Deep Learning Based Hardware System for Factory Leak Detection

The industrial sector grapples with a critical challenge in mitigating the impact of leaks, posing threats to environmental sustainability and operational resilience. Existing technologies, including machine learning and solenoid valves, lack integration into a cohesive system. To address this problem, our proposed solution utilizes a custom dataset of 10,000 images for the development of a deep learning model. The dataset spans 20 diverse areas and surfaces, capturing various lighting conditions, including day, night, evening, and afternoon.Through exhaustive experimentation, the study explores six activation functions, four regularization methods, and four optimizers, resulting in the identification of the top 5 models with the highest test accuracy. Further hyperparameter optimization yields a model with 95.8% accuracy. This optimized model is seamlessly integrated into a practical implementation, involving IP cameras, Raspberry Pi, and solenoid valves, demonstrating its applicability in real-world scenarios. Early detection enabled by this system holds the potential to conserve valuable resources and enhance operational efficiency.