Supporting the Observational Approach in Construction through Bayesian Analysis

Abstract

The validation of design assumptions, as construction works are being carried out, is a vital component in geotechnical engineering. The feedback can be in terms of either the prescribed quality control tests, instrumentation and monitoring, or other site observations. As most of these can be evaluated through systematic logic, the use of data science methods or machine learning procedures is rarely necessary. There are, however, exceptions especially for cases in which: (i) there are several possible causes to the problems which are hard to pinpoint precisely, (ii) the quantum of data is overwhelming, and (iii) there is scatter in the of observed outcomes. Where these features are encountered, it is generally more efficient to process the data using a computer. This paper presents a possible way of interpreting the feedback obtained through observations in construction, via Bayesian programming, which is one of the many methods in machine learning. A case history discussing the performance of ground anchors in a deep excavation project is discussed.