A novel temperature compensation approach of IR gas sensors in coal mines

•Through the experiment test, the constant heating temperature of the gas sensors are 45 ℃.•The precisions of gas sensors before and after temperature compensation were analyzed.•Temperature compensation effect of GA-BP is better than BP for CO2, CH4 and CO. Infrared spectrum analysis is a technical method for quantitative analysis of coal mine gases. However, current infrared (IR) gas sensors are significantly affected by the change of ambient temperature. In this study, experimental research was carried out in the hope of improving the accuracy and applicability of IR gas sensors. Firstly, a test platform whose functions could support sensor temperature compensation experiments was built. The accuracy of sensors without temperature compensation was calculated under the concentrations CO2 of (0.5%, 1%, and 2.43%), CH4 concentrations of (0.1%, 1%, and 4.0%), and CO concentrations of (10 × 10−6, 100 × 10−6, and 500 × 10−6). The calculation results show that the errors are 1.4%∼2.0%, (3,000∼5,000) × 10−6, and (50∼160) × 10−6, respectively, which fail to meet the requirements for error ranges. Secondly, the constant temperature heating test of sensors was carried out in the temperature range of (25–55)℃, and the optimal heating temperature was determined to be 45℃. Then, the accuracy of sensors after temperature compensation was calculated again under the same CO2, CH4 and CO concentrations at 45℃. The results reveal that the errors are only 0∼0.045%, 0∼900 × 10−6, and 0∼9 × 10−6, respectively, which are better than the results before temperature compensation. Thirdly, with the temperature in the high-low temperature test chamber set as (-20∼45)℃, it is found that the errors remain huge. Considering the highly variable temperature conditions in underground coal mines, the genetic algorithm-back propagation (GA-BP) neural network model was established for error analysis. For CO2, CH4, and CO, the mean absolute errors (MAEs) of the GA-BP neural network model are 62.7%, 26.9% and 29.1% lower than those of the back propagation (BP) neural network model, respectively. In summary, this study obtained the appropriate heating range of IR gas sensors, established the error optimization model under different ambient temperature conditions, and provided a technical foundation for the construction of the in-situ on-line monitoring system for coal mine gases.