Research on Gas Detection Using Independent Component Analysis
|School||Harbin Institute of Technology|
|Course||Instrument Science and Technology|
|Keywords||Independent component analysis Gas sensor array Cross sensitivity Back propagation neural network Multiple Regression Analysis|
As the development of environmental science and the environmental detection technique, the cheap measuring gas sensors with good performance have become the new direction of measurement field. Caused by the physical shortcomings of gas sensors, it is hardly for a single semiconductor gas sensor to identify multiple gases. Therefore, using gas sensor array and BSS (Blind Source Separation) algorithm to measure and recognize gases has an enormous applied value. The BSS algorithm plays the crucial role to the behavior of gas identification.Firstly, the research introduces the development and principle of gas measurement and BSS theory, and then goes into particulars of gas identification experiment designing, including the methods of hardware and software designing. Secondly, the research presents ANN (Artificial Neural Networks) and MLR (Multiple Regression Analysis) algorithms, showing the short comes of them, deviating the ICA (Independent Component Analysis) algorithm. The research mainly focuses on recognition technology of ICA, introducing the choice of objective function and common optimization algorithms. Furthermore, FastICA (Fast Independent Component Analysis) is brought and applied to gas measurement and reorganization, the research also gives the processes and flow charts of the programming. Finally, through the cooperating with CETC-49, the research builds experiment system, choosing CH4 and H2 as the samples to measure and recognize single gas and mixed gases respectively. The results of experiment show that the qualitative identification rate reaches 100% in using ICA algorithm, in addition, the research gives the errors and the reasons of quantitative detection and compares with the results from ANN and MLR.