Do-Hyun Chun, Myung-Il Roh, Hye-Won Lee, Seung-Ho Ham, "A Block Lifting Method with Wire Ropes Based on Deep Reinforcement Learning", Proceedings of ICDM(International Conference on Decarbonization and Digitalization in Marine Engineering) 2022, Si-Heung, Korea, 2022.04.28-29
International Conference
2022.04.29 15:53
Do-Hyun Chun, Myung-Il Roh, Hye-Won Lee, Seung-Ho Ham, "A Block Lifting Method with Wire Ropes Based on Deep Reinforcement Learning", Proceedings of ICDM 2022, Si-Heung, Korea, 2022.04.28-29
조회 수 1159
Abstract | The block erection using a gantry crane is an important process for the production of the ships in a shipyard. The motion of the block should be controlled accurately under the external forces to prevent collision with the structures and the excessive loads on wire ropes. However, it is difficult to control the block during the lifting because the movement of the block is indirectly controlled with various objects such as trolleys, hooks, equalizers, and wire ropes. Therefore, we proposed the Deep Reinforcement Learning (DRL)-based block lifting method in this study. The DRL-based block lifting method can control the block under the change of the center of gravity and modelling uncertainty. Furthermore, the DRL-based block lifting method can provide robust control with an unexpected motion of the block due to the unexpected external disturbance. The position, orientation and angular velocity of the block and hoisting speed of wire ropes were set as the input state of the neural network of DRL. The hosting speed of wire ropes was controlled as the output action of DRL. The functions to minimize the change of orientation and to stabilize the speed of the block were set as the reward of DRL. In this study, the deep deterministic policy gradient (DDPG) method of DRL, which is a kind of off-policy actor-critic method, was applied to solve the problem with continuous state space and continuous multi-action space. To verify the DRL-based block lifting method proposed in this study, it was compared with traditional control algorithms for various simulation examples. As a result, the proposed method could effectively control the block with the modelling uncertainty. Also, the proposed method could respond to the unexpected motion of the block effectively due to the unexpected external disturbance. |
---|---|
Publication Date | 2022-04-28 |