Study On Single-walled Carbon Nanotubes Gas Sensor Array And Its Characteristics For Gas Detection In Transformer Oil

Dissolved Gas Analysis(DGA)results provide powerful evidence for power transformer condition detection.Gas sensing technology is the key to achieving on-line analysis of dissolved gases in transformer oil via DGA.The high-performance gas sensing method combined with the gas sensing array technology is the bottleneck for online monitoring of dissolved multi-component gases in transformer oil.In this thesis,Single-Walled Carbon Nanotubes(SWCNTs)and their functionalized derivatives(hydroxyl derivatives,carboxyl derivatives,aminated derivatives via ethylenediamine,aminated derivatives via aniline,Ni-coated,Pd-doped,and ZnO-doped)are applied as gas sensing materials,developed for a single atmosphere and a mixed atmosphere of H₂,CO,and C₂H₂.A SWCNTs-based gas sensor array composed of eight gas sensor array units was fabricated;The gas sensing characteristics of the SWCNTs-based gas sensor array in a single atmosphere and a mixed atmosphere of H₂,CO,and C₂H₂ were tested;The application of machine learning algorithm in the qualitative identification and quantitative analysis of the multi-component mixed atmosphere dissolved in transformer oil was stuidied;The application of machine learning algorithm in the prediction of the service life and the correction of stability of SWCNTs-based gas sensor arrays were explored.The main work and research results of the thesis are as follows:(1)Eight different SWCNTs-based gas sensing materials were selected(unfunctionalized SWCNTs,hydroxyl derivatives,carboxyl derivatives,aminated derivatives via ethylenediamine,aminated derivatives via aniline,Ni-coated,Pd-doped,and ZnO-doped).Based on the planar gas sensor,eight planar gas sensor array units were designed and manufactured,and a SWCNTs-based gas sensor array was integrated through MEMS technology.(2)The temperature characteristics,concentration characteristics,response and recovery characteristics,and selectivity of the SWCNTs-based gas sensor array in a single atmosphere of H₂,CO,and C₂H₂ were tested.The results of gas sensing experiments show that different functionalization methods have different effects on the gas sensing characteristics of each unit of the SWCNTs-based gas sensor array in the single atmosphere of H₂,CO,and C₂H₂.ZnO-doped SWCNTs-based gas sensor array unit exhibits the best gas sensing characteristics in the single atmosphere of H₂ or C₂H₂that is,the optimal working temperature is the lowest(200°C),and the gas sensing response is the highest with the shortest response and recovery time.Ni-coated SWCNTs-based gas sensor array unit exhibits the best gas sensing characteristics in the single CO atmosphere with the highest gas sensing response value and the shortest response and recovery time at the optimal operating temperature(225°C).Eight SWCNTs-based gas sensor array units all show good linearity in the single atmosphere of H₂,CO,and C₂H₂,but also show significant selectivity differences.(3)The concentration characteristics and stability of the SWCNTs-based gas sensor array in a mixed atmosphere of H₂,CO,and C₂H₂ were tested.The gas sensing experiment results show that the gas sensing response value of the SWCNTs-based gas sensor array in the muti-component atmosphere is larger than that in the three-single atmosphere,but it is not a linear superposition.Functionalization improves the long-term stability of SWCNTs-based gas sensor array units in the mixed atmosphere of H₂,CO,and C₂H₂,and the long-term stability of ZnO-doped SWCNTs-based gas sensor array units is the best which presnets the longest service life(1531 times).(4)Combining the experimental results of the concentration characteristics of the SWCNTs-based gas sensor array in the muti-component atmosphere of H₂,CO,and C₂H₂,three machine learning algorithms(Back Propagation Neural Network BPNN,Deep Belief Network-Deep Learning Network DBN-DLN,and Ensemble Learning Boosting-Tree)realize the qualitative identification and quantitative analysis of multi-component mixed atmosphere.The results show that the Ensemble Learning Boosting-Tree model is the most suitable method for the qualitative identification of multi-component mixed atmospheres.The Deep Belief Network-Deep Learning Network DBN-DLN mode is the best method for the quantitative analysis of multi-component mixed atmospheres.(5)Based on the experimental results of SWCNTs-based gas sensor arrays in a muti-component atmosphere of H₂,CO,and C₂H₂,two machine learning algorithms(Boosting-BP and Random Forest RF)were used to achieve the prediction of the service life of the SWCNTs-based gas sensor array in the muti-component atmosphere of H₂,CO and C₂H₂.Combined with the Deep Belief Network-Deep Learning Network DBN-DLN model,the stability of the SWCNTs-based gas sensor array after long-term uses in the muti-component atmosphere of H₂,CO,and C₂H₂ was corrected successfully.