Quantitative information extraction from gas sensor data using principal component regression

Abstract

This paper presents a novel use of the principal component analysis (PCA) and regression methods for quantitative feature extraction from gas sensor data. In this approach, PCA plots are interpreted by observing the locations of samples in the principal component domain. A trainable data processing system that also produces numerical output is designed to validate the method. The main advantages of this system are: 1) retrainability: once it is trained, it can be used for any gas set; 2) flexibility: adaptation to different targets does not require hardware modifications (if a sufficient number and variety of sensors are installed in the sensor cell); and 3) simplicity: all computations are performed with only linear operators, and hence the system does not require complex structures or powerful computation resources. Several experiments are conducted using two industrial gases (toluene and ethanol) to validate the approach. The new approach is also compared with two classic principal component regression (PCR) methods. The results show that the new approach performs better than the classic PCR approaches.