Risk assessment approach for tunnel collapse based on improved multi-source evidence information fusion

Tunnel collapses are common hazards in construction, and they significantly constraint construction progress and safety. Currently, research on the risk assessment of tunnel collapses primarily relies on a single source of information, which leads to distorted evaluations owing to the limitations of data sources. In contrast, using multi-source information offers strong adaptability, high credibility, and complementarity. Therefore, to enhance the accuracy of tunnel collapse risk assessments, this study proposes a novel approach that combines three types of information sources: historical engineering cases, expert knowledge, and on-site practical information. First, artificial neural networks, knowledge evaluation matrices, and cloud models are used to extract evidence from the three types of information sources, thereby acquiring preliminary evidence of collapse risk. When extracting expert knowledge information, an improved similarity aggregation method that comprehensively considers judgment ability and recognition is proposed to reduce the impact of expert subjectivity. Next, to address evidence conflicts in the fusion process, a distance metric based on belief intervals is constructed to calculate evidence credibility, and evidence importance is incorporated to reconstruct the multi-source evidence information. Subsequently, Dempster's synthesis rule is used to fuse the reconstructed evidence, and the collapse risk is calculated by deblurring the fusion results. Finally, the proposed method is applied to the Yanglin Tunnel in China, and the results are consistent with the onsite construction situation. Therefore, the proposed method is feasible and practical, and it can provide a valid reference for risk assessment in similar projects.