Research On MEMS Sensor Array And Algorithm For Artificial Olfactory System

In coal production,it is inevitable to encounter inflammable,explosive,poisonous and harmful gases.Explosion or poisoning caused by gas leakage occurs from time to time,which seriously endangers personal safety.Therefore,gas detection is a necessary means to prevent the occurrence of safety accidents.Conventional gas detection means are faced with the shortcomings of expensive instruments,cumbersome operation and poor sensitivity,so it is not very suitable for the complex background of mine.But the rapid development of artificial olfactory systems has enriched gas detection methods.At present,the research on artificial olfactory system mainly focuses on sensor preparation,pattern recognition algorithm,application equipment and so on.With the development of the Internet of Things,people’s demand for miniaturized,low-power and intelligent artificial olfactory equipment becomes more and more urgent,and the preparation of low-power and miniaturized gas sensor is an important premise to meet this demand.Considering the Micro Electro Mechanical System(MEMS)’s attains in miniature and precision,this thesis prepared MEMS gas-sensitive sensors for methane and carbon monoxide gas based on MEMS technology,composed the sensor array,and after a number of comparisons,A relatively optimal pattern recognition algorithm was determined,and an artificial olfactory system with low power consumption,miniaturization and high sensitivity was successfully constructed to detect mixed gases.The main research work is as follows:(1)A low-power,miniaturized MEMS gas-sensitive sensor is studied to solve the problems of high power consumption,large size and difficult batch production of sensors in artificial olfactory systems.Nano-sized gas sensitive films were prepared by magnetron sputtering as the gas sensitive material layer in MEMS gas sensitive sensors.In order to improve the selectivity of the sensors to methane and carbon monoxide gas,two methods of modifying gas sensitive films with precious metals were proposed.After evaluating the performance of MEMS gas sensor,the suitable sensors were screened and the MEMS gas sensor array was constructed as the sensing front end.(2)Research on mixed gas pattern recognition algorithm based on machine learning.Based on the MEMS gas sensor array,an artificial olfactory system was constructed.Mixed gas response data is collected,and traditional machine learning is used as the core algorithm model to complete the recognition of mixed gas.Traditional pattern recognition needs to be based on the feature data set.In this thesis,the feature data set is made by manually extracting features and combining with the principal component analysis method to reduce the data dimension.The classification results of mixed gas are obtained by using the Support Vector Machine(SVM)and BP neural network model.Experiments show that the classification accuracy of the two models can reach more than 80% in the case of small samples,and the classification accuracy of SVM algorithm is higher.(3)Research on mixed gas pattern recognition algorithm based on deep learning.In view of the complicated steps of traditional pattern recognition,such as signal processing,feature extraction and feature selection,a mixed gas recognition method based on SG filtering and one-dimensional convolutional network is proposed.At the same time,considering the deep application of image recognition network,onedimensional response signal of MEMS sensor array is converted into image by Gram angular difference field(GADF),and mixed gas recognition is completed by combining with image recognition network.The results show that one-dimensional convolutional neural network is the most suitable for sensor data to realize gas mixture recognition,and complete classification can be realized under small samples.Other methods also have recognition accuracy of more than 90%.(4)Comparative analysis of experiments.The artificial olfactory system of different pattern recognition algorithms was tested and compared,the most suitable pattern recognition algorithm was selected,and the completed artificial olfactory system used to recognize mixed gas was constructed.