Using Machine Learning of Sensor Data to Estimate the Production of Cutter Suction Dredgers

Production estimation (the excavated soil per time) helps dredging companies to be able to manage the dredging projects efficiently and enables them to predict the time and cost of the project. They can calculate the production with the density and flow sensors which are installed in a dredging ship. However, due to the high price of the density sensor, many companies avoid purchasing vessels with the density sensor and look for a cheaper alternative. In this study, we explore an alternative way to predict density by leveraging data used during production and applying machine learning algorithms on them. In this article we use a dataset that belongs to a Cutter Suction Dredger (CSD) operating in an African country. Our results exceed a prediction accuracy of 80%. However, our models do not predict the density of the dredger operating in a different location. The reason is that the features used to estimate the density classes are influenced by various factors such as ambient conditions, water dynamics, fuel quality, and soil properties and these factors vary based on the region. Therefore, one of the main contributions of our work is checking the generalizability of our trained models and explaining the features that are important for predicting soil density.