Today, 98% of the world's ships carry out preventative maintenance without connection to the actual status of ship components. Just like the predefined workshop intervals in cars, ship components are regularly maintained based on a predefined schedule. Such an approach leads to unnecessarily high use of spare parts. At the same time, random faults that occur between the intervals are not detected, which can lead to downtime. Overall, this results in unnecessarily high maintenance costs for shipowners.
The main goal of this project is to convince shipowners that a predictive maintenance system based on machine learning can have a direct impact on ships' daily maintenance routines to reduce maintenance costs. A predictive maintenance system means that we monitor and analyze the actual status of the ship components to recommend just-in-time maintenance. This will reduce both the use of spare parts and labor. In addition, we will be able to find random faults that could previously lead to downtime. Overall, this will reduce the shipowners' maintenance costs.
Our self-learning diagnostics and prognostics system is based on comprehensive research results from Ph.D. students at the Department of Ocean Operations and Civil Engineering at NTNU in Ålesund. This project has further developed the system in a commercial direction. The system was installed onboard NTNU's research vessel Gunnerus in the period August to November 2020. During this period, no anomalies were detected even though scheduled maintenance was carried out. These results prove the cost-effectiveness potential and are a clear indication that they are maintaining too much. It is important to emphasize that our results need extended verification over a longer period and on more commercial vessels.
Based on our results from Gunnerus, zero anomalies were detected, even though planned maintenance was performed. Thus, maintenance was performed on healthy engines. This proves the cost-effectiveness potential of reducing the frequency of planned maintenance intervals in the maritime industry. Importantly, our results need verification over a longer time period and on more commercial vessels to prove the concept of predictive maintenance. If successful, we aim to transform planned intervals into dynamic intervals, providing sustainable management control for shipowners.