Application of binary logistic models for identifying high pile rebound using cone penetration test data

The large elastic rebound displacements frequently produced during the pile driving process have several structural and practical impacts; hence, identifying the occurrence probability of such displacement is a critical step in the geotechnical design phase of any piling project. In this research, two probability models for predicting potential rebound soils are developed and evaluated using logistic regression analyses to achieve this. The formulation of the logistic models was based on the use of cone penetration testing data, which were first normalized and averaged at two depth increments, one-foot, and two-feet. The normalized cone measurements utilized as input parameters thus included pore water pressure, cone penetration resistance, the friction ratio, and the in situ state parameter. The proposed models showed promising results in terms of predicting pile rebound with the prediction accuracy of Model 2, based on the two-foot averaged CPT data, being assessed at 67.4%, and the prediction rate of the Model 1, based on one-foot averaged CPT data, being even higher, at 71.3%. The results also showed that the probability of high pile rebound is functionally correlated to cone tip resistance, pore pressure, and the in situ state parameter. However, no substantial relationship was identified between the probability of pile rebound and the friction ratio.