Research On Online Fault Detection Technology Of Automatic Catway Wire Rope For Drilling Rig

Wire rope is an irreplaceable component of large equipment,and its safe operation is related to the overall production of the project;if it fails,serious consequences will befall the staff’s life and safety.In order to timely find the failure of the wire rope and avoid irreparable consequences,troubleshooting in advance is very necessary.This paper takes the automatic catwalk wire rope of a drilling rig as the research object,aims at achieving intelligent fault detection,replace the human eye with a camera,selects a suitable industrial camera model,builds a wire rope image online acquisition device,and constructs a wire rope fault data set so as to lay the foundation for subsequent fault detection of wire rope.For the problem that the traditional image processing process is cumbersome,a convolutional neural network-based fault detection method is proposed,and the network model structure is determined as an 8-layer convolutional neural network.The wire rope fault dataset is obtained as the input of a convolutional neural network model to achieve intelligent detection of wire rope faults with a 79.18% correct fault detection rate.Compared with traditional image processing methods,the proposed method has a simple process and better generality.Aiming at the problem that the structure optimization of convolutional neural network model is difficult,this paper proposes to optimize the structure of convolutional neural network by using attention module and self-distillation algorithm.Firstly,the network structure is optimized by the attention algorithm module,and the fault detection accuracy rate after optimization is increased from 79.18% to 94.53%.Then,the self-distillation algorithm was combined to further optimize the model structure,and the fault detection accuracy rate was improved to 98.11% after optimization,and compared with the optimization results of the attention module algorithm,the accuracy rate was improved by 3.58%.Experimental results show that the optimized network model significantly improves the accuracy of model detection.Moreover,through comparative experiments,it can be proved that the self-distillation algorithm can further improve the detection accuracy of convolutional neural networks without increasing the model size and calculation.