CNN Based Fault Classification and Predition of 33kw Solar PV System with IoT Based Smart Data Collection Setup

A Solar Photovoltaic (PV) System is an energy conversion system that uses the photovoltaic effect to convert sunlight into electricity. A fault in a Solar Photovoltaic (PV) system refers to any abnormal condition or defect that disrupts the normal operation and performance of the solar system. These faults can arise from a variety of factors, including environmental conditions, manufacturing defects, installation errors, and wear and tear of the components. Fault diagnosis in solar PV systems involves the detection, identification, and rectification of faults or abnormalities that can occur due to various reasons. By detecting and addressing faults early, systems can maintain optimal performance levels. Machine Learning (ML) in Solar Photovoltaic (PV) systems refers to the application of algorithms and statistical models that enable computers to perform specific tasks without using explicit instructions, relying instead on patterns and inference. In the context of solar PV systems, ML is used to analyse and interpret vast amounts of data generated by these systems to enhance their efficiency, predict energy production, detect and diagnose faults, and optimize maintenance and operation. By analysing data from sensors and system logs, ML algorithms can identify patterns indicative of faults or inefficiencies, such as shading, soiling, or equipment malfunctions, often before they become serious issues. Convolutional Neural Networks (CNNs) are a class of deep learning algorithms most commonly applied. They are particularly powerful for tasks involving data recognition, classification, and analysis due to their ability to automatically and adaptively learn spatial hierarchies of features. This research presents a unique machine learning model based fault diagnosis and detection method for a 33 KW solar PV system at P.S.R. Engineering College, Sivakasi. The real-time data from the PV system for five years, covering 23,000 instances of eight types of faults such as Cell Cracks or Hot Spots, Partial Shading, sensor fault, Module failure, Ground Faults, Communication Errors, Environmental Factors, Grid Connectivity Issues are collected. CNN is applied to the data and analysed their performance in terms of accuracy, precision, and standard deviation (SD)-score. It is found that CNN achieved the best results, with an accuracy of 98.7% a precision of 95%, a recall of 98%, and an F1 score of 96.5%. Therefore, CNN is used as the fault prediction also. The model is implemented using Python programming language and demonstrated its effectiveness on test cases. The smart data gathering system was achieved utilizing an ESP32 node with several sensors. The obtained data was stored in an authorized Google Sheet and compared to predetermined threshold ranges. When any parameter deviates from its threshold value, the ESP32 node starts a cooling and dust cleaning procedure with a water pump and drip pipe configuration. If the divergence persists, the ESP32 node activates a camera to capture an image of the panel and sends it to the Google Sheet via a connection for further analysis and fault correction.