Solar energy has been gaining significant traction as a clean, renewable energy source for the future. Each solar panel in the PV system experiences an efficiency loss of 0.5% to 1% annually. This gradual decline in performance highlights the importance of continuous condition monitoring and predictive maintenance to ensure optimal energy output. Due to abnormal temperature changes brought on by hotspots, cell cracks, bypass-diode failures, soiling, partial shading, delamination, weak electrical connections, corrosion, and PV module deterioration over time, faults may arise in the solar thermal image. Thus, a classification model is needed to differentiate between faulty and normal thermal images since it might not be visually apparent during human evaluation. In order to classify faults that might be driven by divides or hotspots in solar thermal pictures, this research suggests a new lightweight CNN classifier in the PV system that extracts the significant characteristics from the input thermal images. The model can be deployed on low-power devices because it requires fewer parameters than conventional deep networks. Because of this effective, scalable, and affordable methodology, large solar farms can now accurately diagnose PV issues. The proposed model used the MATLAB platform to compare and simulate the performance of the work. The suggested lightweight CNN outperformed the current models, CNN at 90%, DMOA at 89%, and AQO at 92%, with an accuracy value of 93% at a training percentage of 70%, according to the MATLAB validation of the planned framework. The outcomes demonstrate that the recommended approach works better than the traditional methods of solar thermal image fault detection.
Buy article PDF