Towards Gas Mapping And Gas Location In Confined Space Using An Olfactory Quadruped Robot

Confined spaces are a special category of operating environments that restrict the movement of operators and present potential risks.Confined spaces usually refer to facilities and places such as mines,chemical plant equipment,urban sewers,etc.Operating in confined spaces plays an important role in promoting economic development and protecting the normal and orderly operation of cities.However,in restricted spaces,operators are prone to oxygen deprivation or poisoning,which can seriously affect the safety of operators.Knowing in advance the trend of gas change and gas distribution in restricted space is helpful for managers to arrange the operation time reasonably and protect the life and property of operators.In view of this,this thesis conducts a study on gas mapping and source determination under restricted space.The main contents include:(1)To obtain the trend of gas concentration changes in future moments,this paper combines gated recurrent network,cross-attention mechanism and flamingo search algorithm,and proposes cross-attention gated recurrent network optimized based on flamingo search algorithm(FSCA-GRU).The correlation between different variables are captured by the cross attention mechanism,and the weights of network are optimized by using flamingo search algorithm,which improves the prediction accuracy.Simulation experiments were conducted using a public data set,and the experiments verified the effectiveness and advantages of the improved model in multivariate time series prediction tasks.(2)In order to grasp the gas distribution conditions at future moments,the TD Kernel DM+V algorithm was used to unify the time scales of the predicted gas concentration values and plot the gas distribution conditions at different moments based on the predicted gas concentration values.The number of gas molecule hits at different moments of the gas distribution map is analyzed together to make an accurate estimate of the location of the gas source.Simulation experiments demonstrate that the extrapolation of the TD Kernel DM+V model can accurately estimate the location of a single gas source,and the source determination method based on the number of odor molecule hits is better than the source determination method based on the concentration extremes.(3)In order to explore the gas concentration trends and gas distribution in confined space,a simulated confined space environment was built indoors,and an active suction electronic nose was designed for environmental sampling using a quadruped robot as a motion carrier.The simulation experimental results show that the FSCA-GRU algorithm can achieve high prediction accuracy by taking into account the dependence between multiple variables comprehensively,and the TD Kernel DM+V algorithm can provide an accurate estimation of gas distribution and gas source location for a single gas source,but the gas source location deviates from the actual gas source location in a multi-gas source task.