Formwork pressure prediction in cast-in-place self-compacting concrete using deep learning

Abstract

The prediction of formwork pressure exerted by self-compacting concrete (SCC) remains a challenge not only to researchers but also to engineers and contractors on the construction site. This article aims to utilize shallow neural networks (SNN) and deep neural networks (DNN) using Long Short-Term Memory (LSTM) approach to develop a prediction model based on real-time data acquitted from controllable laboratory testing series. A test setup consisting of a two-meter-high column, ø160 mm, was prepared and tested in the laboratory. A digital pressure monitoring system was used to collect and transfer the data to the cloud on a real-time basis. The pressure was monitored during- and after casting, following the pressure build-up and reduction, respectively. The two main parameters affecting the form pressure, i.e., casting rate and slump flow, were varied to collect a wide range of input data for the analysis. The proposed model by DNN was able to accurately predict the pressure behavior based on the input data from the laboratory tests with high-performance indicators and multiple hidden layers. The results showed that the pressure is significantly affected by the casting rate, while the slump flow had rather lower impact. The proposed model can be a useful and reliable tool at the construction site to closely predict the pressure development and the effects of variations in casting rate and slump flow. The model provides the opportunity to increase safety and speeding up construction while avoiding costly and time-consuming effects of oversized formwork.