Authors:
Emerson Klippel
1
;
Ricardo Augusto Rabelo Oliveira
2
;
Dmitry Maslov
3
;
Andrea Gomes Campos Bianchi
2
;
Saul Emanuel Delabrida Silva
2
and
Charles Tim Batista Garrocho
2
Affiliations:
1
Vale Company, Parauapebas, Para, Brazil
;
2
Computing Department, Federal University of Ouro Preto, Ouro Preto, Minas Gerais, Brazil
;
3
TinkerGen, Seeed Studio, Shenzhen, China
Keyword(s):
Deep Neural Network, Device Edge, Rip Detection, Conveyor Belt.
Abstract:
Failures in the detection of longitudinal rips on conveyor belts are events considered catastrophic in mining environments due to the financial losses caused and the exposure to safety risks of the maintenance teams. The longitudinal rip detection technologies used today have limitations, being the most reliable systems expensive and complex and the simplest and cheapest systems unreliable. In view of this scenario, we studied the implementation of a longitudinal rip detection solution based on images of the conveyor belt. The images will be collected in real time and inference, rip detection, will be carried out locally using a deep neural network model executed on device edge. The results obtained with the prototype, in controlled field tests, were satisfactory and showed the feasibility of using deep neural network algorithms executed on device edge. These results encourage the development of a complete solution for the detection of defects in conveyor belts considering all the op
erational conditions found in the mining environment.
(More)