Incorporation of knowledge and data-driven models applied in shield tunneling: A review

Abstract

Data-driven models have undergone extensive exploration in addressing shield tunneling challenges, propelled by advancements in sensing technology and machine learning (ML) techniques. However, relying solely on data-driven approaches for shield tunneling control presents issues of physical inconsistency, poor interpretability, and a reliance on high-quality and sufficient data. This review meticulously examines the optimization of ML models through the integration of knowledge, tailored to the shield tunneling domain. First, the types of knowledge involved, encompassing world knowledge, scientific knowledge, and empirical laws, are defined and elucidated. Second, existing practices aimed at tackling main issues within this domain, including environmental impacts, geological conditions, and shield operation performance, are elaborated. Subsequently, the fusion strategies based on the ML pipeline are exploited. Building upon this, the challenges and future directions of this innovative model, including knowledge compilation and utilization, model development and evaluation, and practical application in shield construction are discussed. This review deepens the understanding of data and knowledge fusion methods, providing new insights into the development of this approach for aiding in shield tunnel projects.