Bayesian optimization for resistance factor of piles

According to the mathematical statistics theory and Bayesian technique, a method for data processing and optimization in geotechnical engineering is put forward to solve the problem caused by model uncertainty due to lack of enough accurate field data. Meanwhile, the data of bearing capacity of piles in non-cohesive soils (29 piles) and in cohesive soils (59 piles) in South Africa are collected. The model factor of the bearing capacity is defined as the ratio of the measured capacity to the predicted one. By means of the proposed method, the collected data are sorted into three categories, which are good data, ordinary data and abnormal data. The abnormal data are discarded because of their adverse influence on calculation, and the ordinary data are optimized. The first order second moment method, the advanced first order second moment method and the Monte Carlo simulation are employed to calculate the reliability of the bearing capacity. The calculated results show that the sorting and the optimization of data have great influence on the calculated results of reliability and resistance factors. For instance, the calculated results using the good data and the optimized data are larger than those using other data. Finally, the recommended values of resistance factors of driven piles are suggested according to the calculated results and American specifications for load and resistance factor design. The proposed method can offer references to the researchers and for the amendment of relevant specifications.