Dual gas sensor with innovative signal analysis based on neural network

Carbon monoxide and methane dual gas sensor with low system complexity and high stability is proposed in this study. A neural separator based on deep learning is developed to solve the cross-interference problem from the ultra-high spectral overlap between CO and CH4 molecules. A large amount of simulated overlapping spectra of different concentrations CO and CH4 are used to construct, train, tune and test the neural separator, instead of collecting data from onerous experiments. The linear fitting is performed between the predicted concentrations and preset concentrations of CH4 and CO, determination coefficients of R2 = 0.99960 and R2 = 0.99301 are achieved respectively which proves the accuracy of the dual gas sensor is robust and greatly enhanced by the neural separator. In addition, the minimum detection limits of 120.86 ppm (CH4) and 0.5 ppm (CO) are achieved in real-time simultaneous detection of CO and CH4 overlapping environment. This is a successful attempt to apply deep learning method to tunable diode laser absorption spectroscopy (TDLAS) gas sensors to solve the problem of spectral cross-interference, which provides an alternative direction for the realization of simultaneous measurement of multi-component gases.