Simultaneous Prediction of Gas Flow Rate and Concentration in the Micro-Calorimetric Sensor Using Physics-Guided Neural Networks

Dynamic and simultaneous prediction of gas flow rate and concentration is crucial in the fields of industrial gas delivery, environmental monitoring, and energy. The micro-calorimetric sensor can measure gas flow rate and concentration simultaneously due to its advantages such as no moving parts, small size, and short response time. In this study, a physical model of the sensor downstream temperature with respect to the two mutually coupled parameters of mixed gas flow rate and concentration is established. The influence of the concentration of the mixed gas on the sensor output under different flow conditions is analyzed theoretically and experimentally, and the prediction performance of the physical model is evaluated. The physical model is further combined with the neural network to construct a hybrid physical data prediction model hybrid physics–data (HPD). The experimental results show that in the flow range of 0 ~ 11 L/min, the HPD model reduces the concentration prediction error by 79.25% compared with the pure data-driven neural network, and achieves an absolute flow rate prediction error within 0.0015 L/min and an absolute concentration prediction error within 0.65%. This method can be used for the simultaneous dynamic measurement of gas flow rate and concentration and can be applied to gas monitoring and treatment in various industrial fields.