Deep Learning Framework With Essential Pre-Processing Techniques for Improving Mixed-Gas Concentration Prediction

Multiple gas detection in mixed-gas environments is a challenging issue many engineering industries because some of the gases can raise defect rates and reduce production efficiency. For chemo-resistive sensors, precise estimation be measurement variance non-linear nature especially low concentration environment. A simple application deep learning models, however, does not yield sufficiently accurate predictions concentrations multiple mixtures; thus, it essential to develop basic strategies for enhancing accuracy all possible ways. In this study, we framework achieving high prediction by studying pre-processing techniques, task design, architecture design. pre-processing, study several aspects processing time-series sensor data identify key techniques complementing models’ limitations. We utilize design show that multi-task generate synergistic effect. Additionally, further improvement considering on-off classification as part hybrid task. Concerning investigate Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), Recurrent (RNN) models after applying identified techniques. CNN outperformed other joint analysis with The effectiveness our confirmed UCI mixture dataset acquired using chemical platform where 16 sensors are exposed ethylene, CO, methane gases. Using dataset, effective prediction. achieves significant when compared previous studies.