Estabishment And Optimization Of Artificial Olfaction Based On Mixed Potential Type Gas Sensor

The development of artificial intelligence technology is divided into three stages:computational intelligence,perceptual intelligence,and cognitive intelligence.Among them,perceptual intelligence technology which combines sensors with advanced algorithms is a hot research topic in artificial intelligence field.Artificial olfaction is a representative technology in the field of perceptual intelligence.This technology senses the surrounding environment through an array of gas sensors,and uses feature engineering to extract important features from the sensor output signals,from which machine learning algorithms are trained to distinguish gas types or calculate gas concentrations.This technology is widely used in environmental protection,health care,food safety and other fields.Developing high-performance gas sensors,extracting important gas-sensitive features,and constructing advanced algorithms are effective methods to improve the performance of artificial olfaction.Yttrium-stabilized zirconia-based mixed potential gas sensors have the advantages of high sensitivity,low cost,good stability,and all-solid state,etc.By developing sensitive electrode materials,constructing three-phase boundary,and changing the polarization state,the gas-sensitive performance of the sensors can be effectively regulated.Rational design of the device structure of the mixed potential gas sensor can realize the integration of the sensor and multi-signal output,which has significant potential in the practical application of artificial olfaction.Therefore,in this work,an attempt is made to establish the artificial olfaction based on the mixed potential gas sensor.The details of the research are as follows:（1）High-performance gas sensors are the basis of artificial olfaction.In this work,the potential NO₂ sensitive electrode materials are screened with the help of machine learning to establish a screening strategy for mixed-potential gas sensor sensitive electrode materials,which provides a device guarantee for artificial olfaction.In this work,a dataset of potentially sensitive electrode materials is established using materials project,and the accuracy of random forest,gradient boosting tree,weight-based extreme gradient boosting tree and gain-based extreme gradient boosting tree algorithms trained based on this dataset to determine whether the materials are sensitive to NO₂ is over 80%.More than 400 potential sensitive electrode materials were targeted and screened using the above four different algorithms,and 13 representative materials were selected as sensor sensitive electrodes to make mixed potential NO₂ sensors,and all the gas-sensitive performances of these sensors reached the level reported in recent years,which proved that the strategy of machine learning to screen sensitive electrode materials is feasible.（2）In the process of artificial olfaction,multiple signals from gas sensor array are used as the basis for determining gas type or concentration.This requires good homogeneity and stability of multiple sensitive units in sensor array.While the use of multiple sensitive units suffers from high cost and poor stability.The use of single gas sensor with multiple output signals can compensate for the above disadvantages.In this work,the Tafel curve of the mixed potential gas sensor is used instead of the mixed potential as the sensor output signal.Firstly,a finite element model of the mixed potential gas sensor is developed to efficiently output the Tafel curves of the sensor for different concentrations of acetone and ethanol gas mixtures.The accuracy of XGB algorithm based on the Tafel curve to determine whether the sensor is saturated or not reaches 99%.Two regression algorithms,XGB＿A and GBDT＿E,were trained to identify the concentration of acetone and ethanol in the gas mixture using the Tafel curves.The average percentage error of the above two algorithms to identify the actual measured Tafel curves to determine the concentration was 11.3%and 23%,respectively.The artificial olfaction established based on the Tafel curve successfully identifies the concentration of each component of ethanol acetone gas mixture.This method can be used to establish the artificial olfaction to achieve the detection of other gases.（3）For mixed potential gas sensors,multiple output signals can be generated by constructing multiple sensitive electrodes or changing the polarization state of sensitive electrodes.This method can replace sensor arrays to establish artificial olfaction.In this work,the structure of sensor device is optimized so that a single sensor can generate several different output signals.In order to achieve the above functions,three different device structures of mixed potential type gas sensors are designed and fabricated as follows:dual sensitive electrode polarization type,dual lead polarization type,and side reference electrode polarization type.The dual-sensitive electrode polarization type gas sensor mainly consists of dual-sensitive electrode,counter electrode,reference electrode and YSZ electrolyte.Two different sensitive electrodes can produce two different negative responses for VOCs gases.And the gas-sensitive performance of the sensor is regulated by applying a bias voltage to the counter electrode.The linear discriminant analysis is used to reduce the dimensionality of the features,and the K-nearest neighbor algorithm is trained to distinguish the five VOCs gas species with 100%accuracy.The double-lead polarization type gas sensor mainly consists of a sensitive electrode,a counter electrode,a reference electrode,and a YSZ electrolyte.The counter electrode is located on the same side of the YSZ substrate as the sensitive electrode,and the counter electrode changes the polarization state of the three-phase boundary to achieve multiple outputs of the sensor.The polarization state of the three-phase boundary can also be adjusted to regulate the response value,sensitivity,and response recovery time of the sensor.After polarizing the three-phase boundary,the operating temperature of the sensor is significantly reduced,and the power consumption of the sensor is reduced from 1.14 W to 0.625 W.The random forest algorithm trained on the 24 gas-sensitive features can distinguish 9 different VOCs gases with 99%accuracy.The side reference electrode polarization type gas sensor mainly consists of a sensitive electrode,a counter electrode,a reference electrode,and a YSZ electrolyte.The counter electrode and the reference electrode are located on the same side of the YSZ substrate.The polarization state on one side of the three-phase boundary of the sensor can be changed by applying a bias voltage to the counter electrode.So that the sensor can generate two different responses,one positive and one negative.The support vector machine algorithm based on this device with an accuracy of 98.6%in determining nine VOCs gases.Meanwhile,this kind of sensor has a second working mode.The lead wire on the same side of the counter electrode can be saturated by controlling the bias voltage,and the response values for ethanol,acetone and formaldehyde in the concentration range of 10～150 ppm are maintained at 31m V,18m V and 9m V,respectively.The output signal of this sensitive electrode can be used to determine the gas types.And the output signal of the other sensitive electrode is used to calculate the gas concentration.Under this working mode,the gas type and concentration can be determined simultaneously without using complicated algorithms.In summary,based on the nature of the sensitive mechanism of the mixed potential gas sensor,this thesis establishes and optimizes the artificial olfaction based on the mixed potential gas sensor in three aspects:screening new sensitive electrode material,changing the sensor output signal,and optimizing the sensor device structure,so as to lay the foundation for the practical application of this type of sensor.