A miniaturized electronic nose system based on sensor array for C2H4 concentration prediction

Abstract

Ethylene (C₂H₄) is a highly flammable and explosive gas, which may cause severe hazards. To prevent the potential fire or explosion incidents and ensure the safety of chemical plants or surrounding environments, monitoring the concentration of C₂H₄ gas is essential. As a device for detecting gases, the electronic nose (E-nose) can effectively predict concentrations of gases, but traditional E-nose systems face challenges such as poor portability, high cost, and high-power consumption. In this study, a miniaturized, low-cost portable E-nose device was designed utilizing micro-electromechanical systems (MEMS) sensors, integrated with advanced deep learning algorithms, to achieve the real-time prediction of C₂H₄ concentrations. Subsequently, a dataset containing the samples originating from varying concentrations of C₂H₄, ranging from 10 to 100 ppm, was systematically collected. To enhance the prediction accuracy, a CE-CNN model was proposed, which integrates an Efficient Channel Attention mechanism into the Convolutional Neural Network, while Cosine Annealing Warm Restart strategy optimizes the training process by dynamically adjusting the learning rate. Compared to current popular algorithms, CE-CNN achieves the lowest mean absolute error (MAE) of 1.7286 and mean square error (MSE) of 2.1064, as well as the highest R-squared (R²) value of 0.9985 for C₂H₄ gas concentration prediction. Finally, edge computing was implemented on the designed portable E-nose device, enabling independent gas data collection, processing, and prediction, with a maximum concentration prediction error of only 6.80 % of the actual concentration.