Abstract: Identifying the geological conditions ahead of the cutterhead during slurry balance shield tunneling in real time is challenging. To address this, a case study is conducted on the water supply pipeline and road improvement project at Zhijiang road in Hangzhou, China. Tri-axis accelerometers are installed on the shield′s main drive to collect vibration signals. The correlation between these signals and the five typical stratum profiles is examined. The original signals are divided into several times segments, with one rotation of the cutterhead as a cycle, thereby constructing a feature set. Finally, a method combining principal component analysis (PCA) and support vector machine (SVM) is proposed to identify geological types in real time ahead of the cutterhead. The results reveal the following: (1) Geological types considerably affect vibration signals, with the highest vibration amplitude along the Z-axis (tunneling direction). The root mean square acceleration across the three axes shows distinct clustering characteristics in cutterhead torque and penetration parameters under different stratum conditions. (2) The PCA-SVM method achieves an identification accuracy of 97.98% for these five typical stratum types along the project. The stratum identification accuracy using time and frequency domain features extracted by time periods as input parameters is 23.36% higher than that using the original acceleration signals as input parameters. (3) The PCA algorithm effectively reduces data volume, shortens model training time by approximately 29.2%, and improves identification accuracy by approximately 0.3%.

Key words: shield tunneling, stratum identification, vibration signal, principal component analysis-support vector machine method