[Paper Review] Explainable Incipient Fault Detection Systems for Photovoltaic Panels

S. Sairam, Seshadhri Srinivasan+4 more|arXiv (Cornell University)|Nov 19, 2020

Risk and Safety Analysis27 references4 citations

TL;DR

This paper proposes an explainable fault detection and diagnosis system (XFDDS) for incipient faults in photovoltaic (PV) panels by combining an irradiance-based three-diode model (IB3DM) with XGBoost and LIME-based explainable AI. The hybrid system improves accuracy at low irradiance, reduces false alarms via a fault-signature metric, and provides intuitive, actionable explanations for field engineers, demonstrated across multiple PV technologies with real-world fault simulations.

ABSTRACT

This paper presents an eXplainable Fault Detection and Diagnosis System (XFDDS) for incipient faults in PV panels. The XFDDS is a hybrid approach that combines the model-based and data-driven framework. Model-based FDD for PV panels lacks high fidelity models at low irradiance conditions for detecting incipient faults. To overcome this, a novel irradiance based three diode model (IB3DM) is proposed. It is a nine parameter model that provides higher accuracy even at low irradiance conditions, an important aspect for detecting incipient faults from noise. To exploit PV data, extreme gradient boosting (XGBoost) is used due to its ability to detecting incipient faults. Lack of explainability, feature variability for sample instances, and false alarms are challenges with data-driven FDD methods. These shortcomings are overcome by hybridization of XGBoost and IB3DM, and using eXplainable Artificial Intelligence (XAI) techniques. To combine the XGBoost and IB3DM, a fault-signature metric is proposed that helps reducing false alarms and also trigger an explanation on detecting incipient faults. To provide explainability, an eXplainable Artificial Intelligence (XAI) application is developed. It uses the local interpretable model-agnostic explanations (LIME) framework and provides explanations on classifier outputs for data instances. These explanations help field engineers/technicians for performing troubleshooting and maintenance operations. The proposed XFDDS is illustrated using experiments on different PV technologies and our results demonstrate the perceived benefits.

Motivation & Objective

• Address the challenge of detecting incipient faults in PV panels under low irradiance conditions, where traditional model-based methods lose accuracy.
• Overcome the limitations of data-driven fault detection systems, including lack of explainability and high false alarm rates.
• Develop a hybrid fault detection framework that integrates high-fidelity modeling with interpretable AI to improve reliability and maintainability of PV systems.
• Enable field engineers to understand and act on fault diagnoses through human-readable, instance-specific explanations using LIME.
• Demonstrate the system’s effectiveness across multiple PV technologies using real-world fault scenarios, including partial shading and line-to-line faults.

Proposed method

• Propose an irradiance-based three-diode model (IB3DM) with nine parameters that explicitly incorporates irradiance and temperature in parameter computation to enhance accuracy at low irradiance levels.
• Use a non-convex constrained optimization problem with five meta-heuristic approaches (e.g., particle swarm, firefly) to estimate IB3DM parameters from measured PV data.
• Implement an extreme gradient boosting (XGBoost) classifier to detect incipient faults from aggregated PV data, leveraging its strong performance in handling noisy, imbalanced data.
• Introduce a fault-signature metric (FSM) that combines outputs from IB3DM and XGBoost to reduce false alarms by validating classifier predictions against physical model behavior.
• Integrate a local interpretable model-agnostic explanations (LIME) framework into the XFDDS to generate instance-specific, human-readable explanations for fault classifications.
• Trigger explanations only when both XGBoost detects a fault and the FSM exceeds a threshold, ensuring relevance and reducing computational overhead.

Experimental results

Research questions

• RQ1 Can an irradiance-based three-diode model (IB3DM) achieve higher accuracy than conventional models in low irradiance conditions, critical for detecting incipient faults?
• RQ2 How effectively can a hybrid model combining XGBoost and IB3DM reduce false positive alarms in incipient fault detection compared to standalone data-driven methods?
• RQ3 To what extent can LIME-based explanations improve the interpretability and trustworthiness of fault detection decisions for field engineers?
• RQ4 Can the fault-signature metric reliably distinguish between true incipient faults and transient anomalies or false positives in real-time PV data?
• RQ5 How generalizable is the XFDDS framework across different PV technologies and fault types, including partial shading and line-to-line faults?

Key findings

• The IB3DM achieved higher fidelity than standard single- and double-diode models, particularly at low irradiance (e.g., G ≥ 896 W/m²), due to explicit irradiance dependency in parameter computation.
• The fault-signature metric (FSM) significantly reduced false alarms by cross-validating XGBoost predictions with physical model behavior, ensuring only consistent fault signals triggered explanations.
• LIME-based explanations provided intuitive, actionable insights—such as low current (≤9.77 A) and power (≤1354 W) despite high irradiance (≥896 W/m²)—enabling field engineers to quickly identify partial shading faults.
• For line-to-line (LL) faults, explanations correctly highlighted low current (≤7.44 A) and low power (≤1112.52 W) with high voltage (>141.29 V), confirming fault presence and aiding diagnostics.
• False positive cases were effectively resolved by explanations showing high power (>1112.89 W) despite low current, revealing intermittent load behavior and preventing unnecessary maintenance.
• The XFDDS demonstrated robust performance in both binary and multi-class classification tasks across monocrystalline PV panels, with consistent and interpretable results across fault types.