Identifying characteristics of pipejacking parameters to assess geological conditions using optimisation algorithm-based support vector machines

Detecting sudden changes in geological conditions (e.g., karst cavern and fault zone) during tunnelling is a complex task. These changes can cause shield machines to jam or even induce geo-hazards such as water ingress and surface subsidence. Tunnelling parameters that relate closely to the surrounding geology have proliferated in recent years and present a substantial opportunity for the application of data-driven artificial intelligent (AI) techniques that can infer patterns from data without reference to known, or labelled, outcomes. This study explores the potential for support vector machines (SVM) to identify changes in soil type during tunnelling towards reducing the possibility of jamming and geo-hazard development. All tunnelling data were pre-processed to convert time series data into feature-based sub-series. A selection of the most popular parameter optimisation algorithms was explored to improve the accuracy of the AI predictions. Their relative merits were evaluated through comparisons with a recent pipejacking case history undertaken in gravel and clayey gravel soils. The results highlight an exciting potential for the use of optimisation algorithm-based SVMs to identify changes in soil conditions during pipejacking.