The bucket of mining excavator is the main working part of excavator equipment. Its bucket teeth consume a lot in bad working environment. If the bucket teeth fail, it will not only accelerate the consumption of bucket, but also affect the subsequent mining crushing operation and cause serious economic losses. Therefore, aiming at this problem, based on the investigation of bucket tooth fault detection at home and abroad, combined with the existing machine vision technology, this paper studies the bucket tooth fault detection, carries out the relevant work, designs the bucket tooth fault detection system, develops the corresponding algorithm, and realizes the functions of bucket tooth fault detection and fault alarm. The research work is as follows:
(1) The overall design of bucket tooth fault detection system. According to the actual working conditions and working environment of the shovel, the overall scheme of the bucket tooth fault detection system is designed, and the installation mode of the detection system is analyzed in detail according to the actual working characteristics of the shovel. Finally, the data set used in this paper is shown and enhanced to prepare for the training of deep neural network.
(2) The target detection algorithm of bucket tooth fault detection system is realized. This paper takes the bucket tooth target detection as the first link of bucket tooth fault detection, which is the initial detection information of the detection system. In this paper, fast r-cnn is used as the target detection framework, vgg-16 is used as the feature extraction network, and the anchor initial value of the target detection framework is updated according to the characteristics of the bucket tooth target detection data set. Finally, the bucket tooth target detection function is realized and has a good detection effect. In the target detection experiment, this paper compares resnet-50 and zfnet, and finds that the target detection accuracy based on vgg-16 is higher than zfnet, equivalent to resnet-50, and the detection speed is better than resnet-50. In the experiment, it is also found that using the pre training parameter model of the experimental model bucket data set can help the network train the actual data faster and better.
(3) Implementation of bucket tooth fault detection algorithm. In this paper, a full convolution neural network is trained to segment the bucket tooth image. Through simple preprocessing and threshold segmentation, the contour image of bucket tooth is further obtained. In this paper, the standardized Fourier descriptor is used as the feature of bucket tooth profile, and the Euclidean distance is calculated to distinguish the bucket tooth at the end and the bucket tooth at the middle. Then, combined with the spacing relationship between bucket teeth, the position of fault bucket teeth is located. Finally, the image of fault bucket teeth is segmented, and its pixel length is calculated according to the characteristics of bucket tooth contour. Combined with the pixel length of normal bucket teeth, its wear degree is estimated.
(4) User interface design of bucket tooth fault detection system. This paper plans the algorithm flow of the fault detection system, and designs the functions of each part of the detection system. Finally, relying on Python and PyQt4, the user interface of the bucket tooth fault detection system is developed. Through the simple button function, the bucket tooth fault detection, fault alarm and visualization of the detection results are realized.