Neural Network Model For Design Constructability Assessment

Abstract

This paper presents an artificial neural network (ANN) technique of analysis for the assessment of design constructability. The multilayer back-propagation neural network model consists of 12 and 1 output variable. The input variables are the level of applications of constructability factors, which are sub-factors of the most important design phase constructability principles while the output variable is the level of design constructability. The development of the model goes through five main stages: identifying the design phase constructability principles, identifying the degree of importance of the constructability principles, formulating a framework for measuring the level of application of constructability principles and design constructability, collecting historical project data, and applying ANN to assess design constructability. Each stage of the model development is described. Historical project data sets related to beam construction have been collected from various contractors that have at least several years of experience in building construction. A total of 78 data sets were used to test and train the network. The determination of the optimum number of hidden nodes, hidden layers, initial weights of the links connecting the nodes, and the number of epochs for training the networks, are normally based on trial and error. The best architecture was found to consist of 12 input nodes, 5 hidden nodes, and 1 output node.