Fault-tolerant control of bionic robotic fish based on a hybrid Bayesian-Northern Goshawk Optimization framework

When bionic robotic fish operate in complex environments, central pattern generator units are prone to lock-up faults, which can severely degrade locomotion controllability. To address these issues, this study proposes a bio-inspired fault-tolerant proportional–integral–derivative control framework for bionic robotic fish by integrating the Northern Goshawk Optimization algorithm with Bayesian Optimization. First, a Luenberger observer is constructed to enable online state estimation and fault inference. By augmenting the yaw dynamics with an equivalent disturbance term and monitoring its estimate against a decision threshold, it provides a real-time basis for fault-tolerant control decisions. Second, the Northern Goshawk Optimization algorithm is adopted for preliminary adaptive tuning. This algorithm can rapidly explore proportional–integral–derivative parameters over a wide range, enabling preliminary adaptation to model changes caused by faults. Third, an online optimization strategy is designed based on the Northern Goshawk Optimization-Bayesian Optimization Framework. The Bayesian Optimization algorithm is employed for sample-efficient local refinement, thereby achieving accurate compensation and robust tuning of proportional–integral–derivative parameters under fault awareness and effectively regulating the central pattern generator output. Finally, through simulation and experimental analysis, the proposed fault-tolerant control strategy demonstrates superior performance in both posture control of the bionic fish and handling central pattern generator lock-up faults, highlighting its effectiveness in complex and fault-prone underwater environments.