Measuring and benchmarking the productivity of excavators in infrastructure projects: A deep neural network approach

Inefficiencies in the management of earthmoving equipment greatly contribute to the productivity gap of infrastructure projects. This paper develops and tests a Deep Neural Network (DNN) model for estimating the productivity of excavators and establishing a productivity measure for their benchmark. After investigating current practices for measuring the productivity of earthwork equipment during 13 interviews with selected industry experts, the DNN model was developed and tested in one of the 'High Speed rail second phase' (HS2) sites. The accuracy of prediction achieved by the DNN model was evaluated using the coefficient of determination (R2) and the Weighted Absolute Percentage Error (WAPE) resulting in 0.87 and 69.64%, respectively. This is an adequate level of accuracy when compared to other similar studies. However, according to the WAPE method, the accuracy is still 10.36% below the threshold (i.e. 80%) expected by the industry experts. An inspection of the prediction results over the testing period (21 days) revealed better precision in days with high excavation volumes compared to days with low excavation volumes. This was attributed to the likely involvement of manual work (i.e. archaeologists in the case of the selected site) alongside some of the excavators, which caused gaps in telematics data. This indicates that the accuracy attained is adequate, but the proposed approach is more accurate in a highly mechanised environment (i.e. excavation work with equipment predominantly and limited manual interventions) compared to a mixed mechanised-manual working environment. A bottom-up benchmark measure (i.e. excavation rate) that can be used to measure and benchmark the excavation performance of an individual or a group of equipment, through a work area, to a whole site was also proposed and discussed.