Sensor-based Pavement layer change detection using Long-Short Term Memory (LSTM)

Abstract

During construction, pavement projects often suffer from a lack of progress certainty, which leads to cost and time overruns. The pavement construction progress should be monitored in a timely and accurate manner to provide prompt feedback and ensure project success. However, current pavement construction progress monitoring practices (e.g., data collection, processing and analysis) are manual, time-consuming, tedious, inconsistent, subjective and error-prone. The previous research study was limited to only incremental road construction progress measurement. This preliminary study proposes a novel sensor-based method to identify pavement layer changes during construction using a time series algorithm for the approach development of automated as-built measurement of road construction. In this study, data were collected from generating various road construction scenarios in a controlled environment by simulating layer changes using a ground vehicle equipped with a laser ToF (time-of-flight) distance-ranging sensor. Subsequently, Long Short Term Memory (LSTM) was utilized on collected data for feature detection as 'layer up', 'layer down' and 'layer not changed' to classify road layer change. The experimental result demonstrates 84.91% as a promising overall average accuracy of road layer change classification on the control environment data, confirming the potential implementation suitability to detect pavement layers in real pavement construction projects. However, low-performance measures (low precision, recall and F1 score) of layer up and layer down suggest further improvement to enhance the robustness of the proposed model. This method can be extended to automate pavement construction progress monitoring by validating the proposed approach in a real case.