Sistem Identifikasi Kerusakan Bearing Motor Induksi Tiga Fasa Menggunakan Backpropagation Neural Network

Muhammad Shandar Fadillah Faseh, Shandar (2025) Sistem Identifikasi Kerusakan Bearing Motor Induksi Tiga Fasa Menggunakan Backpropagation Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040211076 - Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Motor induksi tiga fasa merupakan komponen vital dalam sistem industri yang keandalannya sangat dipengaruhi oleh kondisi bearing. Kerusakan bearing menjadi penyebab utama kegagalan mesin. Penelitian ini mengusulkan pendekatan berbasis Backpropagation Neural Network (BPNN) untuk klasifikasi kondisi bearing menggunakan data getaran. Penelitian ini bertujuan untuk mengevaluasi akurasi model BPNN terhadap dataset standar SUBF v1.0 dan mengujinya pada data nyata yang diperoleh dari prototipe konveyor. Hasil pengujian menunjukkan bahwa BPNN mampu mencapai akurasi keseluruhan sebesar 99,92% pada dataset standar, mengungguli metode Weighted K-NN berbasis Cepstral Autoregressive dengan akurasi 99,31%. Pada implementasi menggunakan prototipe konveyor, performa klasifikasi per kelas menunjukkan variasi signifikan. Untuk kelas normal, diperoleh precision 98,5%, recall 93,6%, dan akurasi 97,4%, mencerminkan deteksi yang sangat akurat dan stabil. Pada kelas outer fault, recall mencapai 99,0%, namun precision menurun menjadi 65,9%, menunjukkan kecenderungan false positive yang tinggi. Sementara itu pada kelas inner fault, meskipun precision tinggi sebesar 93,7%, namun recall yang rendah (51,3%) yang menandakan model sering gagal mendeteksi kerusakan aktual. Temuan ini mengindikasikan bahwa meskipun model BPNN sangat efektif secara teoretis, terdapat tantangan dalam penerapan praktis terutama dalam mengenali kerusakan inner race secara konsisten. Oleh karena itu, perlu dilakukan optimasi lanjutan pada aspek arsitektur model dan strategi pemrosesan fitur. Secara keseluruhan, penelitian ini menunjukkan bahwa BPNN menjanjikan untuk sistem diagnosis kerusakan bearing berbasis getaran, baik dalam lingkungan terkendali maupun aplikasi nyata pada sistem prototipe.
====================================================================================================================================================================================
Three-phase induction motors are vital components in industrial systems, with their reliability strongly influenced by bearing condition. Bearing faults are a leading cause of machine failures. This study proposes a Backpropagation Neural Network (BPNN)-based approach for bearing condition classification using vibration data. The objective is to evaluate the accuracy of the BPNN model on the standard SUBF v1.0 dataset and to test its performance on real-world data acquired from a conveyor prototype. Experimental results show that the BPNN achieved an overall accuracy of 99.92% on the standard dataset, outperforming the Cepstral Autoregressive-based Weighted K-NN method, which reached 99.31%. In the prototype implementation, per-class classification performance varied significantly. For the normal class, the model achieved a precision of 98.5%, recall of 93.6%, and accuracy of 97.4%, indicating highly accurate and stable detection. For the outer fault class, the recall was 99.0%, but precision dropped to 65.9%, reflecting a high tendency for false positives. Meanwhile, for the inner fault class, the model achieved high precision (93.7%) but low recall (51.3%), indicating frequent failures to detect actual faults. These findings suggest that although the BPNN model performs well theoretically, there are practical challenges, particularly in reliably detecting inner race faults. Further optimization of model architecture and feature processing strategies is necessary. Overall, this study demonstrates that BPNN offers a promising solution for vibration-based bearing fault diagnosis, both in controlled environments and in real-world prototype applications.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bearing, BPNN, Klasifikasi, Getaran
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Vocational
Depositing User:	Muhammad Shandar Fadillah Faseh
Date Deposited:	05 Aug 2025 06:32
Last Modified:	05 Aug 2025 06:32
URI:	http://repository.its.ac.id/id/eprint/125859

Actions (login required)

View Item