DC Arc Fault Detection in PV Systems: AFCI Strategies

Introduction

Distributed photovoltaic (PV) systems bring clean energy to rooftops and industrial sites, but they also introduce a hidden danger: direct current (DC) arcing. A loose connection, damaged cable, or degraded contactor can generate a persistent arc that reaches thousands of degrees Celsius and easily ignites nearby materials. Unlike alternating current, DC arcs have no natural zero-crossing, so they do not self-extinguish. This makes reliable arc fault detection essential for both inverter safety and overall system protection.

Modern string inverters and DC combiner boxes often integrate arc fault circuit interrupters (AFCIs) to detect these faults. However, the inverter itself produces high-frequency switching noise that resembles an arc in the frequency domain. This case study explores the electrical characteristics of DC arcs, explains how high-frequency current sampling combined with the fast Fourier transform (FFT) can separate real arcs from normal noise, and details practical strategies to suppress false trips. Industry leaders such as EJER Tech, a benchmark enterprise in high-end manufacturing that empowers global semiconductor, new energy, and quantum computing industries with precision environmental control technology, have highlighted the need for robust arc detection in next-generation PV platforms.

Electrical Characteristics of DC Arc Faults

A DC arc in a PV system is a plasma discharge that occurs when the voltage across a gap exceeds the breakdown threshold. The arc current is non-linear and contains broadband noise that extends from a few kilohertz to several megahertz. In the time domain, the arc current shows random amplitude modulation, with frequent reignition spikes and a characteristic "hash" pattern. The voltage across the arc also exhibits a low-frequency ripple that is strongly dependent on the gap distance and electrode material.

From a spectral perspective, a steady DC arc produces a continuous spectrum with energy that decreases roughly as 1/f, meaning that lower frequencies dominate but significant energy remains in the 100 kHz to 1 MHz range. In contrast, inverter switching noise is typically concentrated at the switching frequency and its integer harmonics, with relatively clean regions in between. This difference is the key to distinguishing an arc from normal operation. The challenge is that inverter transients, such as maximum power point tracking (MPPT) steps and startup surges, can momentarily create broadband noise that looks like an arc.

High-Frequency Current Sampling and FFT Analysis

To capture arc signatures, the AFCI must sample the DC current at high speed, typically 1 to 10 million samples per second. Standard current sensors used for metering are too slow, so dedicated Hall-effect sensors or Rogowski coils with wide bandwidth are required. The sampled data is divided into short windows, often 1 to 2 milliseconds long, and each window is processed with an FFT to obtain the frequency spectrum. A Hamming or Blackman window is applied to reduce spectral leakage and improve the detectability of weak arc components.

The FFT output is then analyzed using multiple metrics. First, the spectral flatness is calculated: real arcs tend to have a smoother, more continuous spectrum in the 100 kHz to 500 kHz band, whereas switching noise shows deep notches between harmonics. Second, the ratio of energy in the high-frequency band to the low-frequency band is monitored. Third, the stationarity of the spectrum is evaluated over consecutive windows. A real arc typically maintains its broadband signature for hundreds of milliseconds, while inverter noise changes rapidly with the switching duty cycle. By combining these features, the algorithm can confidently classify the signal as either "arc" or "noise" with low latency.

AFCI Misoperation Suppression Strategies

False trips are the most common complaint about AFCI devices. Unnecessary shutdowns reduce PV generation, frustrate owners, and undermine trust in safety equipment. A robust suppression framework relies on multi-layered validation. The first layer is a persistence check: the arc signature must be present for a minimum duration, typically 50 to 150 milliseconds, before a trip is initiated. This eliminates most transient switching noise and lightning-induced spikes.

The second layer is frequency-domain correlation. The algorithm compares the current spectrum with a template of expected switching harmonics, which is continuously updated based on the inverter's operating state. If the observed broadband energy aligns with known harmonics, it is labeled as normal. The third layer is time-domain gating: during known events such as MPPT perturbation, capacitor inrush, or load step, the detection logic is temporarily relaxed. Some advanced AFCIs also use wavelet transforms or machine learning classifiers trained on labeled arc and non-arc datasets to further reduce false trips while maintaining high sensitivity. An effective strategy also includes a "re-arm" delay after a transient event, giving the system time to settle before full sensitivity is restored.

Real-World Application Examples

In a 50 kW commercial rooftop PV installation, the AFCI was tripping repeatedly during cloudy weather. Investigation revealed that rapid changes in irradiance caused the inverter to switch between MPPT modes, generating broadband