| The submarine pipeline autonomous inspection robot is an unmanned vehicle that can independently detect and follow the submarine pipeline.The goal is to independently detect various pipelines and submarine cables laid in the sea,follow the movement of the pipeline,and detect whether there is a fault in the pipeline.As tens of thousands of kilometers of oil and gas pipelines,submarine optical cables and submarine cables are laid under the sea,it is difficult and dangerous for such a huge system to rely on human for fault detection.Therefore,autonomous inspection robot for submarine pipelines has broad application prospects.In order to improve the robustness and stability of the depth control of the inspection robot,the motion on the vertical plane of the inspection robot is analyzed,and various methods currently applied in autonomous vehicles are studied and compared.First of all,the kinematics and dynamics of the inspection robot were analyzed,and its complete equations of motion in six degrees of freedom were established by Newton’s second law.Combined with the research on depth control problem,the equations of motion in three degrees of freedom,namely axial motion,vertical motion and pitching motion,were selected and decoupled,and the differential equations of motion in three degrees of freedom were obtained.Secondly,the effect of applying the traditional PID controller to AUV depth control is studied and evaluated,and it is concluded that the system is prone to jitter when the control time is long and the system has various interference such as time-varying and unpredictability.In order to improve the stability and robustness of the control,this paper introduces the actor-critic algorithm in reinforcement learning.By combining it with PID control,the actor-critic PID control is established to make up for the immutable parameters of PID control and the uncertainty of the model.The effectiveness of the algorithm is verified by simulation and compared with PID control.It is proved that the algorithm improves the control effect of the system.In order to solve the problem of long training time of reinforcement learning,S-function is introduced as an integral separated PID controller,and the reward function in reinforcement learning is set as a piecewise function to accelerate the learning speed and further improve the control effect.Then,in this paper,two driving modes of vertical depth determination and pitch depth determination of autonomous inspection robot of submarine pipeline are studied.By establishing S-function decision maker,fuzzy decision maker and neural network decision maker,simulation and comparison are made on the driving choices of inspection robot in the taxi process,and the conclusion is that the neural network decision has better effect.Finally,the improved control method and control strategy method are verified by pool experiment.According to the comprehensive analysis of the experimental results,it is proved that the improved actor-critic PID control has higher stability and robustness than PID control,and also verifies the feasibility of the neural network strategy in driver selection. |
