Design of a decision support system to achieve condition-based maintenance in ship machinery systems

A decision support system is proposed for the condition-based maintenance of ship machinery systems based on the adaptive neuro-fuzzy inference system (ANFIS) approach. A case study is conducted for a container ship's main diesel engine. Within the scope of the methodology, the main engine power is predicted based on exhaust gas outlet temperatures of cylinders and the main engine shaft RPM. In this regard, firstly, two different strategies such as the creation of the cylinder-basis model and the development of the overall system model are developed to determine the ANFIS model structure for the analysis. Then, comparative analyses are carried out to select a suitable ANFIS structure and its specific membership functions. In addition, the estimation process is also performed by the artificial neural network (ANN) model, and its results are compared with the findings of the best ANFIS structure. The success of the constructed models is evaluated by some error metrics. The overall ANFIS model with 5 membership functions is determined as the best approach by scores of 0.9806 for R2, 1.6588 MW for RMSE, and 3.2703 for MAPE. As a result of the estimation procedure, a decision support system to assist marine operators in maintenance operations is developed. The proposed strategy can be applied to different types of systems in the ship engine room such as the fuel oil system and can be improved by including more parameters obtained from the system's significant points in the analysis.

Automated summary

A decision support system based on the adaptive neuro-fuzzy inference system (ANFIS) approach is proposed for condition-based maintenance of ship machinery systems. A case study is conducted for a container ship's main diesel engine to predict the main engine power based on exhaust gas outlet temperatures of cylinders and the main engine shaft RPM. Comparative analyses are carried out to select a suitable ANFIS structure and its specific membership functions. The success of the constructed models is evaluated using error metrics, and the overall ANFIS model with 5 membership functions is determined as the best approach.