Abstract: Expansive soil is prevalent in the Chengdu area, and rainfall is a significant factor triggering deep foundation pit engineering accidents. Through the investigation of engineering cases, the typical deformation and failure processes of deep foundation pits in the Chengdu area, supported by piles under rainfall infiltration, were analyzed and summarized. The measured horizontal deformation curves of supporting piles in foundation pit engineering were categorized into three types: steep, gradual, and stable types. The early risk prevention of deep foundation pits in Chengdu’s expansive soil area can be based on the horizontal deformation of supporting structures, as indicated by the deformation developments of unstable pits and the three types of measured deformation curves during early rainfall after excavation. Using wavelet analysis, artificial neural networks, and Copula random variable correlation analysis, a prediction model for the horizontal deformation of supporting structures in foundation pits, considering rainfall influence, was established. The actual deformation curves of deep foundation pits were predicted based on this model. Finally, the prediction results enable early risk warnings based on deformation predictions. The predicted deformation results align well with the measured data, preliminarily confirming the validity of the proposed model. Using the same deformation warning index, the risk warning based on predicted deformation can significantly advance the warning time, providing a basis for optimizing the treatment scheme.

Key words: Chengdu expansive soil, deep foundation pits, foundation pit stability, deformation prediction, neural networks