In a significant stride towards enhancing marine propulsion system maintenance, researchers have developed a novel, lightweight prediction model that could revolutionize condition-based monitoring at sea. The study, led by Yongjun Yao from the School of Mechanical Engineering at Hubei University of Technology in Wuhan, China, introduces BSAformer, a Transformer-based architecture designed to predict marine shaft centerline trajectories with remarkable accuracy.
Marine propulsion shafts are subjected to complex, nonlinear forces while at sea, making their trajectories difficult to predict with conventional models. These models often struggle with long-term accuracy and require substantial computational power, limiting their practical use aboard vessels. Yao and his team addressed these challenges by creating a model that integrates bidirectional spectral attention, reconfigured SPSA, and memory-trimming BASS modules. In plain terms, this means the model can look at data from both past and future perspectives, capturing dependencies that other models might miss. It also separates trends related to bearing wear from the regular, periodic movements of the shaft, making it more sensitive to early signs of faults.
The results are impressive. When tested on a public dataset, BSAformer outperformed existing models like Informer and Autoformer, with trajectory errors below 1.5% at lower speeds and below 4% at higher speeds. Even when forecasting up to 720 steps ahead, it reduced dynamic-mode error by 42% compared to the best baseline. As Yao explains, “The BSAA captures both forward and backward frequency-domain dependencies, enhancing the model’s robustness to residual non-stationary dynamics and mitigating long-term error propagation.”
So, what does this mean for the maritime industry? Accurate, real-time prediction of shaft trajectories enables better condition-based maintenance, reducing downtime and repair costs. It also facilitates early fault detection, preventing catastrophic failures and improving safety. The model’s lightweight nature makes it suitable for on-board use, providing a deployable, low-latency tool for health monitoring of ship propulsion systems. As Yao puts it, “The results enable a deployable, low-latency tool for real-time health monitoring and early fault detection of ship propulsion systems.”
The study, published in the journal ‘Results in Engineering’ (translated from the original Chinese title), represents a significant advancement in marine propulsion system maintenance. By providing a more accurate, efficient, and practical tool for predicting shaft trajectories, it opens up new opportunities for the maritime sector to improve operational efficiency, reduce costs, and enhance safety. As the industry continues to embrace digitalization and smart technologies, such innovations will play a crucial role in shaping the future of maritime operations.