Shallow Karst Cave Identification Technique With Pile Hammer Excitation Based On Machine Learning

Because the construction of infrastructure requires stable foundation,so the existence of caves has great hidden trouble for infrastructure,which might cause immeasurable loss when we cannot explore it in time.Although there are many mature ways to detect caves such as seismic reflection method,ground penetrating radar,standing wave tube method,etc.However,those methods still have some shortcomings.Ground penetrating radar method requires higher cost,and seismic reflection method highly relies on inspectors’ experiences,and standing wave tube method has limited detection range.So it’s very important to explore a high-safety,high-efficiency and low-cost intelligent cave detection technique.With the development of science and technology,machine learning has become a hot research area because of its self-learning ability.On the one hand,the principle of shallow seismic reflection method is used to obtain the reflection signal of caves from a small number of sensors on the ground surface,by using the pile hammer shock during foundation construction as excitation instead of explosives.On the other hand,through finite element simulation a large number of ground acceleration response data under various scenarios are obtained.Subsequently,the intrinsic features of the time history data are extracted as the input for machine learning algorithms.Thus,a high-efficiency and low-cost detection method of caves can be realized.Three machine learning methods such as decision tree,random forest and k-nearest neighbor was used to build up this model.The results show that the k-nearest neighbor model based on the six-sensor placement strategy achieves the highest accuracy.The accuracy of the prediction of the location and diameter for caves can reach 98.1% with a tolerance error of two meters.Later we studied that when there was proper layout error for the sensor,the model’s accuracy was not affected.So the innovation point of this model is that we replace the power with hammer-hit,which can improve safety and reduce the cost.Furthermore,with the help of machine learning,the output of the model is more objective.This paper verifies the feasibility of developing a high-efficiency and low-cost intelligent cave detection technique,which can provide technical support for the assessment of geological conditions in the early stage of construction.