Research On Working State Detection System Of Shovel Teeth

Excavator in open-pit mine production for a long time in the complicated work under the harsh environment of relieving prone to wear and loss failure,failure relieving seriously affect the efficiency and economic benefit of opencast wear if not timely find that will cause the bucket wear,and loss of relieving mixed in the coal into the broken chance to generate serious security hidden danger,to solve this problem,In this paper,the image processing technology is applied to mine production,and the work status detection system of shovel teeth is studied.The aim is to realize the all-weather status monitoring of shovel teeth throughout the year,and to alarm and remind shovel drivers in the first time when there is abnormality.Firstly,combined with the working characteristics of the electric bucket teeth,the abnormal state of the shovel teeth such as fracture and wear was analyzed.Combined with the actual requirements of the project site,the detection scheme of the shovel teeth falling off and wear was designed by using the deep learning-based target detection and instance segmentation technology,and the hardware platform was built to collect the key frame images of the shovel teeth.In order to solve the problems of poor night imaging quality,fog and noise of the original data,the mixed light filling,dark channel fog removal and filtering pretreatment were carried out.In order to solve the problems of high repetition rate of the original data samples,the image enhancement technology was used to enrich the samples,and the shovel tooth target detection and instance segmentation data sets were established respectively.Based on YOLO-V4 algorithm,the detection model of shovel teeth was built.According to the characteristics of relatively fixed aspect ratio of shovel teeth samples,K-means algorithm was used to re-cluster the prior box in the model.The mean mean accuracy(MAP)of the optimized shovel teeth detection model reached 0.982,and the average crosscut-merge ratio(AVG?IOU)was 0.877.The detection time of a single frame is 19 ms,and both accuracy and real-time performance can meet the requirements of tooth shedding detection.A tooth shedding judgment algorithm is designed to sort the position of the prediction box output by the model and judge whether the tooth shedding occurs.The image mask of shovel teeth was extracted by example segmentation,and the model of shovel teeth segmentation was built based on MS RCNN segmentation network.Res Net-101 was selected as the backbone network to improve the FPN feature pyramid network,and a feature fusion network from bottom to top was added to make each layer make full use of the detail features of bottom shovel teeth.On the test set,the average cross and merge ratio(MIOU)of shovel teeth was 0.8362,and the average pixel accuracy(MPa)was 0.9076.A binary mask based wear algorithm of shovel teeth was designed to calculate the pixel area of the segmentation mask of each shovel teeth,and the current wear degree of shovel teeth was accurately judged by comparing it with the unworn initial value.Finally,the function of the system is tested by field test.Install hardware equipment,structures,software environment,the debugging on the open pit mine excavator,the relieving off alarm and wear testing experiments,the experimental results show that the relieving state detection system in relieving loss under different wear conditions and can effectively identify and relieving loss accurate alarm rate can reach more than 90%,the wear degree of detecting accuracy can reach more than 95%,It has strong robustness to environment and high engineering application value.