Enhancing Accuracy of Nanocomposite Hydrogen Sensors in Various Environmental Situations through Machine Learning

This paper presents a proof of concept that combines a nano-composite hydrogen detecting sensor and machine-learning technique to achieve accurate and fast detection of hydrogen leakage. The nano-composite hydrogen detecting sensor is fabricated by depositing MoS₂ on a SiO₂/Si wafer using chemical vapor deposition, followed by forming discrete Pd nanoparticles through DC (Direct current) sputtering. This sensor shows high sensitivity of 2.77 and fast response time of 4 to 5 seconds at room temparature, but has a significant dependency on environmental factors such as temperature, and humidity. A machine learning technique, i.e. random forest, was incorporated to filter out the environmental factors. Experimental results show that the combination, i. e. MiCS-2714 sensor not only retains sensitivity, response time of the nan-ocomposite but also attains R² score of 0.994, MSE 0.0506, and the state classification accuracy of 0.979.