Abstract: With the rapid development of underground projects such as subway construction in the major cities, much attention has been paid to the health maintenance of the shield equipment. As one of the key components of the shield machine, the cutterhead is easy to wear out, but difficult to be measured directly, and may significantly affect the efficiency and delay the project of shield tunneling. The traditional analysis methods based on the cutter mechanism models usually have poor performance due to various practical engineering conditions and complicated structures of the shield machine system, which makes it difficult to accurately evaluate its health status. In this paper, a data-driven method for evaluating the health of the shield machine cutterhead is proposed, which tries to model the relationships between the sensor data of the shield machine and the health state of the cutterhead, and to quantify the degradation of the cutterheader's performance by using the t-SNE (t-distribution stochastic neighbor embedding) model. The main steps of the proposed method include:① the sensor data pre-processing and initial feature extraction; ② the analysis of the intrinsic manifold distribution in the feature space, and the optimization of the low dimensional feature space obtained by the t-SNE model; ③ a Mahalanobis distance-based metric is designed to quantify the performance degradation of the cutterhead in the optimized feature space. The results have shown that the proposed method can accurately evaluate the performance of the cutterhead based on the actual operation data of the shield equipment.

Key words: health evaluation, Mahalanobis distance, shield equipment, t-distribution stochastic neighbor embedding