Predicting geotechnical axial capacity of reinforced concrete driven pile using machine learning technique

• Main Authors: Xun, Ooi Zi, Abdullah, Rini Asnida
• Format: Article
• Language: English
• Published: Penerbit UTM Press 2023
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utm.my/108527/1/XunOoiZi2023_PredictingGeotechnicalAxialCapacityofReinforced.pdf; http://eprints.utm.my/108527/; http://dx.doi.org/10.11113/mjce.v35.20544

Modified Meyerhof method is a popular method to calculate pile geotechnical axial capacity in Malaysia currently. From past experience, pile design based on empirical and analytical method produce variability of predicted capacity, in which, there is a wide scatter of predicted capacities and tendency for the predictions to be conservative, i.e. to underestimate the load capacity. This study provides options of machine learning and statistical approach for prediction of pile capacity based on soil investigation and dynamic pile load test result. It serves as an additional checking for engineer during design of pile based on conventional empirical method. It also helps to provide deeper insights of non-linear variables related to pile capacity through machine learning and statistical approach. This study helps engineer to design pile foundation optimally, economically and safely. The prediction of pile geotechnical axial capacity with machine learning technique and statistical approach for local marine clay soil in Penang, Malaysia is proposed in this study. The information from soil investigation report and dynamic pile load test report are gathered from six projects at Batu Kawan and Nibong Tebal located in Penang state that contributed 439 numbers of data. The skin friction factor, end bearing factor and pile geotechnical axial capacity are computed and predicted using empirical method, machine learning model and statistical model. Support Vector Machine illustrate best fit model for predicting skin friction factor with R2 of 0.517 while Random Forest seems to be the best fit model for predicting end bearing factor with R2 of 0.264. Random Forest is found to be the best model in predicting the geotechnical pile axial capacity compare to other models as it explains 96.2% of the variability of pile capacity.