With the increasing demand for shipbuilding, automating painting processes has become vital for safety and efficiency. This study introduces a fault prediction and alarm system for a multi-joint painting robot capable of coating both the interior and exterior of ship blocks. Conventional safety monitoring methods depend on strain gauges that measure stress relative to allowable limits. Although effective, they require additional sensors, which increases cost and operational complexity. To address these limitations, a deep-learning-based surrogate model was developed to predict structural safety using only the reaction forces and moments generated during robot operation. Over 4,000 finite element simulations incorporating dynamic loads were performed using Abaqus and Isight software to generate stress data. The results were processed in Python to build a dataset, which was then split at a ratio of 9:1 for training and testing to train a deep neural network. The hyperparameters were optimized using Bayesian optimization. The proposed model achieved a coefficient of determination (R²) of 0.9985 and error levels below 1%, enabling the real-time prediction of maximum stress and sending an immediate alarm when limits were exceeded. This system enhances safety while lowering costs by reducing the reliance on external sensors. Extending the simulation range beyond the expected loads further improves robustness, supporting practical deployment in shipbuilding automation.