A Machine Learning Approach to Predictively Determine Filter Clogging in a Ballast Water Treatment System

Abstract: Since the introduction of the Ballast Water Management Convention, ballast water treatment systems are required to be used on ships for processing the ballast water to avoid spreading bacteria or other microbes which can destroy foreign ecosystems. One way of pre-processing the water for treatment is by straining the water through a filtration unit. When the filter mesh retains particles, it begins to clog and could potentially clog rapidly if the concentration of particles in the water is high. The clog jeopardises the system. The thesis aims at investigating if machine learning through neural networks can be implemented with the system to predictively determine filter clogging by investigating two popular network structures for time series analysis. The problem came down to initially determine different grades of clogging for the filter element based on sampled sensor data from the ballast water treatment system. The data were then put through regression analysis through two neural networks for parameter prediction, one LSTM and one CNN. The LSTM predicted values of variable and clogging labels for the next 5 seconds and the CNN predicted values of variable and clogging labels for the next 30 seconds. The predicted data were then verified through classification analysis by an LSTM network and a CNN. The LSTM regression network achieved an r 2 -score of 0.981 and the LSTM classification network achieved a classification accuracy of 99.5%. The CNN regression network achieved an r 2 -score of 0.876 and the CNN classification network achieved a classification accuracy of 93.3%. The results conclude that ML can be used for identifying different grades of clogging but that further research is required to determine if all clogging states can be classified.