The Development Of New MEMS Gas Sensor Based On Neural Network Temperature Compensation

As the largest country in the coal production,mining accidents often occurs in China,especially the roadway gas accident.It’s very important to monitor the gas concentration.At present,gas sensors based on the principle of catalytic combustion is commonly used in China.It has simple structure and it is easy to use,but it has three main defects: the one is power consumption.Most catalytic combustion type gas sensor’s power consumption is between 150~200 m W.The two is long response time,normally about 15 s,and the three is that the catalyst is easy to be poisoned.The response time not only affects the speed of gas detection,but also affects the power consumption.The longer the time needed to sample the gas concentration,the higher the power consumption.Catalyst poisoning is something about sensor life.Therefore,the main defects of the catalytic combustion gas sensor can be summed up as power consumption and life.The laboratory has designed a new thermal conductivity gas sensor through MEMS process(the sensor is only sensitive probe),its power is only90 m W,and the response time is 15 ms.Heating elements is also the gas sensitive element,without catalyst or catalyst carrier,so it will not be poisoned,which has great advantages and potential applications compared with the existing catalytic combustion type gas sensor.However,because of semiconductor processing,its output signal is affected by the temperature,resulting in measurement is not accurate.Therefore,it is necessary to make temperature compensation for application.Firstly,the principle of the detection of the new MEMS methane sensor is analyzed theoretically and its main characteristics are studied and analyzed through experiment.The sensor’s working range,rated power,sensitivity,response time and other parameters are determined.Through the study of the sensor’s temperature characteristics,it was found that the sensor was severely affected by temperature and needed temperature compensation.Secondly,BP and RBF neural networks are adopted for the sensor temperature drift.BP neural network is easy to fall into the local minimum,so simulated annealing algorithm(SAA)is used to improve the measurement accuracy.However,the traditional SAA will slow convergence speed,so an improved SAA based on VFSA algorithm with the idea of non-uniform change is proposed.A small number of search cycles with reduced optimization radius in the process before and after annealing is added in this algorithm.The simulation results show that the improved algorithmimproves the convergence speed and the maximum absolute error decreases from0.455% to 0.1206%.RBF neural network has high accuracy in temperature compensation,but it has low efficiency in temperature drift with better linearity.Therefore,this paper proposes a hybrid optimization algorithm,which uses least squares method in low temperature linear segment and RBF neural network with better performance in high temperature nonlinear section.The simulation results show that the hybrid optimization algorithm can not only meet the requirements,but also improve the compensation efficiency.Finally,based on the temperature compensation,a new low-power MEMS wireless gas sensor based on Zig Bee technology is designed according to the functional requirements of the gas sensor in the coal mine.In order to reduce power consumption,energy-saving strategy is proposed from the hardware selection,micro-controller power mode and data compression.Energy analysis shows that the new MEMS gas sensor has a theoretical life of up to 8 years.The data acquisition and temperature compensation,wireless communication performance of the new MEMS gas sensor are tested.The results show that the sensor is sensitive to the gas and the packet rate is 100% when communication distance is within 80m.