Research On Shovel Teeth Of Mining Excavator Fault Detection

Electric shovel is a mining excavator widely used in open-pit mining.During the mining process,the shovel teeth will directly contact with rock,ore and other materials.Due to the large reaction force,the shovel teeth often suffer from wear,fracture,and shedding.The faulted shovel teeth would be mixed with materials and sent into the crusher,causing great damage to the crusher as the result of the hard tooth,which will further lead to the failure of the whole mining-crushing production line.Such kind of failure will cause huge economic losses each year.To solve this problem,a shovel fault detection system was built combining with deep learning and image processing technology to develop related algorithm.Firstly,a dependable camera was selected to build a detection system of electric shovel to accommodate its harsh working conditions.To meet the demand of subsequent detection algorithms,suitable datasets were established and augmented.Secondly,object detection was performed on the bucket image.Preliminary detection of shovel teeth failure was realized by comparing the expected number of shovel teeth with the boundingbox count.In order to meet the requirements on feasibility and accuracy,the Faster R-CNN object detection algorithm was adopted.In this thiese,the anchor initial value of Faster R-CNN was adjusted based on the shovel teeth dataset.From the aspects on accuracy and speed of feature extraction network,VGG16 was more suitable compared with ZFNet and ResNet-50.Also,the model bucket data was used as pre-training parameters to train the real bucket data to improve detection accuracy.Then,a fully convolutional network was used to segment foreground and extract the shovel teeth contours in order to distinguish the end and middle shovel teeth and achieve the order calibration of the shovel teeth after obtaining the position information of the bucket teeth in the image.End and middle bucket teeth distinguishment was conducted by Euclidean distance using standardized Fourier descriptor as contour feature.Position relationship between the shovel teeth was used to predict the position of the fault bucket teeth by least square method.By analyzing the height relationship of the contour points,the normal bucket tooth and the fault length was calculated,and further output the wear degree.Finally,the software of the shovel teeth fault detection system was developed.The user interface of the shovel bucket tooth fault detection system was designed based on the GUI platform of Python's PyQt4.Applications such as image reading,teeth object detection,teeth fault detection,and fault alarm was realized in the proposed system.The algorithm of the whole detection system was verified through the experiments,and the reasons of errors were analyzed.There are 81 figures,14 tables and 91 references in the thesis.