**1. Introduction**

The rapid development of technology and social advancements has led to the skyrocketing of energy demand, which has, in turn, resulted in an increase in fossil fuel generation of energy [1,2] This has raised concerns of high CO2 emission into the atmosphere due to the combustion of fossil fuels [3,4], which leads to global warming, GHG emissions, climate change, and other environmental issues [5]. Owing to the global commitment to overcome these issues by reducing fossil fuel energy generation to the bare minimum, the renewable industry has experienced an exponential growth and development in recent years. Renewable energy sources, especially solar [6,7], have been increasingly adopted for residential, commercial, and industrial applications [8–10]. The 2020 first quarter (Q1 2020) report of the National Renewable Energy Laboratory (NREL) stated that at the end of 2019, the installed solar PV capacity totaled 627 GWDC, an increase of 115 GWDC from the previous year [11].

Solar PV systems, however, need constant maintenance in order to efficiently operate over time. Therefore, strategies have to be in place to effectively monitor and maintain these systems. Various conventional methods are deployed by experts to carry out preventive, corrective, and predictive maintenance activities [12]. These methods usually equip the PV system with ground fault detection interrupters (GFDI) and overcurrent protection devices (OCPD). However, most of the time, they are not sufficient enough for detecting certain faults due to low irradiance conditions, nonlinear output characteristics, PV inverter maximum power point trackers (MPPT) or high fault impedances [13]. The need for more adequate and intelligent strategies of detecting and diagnosing faults in PV systems has encouraged the adoption of AI-based methods. These methods utilize machine learning to

**Citation:** Abubakar, A.; Almeida, C.F.M.; Gemignani, M. Review of Artificial Intelligence-Based Failure Detection and Diagnosis Methods for Solar Photovoltaic Systems. *Machines* **2021**, *9*, 328. https://doi.org/ 10.3390/machines9120328

Academic Editor: Christoph M. Hackl

Received: 26 October 2021 Accepted: 26 November 2021 Published: 1 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

train models to detect and locate various faults, monitor the general health status, and help the maintenance engineers of PV systems to rapidly expedite system recovery [13].

Over the years, AI has been utilized in PV system's design [14], MPPT [15], power prediction [16], as well as failure detection and protection [17–21]. The authors of [22] detailed the key characteristics that distinguish fault detection and diagnosis methods in PV systems. According to the authors, these characteristics include rapidity in detecting defects, input data (climate and electrical data), and selectivity (ability to distinguish between different faults). In addition, the study divided AI-based fault detection and diagnosis methods into two categories. First, visual and thermal methods for detecting discoloration, browning, surface soiling, hotspot, breaking, and delamination. Second, electrical methods for detecting and diagnosing faulty PV modules, strings, and arrays, such as arc faults, grounding faults, and diode faults. In [17], the adaptive neuro-fuzzy inference framework was adopted for the development of a smart fault detection approach for PV modules, while in [18], a dual-channel convolutional neural network model with a feature selection structure was proposed for PV array fault diagnosis. K-NN was used by the authors of [19] for the modeling of PV systems based on experimental data for detection, while in [20], a wavelet-based protection strategy was presented for the detection of a series of arc faults interfered by multicomponent noise signals in grid-tied PV systems. In [21], a PV fault detection algorithm that integrates two bi-directional input parameters based on the artificial neural network was presented. A novel extreme learning machine (ELM) based modeling method, featuring high training speed and generalization performance was proposed in [23], using current-voltage (I-V) curves measured at different operating conditions, for the characterization of the electrical behavior of PV modules. The authors in [24] attempted to improve the integration of PV systems into the electrical network by controlling the converter and inverter. This is achieved through the introduction of an adaptive reference PI (ARPI) for the inverter aimed at enhancing the system performance by supporting low voltage ride through (LVRT) capability and smoothing of the PV power fluctuations during variable environmental conditions. The authors of [25] proposed a model based on the geographic information system (GIS)-based reinforcement learning, for the optimal planning of rooftop PV system. The model considers the uncertainty of future scenarios across the buildings lifecycle. Another AI application is PV systems, which is utilized in [26] for the optimal dispatch of PV inverters in unbalanced distribution systems using reinforcement learning, while the authors of [27] used AI for the optimal design of a phase change material integrated renewable system with on-site PV, radiative cooling, and hybrid ventilations. In addition, reinforcement learning with the fuzzified reward approach was used in [28] for controlling the MPPT of PV systems. Another MPPT algorithm of the PV system, which is based on irradiance estimation and the multi-Kernel extreme learning machine, was presented in [29] in order to reduce investment costs and improve PV system efficiency, while in [30], the deep reinforcement learning approach was used for MPPT control of partially shaded PV systems in smart grids. Two other literatures on the application of AI for MPPT control are provided in [31,32]. The authors of [31] presented a new combined ELM variable steepest gradient ascent MPPT for the PV system, while the authors of [32] presented a novel meta-heuristic optimization algorithm based MPPT control technique for partially shaded PV systems. As stated earlier, AI was also utilized for PV power prediction as in the case of [33], where the short-term PV power prediction was achieved using a hybrid improved Kmeans-GRA-Elman model based on multivariate meteorological factors and historical power datasets. Moreover, AI was used for a similar application in [34], where the deep learning and wavelet transform integrated approach was used for short-term solar PV power prediction. The authors of [35] developed an intelligent real-time power management system, where an incremental unsupervised neural network algorithm was used to predict the output power and then detect the power fluctuations occurrence of a grid-tied PV system. A comparative study on short-term PV power prediction using the decomposition based ELM algorithm was presented in [36], while in [37], a short-term PV power forecasting model using the hybrid deep learning approach was proposed.

