The increasing complexity of vessels’ facilities and systems, the high economic impact of maintenance costs, and the need for high reliability and efficiency have led to a growing interest in condition-based maintenance (CBM). In this contest, machine learning (ML) has proven to be a powerful tool in approaching CBM tasks, as it can handle high-dimensional and multivariate data and extract hidden relationships within them in complex and dynamic environments. Here we propose a diagnostic model for monitoring the state of marine diesel engines and their components through the analysis of process parameters and ML tools. The model employs an artificial neural network (ANN) to estimate the optimal engine parameter value for normal condition operation. Deviations between measured real-time signals and optimal estimated values are used as indicators of potential faults, facilitating early detection and prevention. The calibration and performance evaluation of the diagnostic model are conducted on simulated fault data generated through GT-Power simulation tool. This approach to engine monitoring may overcome some limitations of ML algorithms based on supervised models, which rely on historical data containing information about anomalous activities or faults that are often not available.