Deep Learning Based Interference Suppression Method for Surface Wave Detection Using Tunnel Boring Machine (TBM) Drilling Noise

Abstract

Tunnel Boring Machines (TBMs), characterized by superior advancement rates and integrated operational sequences, have emerged as the predominant construction methodology for urban subsurface tunnel development. However, TBMs are less capable of coping with faults, isolated rocks, and other adverse geological conditions common to shallow urban subsurface. This has led to serious challenges in the safe construction and disaster prevention of urban subsurface tunnels. Due to the strong energy and large amplitude of surface waves, analyzing them provides an effective way to characterize the subsurface and ensure the safety of urban shallow tunnel construction. Inspired by the seismic-while-drilling method, surface-wave detection using tunnel-drilling noise has been widely applied due to its advantages of real-time detection and non-invasiveness. However, it is difficult to remove the body-wave interference signals brought about by factors such as the excitation mode of the seismic source, which in turn affects the subsequent extraction of the dispersion curves. Thus, we analyze the three main types of interference signals present in tunnel drilling noise-based surface-wave data, and a synthetic dataset incorporating realistic field interference signal characteristics is constructed. Then, we achieve the suppression of interference signals by constructing a special convolutional self-encoder that pays attention to the morphological features of the signal. Besides, a transfer learning method is proposed to suppress the inter-trace interference due to irregular acquisition and to make the method directly applicable to field data. The results of synthetic and field data processing show that the proposed method has good denoising capability and exhibits higher accuracy in subsequent dispersion-curve extraction.