Abstract
In order to develop safe and efficient overhead cranes, we developed an estimation technology for rope tension and load that can be applied to the initial sway suppression and overload detection during lifting from ground. To estimate the state of hoisting motor with nonlinear properties, we adopted multi-task learning using deep learning. The driving frequency, the rotation speed, and the current of the motor were input to the neural network, and the rope tension and the load were estimated simultaneously. The training data used for learning was created by image analysis of the movement of markers attached to load block and suspended load. The training and the validation data were prepared separately. To verify the effective of the learning results, we added data with different horizontal position and lifting length for the validation data. As a result, it was clarified that the tension condition of the rope and the load can be estimated simultaneously. The average delay of rope tension detection was 0.02 seconds, and the load estimation error was less than 7% of the maximum load.
