Authors :
Philip O. Adejumobi; Oluwadare A. Adebisi; Iyabo I. Adeaga; Abiodun A. Baruwa; Kolawole M. Ajala
Volume/Issue :
Volume 8 - 2023, Issue 4 - April
Google Scholar :
https://bit.ly/3TmGbDi
Scribd :
https://bit.ly/42M1hi3
DOI :
https://doi.org/10.5281/zenodo.8265646
Abstract :
Mechanical parts of engines help to reduce
friction and carry weight for linear or rotating motion.
Modern engines are complex systems with structural
elements, mechanisms, and mechanical parts. The
building blocks of the engines are joined together using
several mechanical components that are similar in shape
and size. During the assembly and disassembly of these
complex engines, the mechanical components get mixed
up. The traditional classification techniques for
components are laborious with high costs. Existing
research for classifying mechanical components uses
algorithms that work based on shape descriptors and
geometric similarity thereby resulting in low accuracies.
Hence, there is a need to develop an automatic
classification technique with high accuracy. This study
classified four mechanical components (bearing, nut,
gear, and bolt) using four deep learning models
(AlexNet, DenseNet-121, ResNet-50, and SqueezeNet). In
the result, Densenet-121 achieved the highest
performance at an accuracy of 98.3%, sensitivity of
95.8%, specificity of 98.5%, and area under curve of
98.5%.
Keywords :
Mechanical Components, Deep Learning, Convolutional Neural Network, Engine, Transfer Learning.
Mechanical parts of engines help to reduce
friction and carry weight for linear or rotating motion.
Modern engines are complex systems with structural
elements, mechanisms, and mechanical parts. The
building blocks of the engines are joined together using
several mechanical components that are similar in shape
and size. During the assembly and disassembly of these
complex engines, the mechanical components get mixed
up. The traditional classification techniques for
components are laborious with high costs. Existing
research for classifying mechanical components uses
algorithms that work based on shape descriptors and
geometric similarity thereby resulting in low accuracies.
Hence, there is a need to develop an automatic
classification technique with high accuracy. This study
classified four mechanical components (bearing, nut,
gear, and bolt) using four deep learning models
(AlexNet, DenseNet-121, ResNet-50, and SqueezeNet). In
the result, Densenet-121 achieved the highest
performance at an accuracy of 98.3%, sensitivity of
95.8%, specificity of 98.5%, and area under curve of
98.5%.
Keywords :
Mechanical Components, Deep Learning, Convolutional Neural Network, Engine, Transfer Learning.
