Banca de DEFESA: WALLISSON FERNANDES MARTINS DOS SANTOS

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : WALLISSON FERNANDES MARTINS DOS SANTOS
DATE: 26/07/2023
TIME: 10:30
LOCAL: Sala virtual Google Meet
TITLE:

Bearings failures detection and diagnosis, under different loads and speeds, by using Convolutional Neural Networks

KEY WORDS:

failure detection and diagnosis, convolutional neural network, artificial intelligence techniques, bearing failures, machine learning.

PAGES: 83
BIG AREA: Engenharias
AREA: Engenharia Mecânica
SUBÁREA: Projetos de Máquinas
SPECIALTY: Máquinas, Motores e Equipamentos
SUMMARY:

With the increasing complexity and costs of industrial systems, management measures aimed at preventing or mitigating the loss of reliability, decreased productivity and safety risks, caused by process abnormalities and component failures, become increasingly important. . In this context, Artificial Intelligence (AI) has been consolidating itself as an effective and challenging means in the process of monitoring, detecting and diagnosing failures in equipment and industrial systems. Among the equipment, which are frequently the object of studies, bearings stand out, which are critical mechanical components of rotating machines. Vibration monitoring is the most widely used technique for detecting, locating and distinguishing bearing faults. Faced with the efficient and increasing performance of AI techniques and the importance of bearings in industrial processes, this work implements a Convolutional Neural Network (CNN) for detection and diagnosis of faults in bearings, under different loads and speeds in the motor and different types and depth of bearing failures. For the development of the proposed approach, the Case Western Reserve University (CWRU) bearing test database was used. The raw vibration signals were pre-processed through the Continuous Wavelet Transform (TWC) and converted into images, which were fed directly into the developed CNN structure. When compared to other CNN-based methods that used the same database, the proposed approach demonstrated superiority or was at least as successful, achieving an accuracy of 97.7% when tested with files under operating conditions other than operating conditions. training.

COMMITTEE MEMBERS:
Presidente - 1451883 - FABIO MENEGHETTI UGULINO DE ARAUJO
Interno - 1328152 - CARLOS EDUARDO TRABUCO DOREA
Interno - 1242315 - PABLO JAVIER ALSINA
Externo à Instituição - MARCELO ROBERTO BASTOS GUERRA VALE - UFERSA