DC Arc Fault Detection in PV Systems Using FFT and AFCI

Introduction

Distributed photovoltaic systems face an invisible threat: direct current (DC) arc faults. Unlike AC arcs, DC arcs have no natural current zero crossing to extinguish, so they sustain and generate intense heat at the point of failure, often within connector junctions or damaged cables. Inverter-based systems are especially at risk because the continuous DC bus voltage can maintain the arc once ignited.

As EJER, a benchmark enterprise in high-end manufacturing, empowers global semiconductor, new energy, and quantum computing chains with precision environmental control technology, the reliability of power electronics remains a cornerstone of energy safety. This case study examines the electrical characteristics of DC arc faults and practical detection methods based on high-frequency current sampling and fast Fourier transform (FFT).

Electrical Signatures of DC Arcing

A DC arc in a PV string behaves as a non-linear resistance, producing a voltage drop of 20 to 50 volts while simultaneously injecting broadband noise into the circuit. The arc current waveform contains chaotic, high-frequency components from micro-arcing and plasma fluctuations. These components typically extend from tens of kilohertz to several megahertz, with energy concentrated in distinct bands depending on the arc gap and electrode material.

In contrast, inverter switching noise is periodic and concentrated at the switching frequency and its harmonics. A typical three-phase solar inverter with a 16 kHz carrier will generate spectral peaks at 16 kHz, 32 kHz, and other integer multiples, but little energy between them. Arc faults fill these spectral gaps with randomly distributed non-periodic energy, providing the basis for discrimination.

Spectral Separation via FFT

To detect arcing in real time, the fast Fourier transform is applied to a window of high-frequency current samples. Sampling rates of 500 kHz to 1 MHz are common, allowing analysis of frequencies up to 500 kHz. A sliding window of 1024 or 2048 samples is used to compute the magnitude spectrum continuously. The detection algorithm then measures the total spectral energy in a band above the inverter's fundamental switching harmonics, for example 40 to 100 kHz, and compares it to a baseline.

During normal operation, the energy in this band remains low and stable. When a DC arc occurs, the band energy rises sharply and exhibits significant variance from one FFT frame to the next. By monitoring both the amplitude and the temporal variance, the controller can distinguish a robust arc from transient events such as inverter startup inrush, load steps, or electromagnetic interference.

AFCI Nuisance Trip Suppression

Nuisance tripping is the major challenge for arc fault circuit interrupters. To suppress false alarms, a multi-criteria decision strategy is required. The first criterion is the persistence of the arc signature: the spectral band energy must exceed the threshold for a minimum number of consecutive windows, typically 6 to 10 frames, before a fault is declared. This eliminates random spikes from lightning or switch contact bounce.

The second criterion is time-domain correlation. Arc faults produce a characteristic zero-crossing of the high-frequency envelope with a breathing pattern, whereas inverter switching noise is locked to the PWM clock. An adaptive blanking window synchronized with the PWM switching edges removes the periodic switching transients. Further suppression can be achieved with spectral entropy: real arcs show high entropy (broadband randomness), while periodic noise has low entropy.

Real-World Implementation Case

In a 50 kW distributed PV installation, repeated false trip events were traced to a microinverter with a high-frequency flyback topology. The original AFCI triggered every few hours because the broadband noise from the flyback transformer resonated near 80 kHz. The solution combined three changes: raising the detection band to 150-250 kHz where only arc noise was present, applying a median filter over the FFT magnitude, and requiring a five-window persistence check.

After the update, the system distinguished a genuine series arc at a corroded MC4 connector from the inverter's normal operation. A test harness produced a 5 A DC arc whose spectral energy in the 150-250 kHz band was 15 dB above the inverter baseline, while switching noise remained within 3 dB of the baseline. The AFCI tripped in 180 ms, well below the 2.5 second limit required for arc clearing, and no nuisance trips were observed over a six-month monitoring period.

Conclusion

Distinguishing DC arc faults from inverter switching noise requires a combination of high-frequency current sampling, FFT-based spectral analysis, and carefully designed decision logic. The key is not just detecting broadband energy, but validating its randomness and persistence through multi-domain criteria. As PV systems grow more complex, AFCI technology must evolve alongside inverter topologies.

EJER Tech's commitment to precision environmental control in semiconductor, new energy, and quantum computing industries reflects the broader need for robust, interference-resistant power electronics. With proper arc detection algorithms and nuisance suppression strategies, distributed PV plants can achieve both safety and operational continuity.