Complex equipment plays a critical role in the defense and industrial sectors. Determining the optimal maintenance timing for such equipment can effectively prevent financial losses caused by unplanned downtime. To address problems such as insufficient monitoring data affecting model accuracy and model optimization results that often deviate from engineering practice in current methods for determining the optimal maintenance timing, this paper proposes a model for determining the optimal maintenance timing based on monitoring image-assisted enhancement. First, a “deep learning-based object recognition combined with motion analysis” approach is employed to extract health information from monitoring images. Furthermore, a 3 dimensional attention mechanism is utilized to fuse the extracted image data with sensor data, thereby achieving data augmentation. Second, an optimal maintenance timing determination model is established based on evidential reasoning rules to perform an initial determination of the optimal maintenance timing. Third, optimization constraints are defined based on engineering realities, and model parameters are optimized to enhance model accuracy while ensuring the interpretability of the results. Finally, the validity of the proposed model was verified through comparative experiments using a liquid rocket structural health simulation system.