Real-time isooctane and pentane gas identification based on machine learning analysis techniques

Artificial intelligence has made it possible to advance in the development of electronic systems that aim to replicate human senses such as vision, hearing and smell in intelligent applications or robotic systems. Today, there are commercial sensors for detection of conventional gases such as methane, butane, carbon monoxide, among others, but not for gases of a more complex chemical nature which are widely used in the petrochemical industry. In this paper we present the development of a recognition system between isooctane and pentane gases based on both low-cost sensors used in the detection of conventional gases and machine learning analysis techniques. For this, a novel gas analysis approach is proposed based on the use of Principal Component Analysis (PCA), both Linear and Quadratic Discriminant Analysis (LDA and QDA), and Support Vector Machine (SVM) algorithms. The experimental results showed that PCA-LDA method achieved the best prediction rates with 85.93% testing accuracy. Our approach also achieved 73.33% precision and an average prediction time of 0.003 seconds when tested in a simple real-time application.