Machine learning approach for approximating design parameters from engineering charts

Geotechnical engineers have traditionally relied on engineering charts for the analysis and design of specific geotechnical problems. However, interpolating target design parameters, particularly on logarithmic scale charts, can be time consuming and susceptible to human error. Recent advancements in machine learning enable engineers to efficiently approximate design parameters by training models on extensive datasets, thereby minimizing both time and manual intervention. Furthermore, coefficients for closed-form equations can be derived from these models in some cases, streamlining computational analysis and enhancing design workflows. This paper presents two case studies: one focused on shallow footing settlement assessment and the other on single pile settlement assessment. It illustrates the application of non-linear regression, high-degree polynomial regression, Gaussian process regression and fully connected neural networks in developing effective machine learning models for graphical approximation.