Deep learning enables safety concerns of fuel ignition recognition in thermal-hydrodynamic process with ship motion-sloshing interaction

Identifying initial accidental indicators is crucial for finding the hidden patterns underlying thermal-hydrodynamic data, thus analyzing combustion performance of marine fuel in engine room. Recent advancements in deep learning (DL) have emerged as data-driven techniques for handling complex parameters and predicting potential ignition phenomena. Current study investigates the effectiveness of DL algorithms, particularly heated surface ignition of leaking marine fuel under different ship motion conditions, in finding new approach for recognition of initial ignition. Experimental and FLUENT-based data from marine fuel ignition process are collected and analyzed, focusing on multiple dominant motions, including ship roll, pitch and heave. The potential applications of DL models (single and hybrid) in ignition indicator are addressed, especially in the presence of limited actual data and ship motion-sloshing interactions. Saliency analysis reveals that the long short-term memory (LSTM) and bidirectional long short-term memory (BiLSTM) models for ignition characteristics in inaccurate capture without intense flames, which suggests that it generates conspicuous discrepancies to predictions. The results indicate that the convolutional neural network (CNN) model achieved the lowest deviation rate at 3.22 %, yielding markedly reduce deviation rate between predicted and actual heated surface temperature fluctuations by CNN-LSTM (1.65 %) and CNN-BiLSTM (0.28 %). The proposed CNN-BiLSTM model demonstrates high accuracy in predicting ignition temperature under elevated ship motions, with mean absolute percentage error (MAPE) and root mean square error (RMSE) of 0.56 and 0.70, respectively. These findings highlight the capability of DL models particularly CNN-BiLSTM, in effectively capturing thermal behavior interaction with dominant ship motions. This study emphasizes the importance of leveraging DL techniques for marine fuel ignition prevention and suggests future work should focus on enhancing model accuracy through diverse scenarios in engine room and comprehensive datasets.