This study presents an extensive review of AI-based methods and techniques of fault detection and diagnosis reported in various literatures. The contribution of the study is in outlining the characteristics of the reviewed AI-based methods and their effectiveness in rapidly and efficiently detecting faults with minimal error, since the effectiveness of a fault detection and diagnosis method depends on the following factors: Its ability to detect a fault and pinpoint its location in the shortest possible time; its relative affordability; and ease of use [38]. The structure of the remaining part of the paper is as follows. Section 2 discusses the various types of faults that occur in PV systems; Section 3 introduces artificial intelligence and machine learning; Section 4 provides a review of the AI-based fault detection and diagnosis methods proposed in various literatures; and Section 5 concludes the present work and discusses its perspective.

#### **2. Types of PV System Faults**

Over time, PV systems experience fault occurrences that affect the system's operating efficiency, may cause damage to the system components, and may also lead to dangerous fire threats and safety hazards. PV system faults are classified as physical, environmental or electrical faults [39]. Panel faults, such as PV cell internal damages, cracks in panels, bypass diodes, degradation faults, and broken panels are classified as physical faults [39]. Shade faults due to bird dropping, dust accumulation, cloud movement, and tree shadows are classified as environmental faults [39]. Faults that are classified as electrical faults include MPPT faults, open-circuit faults, ground faults, line-line faults, short-circuit faults, arc faults, and islanding operation [39,40]. This section briefly discusses the different types of faults peculiar to PV systems.

#### *2.1. Shading Faults*

Shading occurs when objects, such as trees, neighboring buildings, and overhead power lines, cast shadows on PV modules [41,42]. Shading in PV arrays could be homogeneous, where there is a balanced reduced irradiation across the PV panels or nonhomogeneous, where there is an unbalanced reduced irradiation across the panels [39].

### *2.2. Arc Faults*

A frequent high-power discharge of electricity through an air gap between conductors causes this type of fault [43–45]. The two forms of arc faults include first, the series arc faults that usually originate from solder separation, connection corrosion, cell damage, rodent damage or abrasion from numerous sources. Second, parallel arc faults that result from insulation failure in current-carrying conductors [43,44].

#### *2.3. Line-Line Faults*

A line-line fault is an unintentional short-circuit between two points with differing voltage potentials [46–48]. These faults are more difficult to detect than other faults and are frequently misinterpreted as short-circuit faults in grounded PV systems, since the fault current is determined by the voltage differential between two fault spots [39]. The most common types of line-line faults are intra string faults, which are short-circuit faults between two locations on the same string, and cross string faults, which are short-circuit faults between two places on separate threads [39].

#### *2.4. Ground Faults*

To protect users from a possible electric shock, it is common practice that the metallic parts of the PV array are grounded using earth-grounding conductors (EGC) [39]. The term "ground fault" refers to any unintentional connection between a current-carrying conductor and an EGC that results in a current flow to the ground [45,46,49,50].
