Determination of Automated Construction Operations from Sensor Data Using Machine Learning (original) (raw)

Automated construction creates an intricate working environment involving workers and machines. The added complexity of automated construction demands a rigorous monitoring system compared to conventional construction. The first stage of developing such a monitoring system is the identification of construction operations. This paper discusses a methodology for the identification of construction operations from sensor data. The methodology is illustrated using the case study of a coordinated lifting equipment implemented in a laboratory. The data is collected from a small scale structural frame consisting of steel modules in a controlled laboratory condition. The automated system follows a top-down construction method where the major construction operations are performed at the ground level and the structure is lifted upwards in stages. Strain and acceleration measurements were collected from the structure during construction. Each operation is associated with a unique pattern of measurements at each sensor location. The measurement data is used for analysis by support vector classification. Parameters like error penalty (C) and width of Gaussian kernel (σ) were varied to obtain the best prediction results. The results of the analysis show that the linear classification gives better results compared to the nonlinear classification for all operations except coordinated lifting. However, coordinated lifting is the best-predicted operation with an accuracy of 96%. Selection of optimal values of C and σ enhances the accuracy of classification. The features extracted from data seems to highly influence the learning of the algorithm and the performance of prediction. The results show the potential for using machine learning techniques for monitoring automated construction operations.