Machine Learning for Engineering Processes

Abstract

Buildings are realized through engineering processes in projects, that however tend to result in cost and/or time overrun. Therefore, a need is highlighted by the industry and the literature, to develop predictive models, that can aid in decision-making and guidance, especially in a preparation effort before production is initiated. This study aims at investigating what are possible applications of machine learning in building engineering projects and how they impact on their performance? First, a literature review about machine learning (ML) is done. The first case is drawing on a productivity survey of building projects in Sweden (n = 580). The most influential factors behind project performance are identified, to predict performance. Features that are strongly correlated with four performance indicators are identified: cost variance, time variance and client- and contractor satisfaction and a regression analysis is done. Human related factors predict success best, such as the client role, the architect performance and collaboration. But external factors and technical aspects of a building are also important. The second case combines constructability and risk analysis on a basis on civil engineering project from several different countries and with very different character; a town square, a biogas plant, road bridges and sub projects from an airport. The data encompasses 30 projects. The development build on literature study, expert interview, unsupervised and supervised learning. The strength lies more in the conceptual work of risk sources enabled by ML. Human reasoning is needed in building projects. Also after the introduction of ML.