Predictive maintenance for ballast pumps on ship repair yards via machine learning

Abstract

Ballast pumps play a pivotal role in the operations of floating docks during docking and undocking of a marine vehicle that is due for inspection, repairs, or maintenance. Any failure of the ballast pumps during the docking/undocking operations may pose a threat to the safety of the dock crew, loss of revenue to the dock owners/operators, and disorganizes the schedules. Failure of the ballast pumps has been attributed to maintenance approaches used to predict such occurrences. This paper aims to provide a predictive maintenance approach towards an early warning maintenance/failure warning system using machine learning as opposed to sensor technology. A machine learning methodology was used to process and analyze the dock pump operating parameter with the aim of drawing inferences from data via MATLAB. Principal component analysis for the operating parameters was carried out to define which parameters will accurately predict the pump failure. Back pressure, flow rate, amperage, RPM and suction pressure, operating parameters, were monitored for 40 weeks. The predictive maintenance tool predicted that the dock pump may fail or requires maintenance between the 7th and 8th weeks. This prediction deviated from the actual 9 weeks and 2 days, that it took the dock pump to fail. A 13.85% deviation from the actual failure time could be attributed to the quality and low volume of the operating data recorded. Nevertheless, with less ambiguity of the data, the maintenance prediction tool can be used as a basis before sensor technology on the dock pumps is implemented.