Accurate fault detection is the key condition to ensure the durability and stability of photovoltaic (PV) systems. Traditional methods for detecting these faults generally depend on high-bandwidth cloud computing or expensive sensing devices. This dependence usually leads to excessive delay times and operating costs. To minimize this obstacle, this research work presents a lightweight Tiny Machine Learning (TinyML) infrastructure. It diagnoses the typical PV faults such as short-circuit, open-circuit, and partial shading locally. The final simulated system shows the accuracy of 94.31% for the purpose of fault detection. It uses a high-quality electrical dataset for training the quantized Deep Neural Network (DNN). It has also implemented 8-bit integer quantization to overcome the memory restrictions of the ESP32 microcontroller. This process reduces the storage requirement by a factor of 10.2 which is 17.7 KB. Experimental assessment also shows that the on-device processing time was approximately 22 ms, which is a 25-fold reduction in latency compared to traditional cloud-based methods. With these outcomes, quantized intelligence on the edge may offer a cost-effective and viable way to create protective and self-reliant renewable energy systems.