The Conveyor Belt (CB) is the most versatile and widespread means of transporting some materialsin industrial facilities. In this apparatus, the rollers are equipment that provide support and guidethe belt. When the roll which rests on the belt support fails, the operation of the conveyor belt isimpaired. Fault detection on this equipment is performed by maintenance personnel who rely onnoise produced by defective rollers, among other factors, to make a diagnosis. In addition, since theCB can be hundreds of meters long and extend over hard to reach places, determining which roll isdefective becomes a difficult, expensive and even dangerous task. In this work, a proposal for thedetection of defective rollers through sound samples and computational intelligence is presented.The study was conducted at the port of Tubarao, ES, Brazil, where the iron ore produced by thecompany Vale SA is carried to its ships through various CBs. A directional microphone was used tocapture sound samples from dozens of rolls in good and bad condition, forming our database.Having the data, they were initially converted to the frequency domain via Fast Fourier Transform(FFT) and segmented. Part of the pre-processed dataset was used to develop a random-forest-baseddetector and the rest of the data was kept aside for blind testing purposes. The method of extractingsound features proved to be a very efficient and promising tool in classification problems. Onaverage, the Random Forest algorithm had accuracy and recall over 90 %, which shows itseffectiveness. Last but not the least, it is noteworthy that Vale SA is developing a robot for inspectionservices in CBs and the failure detection system presented here is one of the modules for thatinspection robot.
展开▼
