Development of Hoisting State Estimation Methods of Crane using Machine Learning Based on Slip Value of Induction Motor

Abstract

In order to develop a overhead crane that can provide high safety and productivity, the authors developed a technology for rope tension detection and load estimation of payload. The technology can be applied to detection of overload and lifting off from ground. Multi-task learning using neural network was adopted to estimate the output of induction motor with nonlinear properties. The driving signals were input to the neural network, and the rope tension state and the load were estimated at the same time. The authors proposed a method to create teacher data by using slip value of induction motor to determine the rope tension point based on actual load. As a result, it was clarified the proposed method is effective for evaluating the rope tension point based on the increase of actual load. In addition, the accuracy of the learning result that was learned by the proposed method was improved from the previous method. As for the accuracy of the proposed method, the deviation of the rope tension detection error was less than ±0.5 second, and the load estimation error was 2% of the maximum load.