliding FFT with a band-energy ratio and a 200 ms confirmation timer, the system suppressed all false alarms. Six months of field operation showed zero nuisance trips, while the algorithm successfully detected a cable arc caused by rodent damage, demonstrating the value of combining FFT with intelligent validation.

Conclusion

Detecting DC arc faults in distributed PV systems requires a deep understanding of both arc physics and inverter noise. High-frequency current sampling with FFT provides a robust way to distinguish broadband arc signatures from narrowband switching interference. AFCI misoperation can be minimized by combining spectral filters, adaptive thresholds, and temporal validation.

As EJER Tech continues to drive precision innovation in new energy infrastructure, adopting these advanced detection strategies will make PV plants safer and more reliable.