Edge AI-based early anomaly detection of LNG Carrier Main Engine systems

This paper proposes a novel deep learning model and Edge-AI technology for early detection of anomalies in the main engine system of LNG carriers. The main engine system is a critical component of a ship, and any abnormalities can lead to serious accidents. Conventional anomaly detection methods do not consider the residuals of time-series forecasting and the transferability of the model to multiple ships. The proposed deep learning model consists of two LSTM-based Revin-AutoEncoder models, which utilize the Revin technique to remove non-stationary information and compensate for the residual generated by unstable time-series forecasting. Furthermore, Edge-AI technology is employed to perform model inference without communicating with a central server, enabling fast detection and response to abnormalities, preventing network congestion, and reducing costs. The effectiveness of the proposed method in detecting anomalies in a new LNG carrier with various equipment and systems is experimentally demonstrated, overcoming the challenge of anomaly detection caused by the diverse equipment and systems of a ship. The experimental results showed a maximum recall performance of 0.78. The proposed system show the possibility of the learned model performing early anomaly detection on another ship and is expected to contribute to the development of anomaly detection technology in various industries.