Time series clustering for TBM performance investigation using spatio-temporal complex networks

This paper proposes a network-enabled approach for analyzing time series data related to tunnel boring machine (TBM) excavation behavior in the nature of high dimensionality and nonlinearity. To fill the gap between time series data analysis and complex network theory, the main objective is to capture spatio-temporal patterns of TBM dynamic excavation behavior from a topological structure perspective, which can provide valuable insights into geological information and over excavation ratio for intelligent tunneling project management. To accomplish this goal, the principal component analysis (PCA) is firstly utilized to reduce the dimension of the multivariate and heterogenous dataset for simplicity. Afterward, the time-lagged cross correlation (TLCC) and dynamic time wrapping (DTW) are implemented to measure the similarity between two segment rings for network graphing, resulting in a holistic view of the TBM excavation performance. Through the case study, network analysis results indicate that: (1) Leiden outperforms other state-of-the-art community detection algorithms in dividing the whole network into four high-quality communities. (2) There is a trend for segment rings with more similar excavation behavior and geological conditions to be gathered into the same community. (3) Since the over excavation ratio in four communities derived from the DTW-based complex network is proven to be significantly different, global sensitivity analysis is deployed to find out the most crucial features for decisionmaking in TBM control. The novelty to be highlighted is the developed time series analysis approach relying on the complex network perspective, which is helpful in effectively detecting relationships along with hidden engineering knowledge among rings. This has potential value in better understanding and improving the TBM tunneling performance under the underground environment with great complexity and uncertainty.

Automated summary

This paper proposes a network-enabled approach for analyzing time series data related to tunnel boring machine (TBM) excavation behavior. The main objective is to capture spatio-temporal patterns of TBM dynamic excavation behavior from a topological structure perspective. The novelty is the developed time series analysis approach relying on the complex network perspective.