0885-8993 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TPEL.2019.2911594, IEEE Transactions on Power Electronics A Digital Twin Approach for Fault Diagnosis in Distributed Photovoltaic Systems Palak Jain, Student Member, IEEE, Jason Poon, Student Member, IEEE, Jai Prakash Singh, Costas Spanos, Fellow, IEEE, Seth R. Sanders, Fellow, IEEE, Sanjib Kumar Panda, Senior Member, IEEE Abstract—Rooftop and building-integrated distributed photo- voltaic (PV) systems are emerging as key technologies for smart building applications. This paper presents the design methodol- ogy, mathematical analysis, simulation study, and experimental validation of a digital twin approach for fault diagnosis. We develop a digital twin that estimates the measurable characteristic outputs of a PV energy conversion unit (PVECU) in real-time. The PVECU constitutes a PV source and a source-level power converter. The fault diagnosis is performed by generating and evaluating an error residual vector, which is the difference between the estimated and measured outputs. A PV panel- level power converter prototype is built to demonstrate how the sensing, processing, and actuation capabilities of the converter can enable effective fault diagnosis in real-time. The experimental results show detection and identification of ten different faults in the PVECU. The time to fault detection in the power converter and the electrical sensors is less than 290 μs and the identification time is less than 4 ms. The time to fault detection and identification in the PV panel are less than 80 ms and 1.2 s, respectively. The proposed approach demonstrates higher fault sensitivity than that of existing approaches. It can diagnose a 20% drift in the electrical sensor gains and a 20% shading of a solar cell in the PV panel. Index Terms—Fault diagnosis, fault location, estimation, solar power generation, converters I. I NTRODUCTION Photovoltaic (PV) energy conversion architectures based on panel or subpanel-level distributed power electronic converters are increasingly ubiquitous for rooftop and building-integrated PV (BIPV) systems [2]–[5]. The unique advantages of a distributed power electronic-based PV system include higher energy yield (particularly, in partial shading conditions), higher performance reliability, lower installation costs, plug-and-play operation, and enhanced system flexibility, modularity, and scalability [3], [6]–[8]. The building block of a distributed PV system can be referred to as a PV energy conversion unit This work is supported in part by Republic of Singapore’s National Research Foundation through a grant to the Berkeley Education Alliance for Research in Singapore (BEARS) for the Singapore-Berkeley Building Efficiency and Sustainability in the Tropics (SinBerBEST) Program and in part by the NSF through the Graduate Research Fellowship program. This paper was presented in part at the 2016 IEEE 17 th Control and Modelling for Power Electronics (COMPEL) on June 30, 2016 [1]. (Corresponding author: Dr. Sanjib Kumar Panda) P. Jain and S. Panda are with the Department of Electrical and Computer Engineering at the National University of Singapore (NUS), Singapore 117583 (email: palakjain@u.nus.edu; sanjib.kumar.panda@nus.edu.sg). J. P. Singh is with the Solar Energy Research Institute of Singapore, NUS, Singapore 117574 (email: jaiprakash.singh@nus.edu.sg). J. Poon, C. Spanos, and S. Sanders are with the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley, CA 94720 USA (email: {jason, spanos, seth.sanders}@berkeley.edu). (PVECU). The PVECU constitutes a PV source (e.g. a PV panel or a PV subpanel) and a source-level power converter. The distributed PV systems are vulnerable to a variety of faults due to their complex outdoor installations, increased number of power electronic converters at the PV panel-level, harsh mission profiles, manufacturing defects, and aging. A comprehensive analysis of common failure modes that could occur in these PV systems and their impacts on performance and reliability are given in [9]–[11]. These failures degrade system performance, and endanger the safety and security of the buildings and its occupants. Moreover, these failures are difficult to locate and repair in building applications because rooftop and BIPV systems have a very large number of PVECUs which can be physically inaccessible due to complex installations (exacerbating maintenance and inspection that incur high cost). Thus, robust and cost-effective methods for ensuring their dependability and fault tolerance are necessary. Strategies for fault diagnosis have been explored to improve dependability and fault tolerance of complex systems. Fault diagnosis is a monitoring scheme that is comprised of two functions: 1) fault detection (FD), which is a binary decision on the occurrence of a fault; and 2) fault identification (FI), which is the process of classifying the precise fault type [12]. The successful fault diagnosis in a distributed PV system enables online fault remediation, increasing its availability and decreasing its maintenance costs. Prior works in the literature address fault diagnosis of var- ious subcomponents of PV systems. For example, the authors in [13]–[15] investigate various fault diagnosis approaches for power electronic converters in PV systems. Similarly, the authors in [16]–[18] report various methodologies for PV panel fault diagnosis. However, there is a lack of research work towards a holistic fault diagnosis approach for a complete PVECU. Table I presents common failure modes of each subcomponent in the PVECU. The holistic fault diagnosis approach should, at a minimum, be able to detect and identify these failure modes. Commercial PV monitoring solutions are available that provide web-based monitoring and use data analytics to detect a fault event [24], [25]. However, these solutions typically lack: 1) the ability to perform FI; 2) require high bandwidth communication for data acquisition and control; 3) require additional central processing units (CPUs); and 4) add latency in the diagnostics of system performance and faults. This work applies the concept of a digital twin to de- velop a holistic fault diagnosis approach. A digital twin is a digital emulation of a physical system that estimates the