Development Of A Fault Diagnosis Model For Multi And Single Rotating Machinery Fault Data Using Xgboost And RFECV On Laboratorium Experiment Data

Aryasatya, Daniswara Arkananta (2025) Development Of A Fault Diagnosis Model For Multi And Single Rotating Machinery Fault Data Using Xgboost And RFECV On Laboratorium Experiment Data. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5007211193-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Diagnosis kerusakan pada mesin berotasi melalui analisis sinyal getaran telah menjadi komponen krusial dalam pemeliharaan prediktif dan manufaktur. Studi ini mengembangkan dan memvalidasi sebuah model diagnosis kerusakan berkinerja tinggi yang mampu mengklasifikasikan kondisi kerusakan tunggal maupun multi-kerusakan yang kompleks. Kerangka kerja yang diusulkan mengintegrasikan Recursive Feature Elimination with Cross-Validation (RFECV) untuk pemilihan fitur yang optimal dengan algoritma Extreme Gradient Boosting (XGBoost) untuk klasifikasi. Data getaran dikumpulkan secara eksperimental dari test rig laboratorium di bawah sepuluh kondisi yang berbeda, termasuk kondisi sehat, unbalance, misalignment, kerusakan bantalan cincin dalam dan luar, serta berbagai kombinasinya. Dari set awal fitur frekuensi, RFECV berhasil mengidentifikasi sebuah subset optimal yang terdiri dari delapan fitur yang dapat memaksimalkan kinerja model. Model XGBoost akhir, yang dioptimalkan menggunakan Bayesian Optimization bisa mencapai akurasi, presisi, recall, dan F1-score sempurna sebesar 100% di semua sepuluh kelas. Ketahanan dan kemampuan generalisasi model ini divalidasi lebih lanjut melalui skor rata-rata validasi silang 10-lipatan (10-fold cross-validation) sebesar 0.997. Dalam analisis perbandingan, XGBoost juga terbukti menjadi model yang paling efisien secara komputasi, mengungguli RandomForest dan Gradient Boosting dalam hal kecepatan pelatihan sambil tetap mempertahankan akurasi tingkat atas. Hasil penelitian ini secara meyakinkan menunjukkan bahwa pendekatan terintegrasi RFECV-XGBoost menyediakan solusi yang akurat, tangguh, dan efisien untuk diagnosis kerusakan yang kompleks, serta menawarkan potensi untuk meningkatkan keandalan sistem pemantauan kondisi di industri
==========================================================================================================================================
The diagnosis of faults in rotating machinery through vibration signal analysis has become a crucial component in predictive maintenance and smart manufacturing. This study develops and validates a high-performance fault diagnosis model capable of classifying both single and complex multi-fault conditions. The proposed framework integrates Recursive Feature Elimination with Cross-Validation (RFECV) for optimal feature selection with the Extreme Gradient Boosting (XGBoost) algorithm for classification. Vibration data was experimentally collected from a laboratory test rig under ten distinct conditions, including healthy, unbalance, misalignment, inner and outer ring bearing faults, and their various combinations. From an initial set frequency domain features, RFECV identified an optimal subset of eight features that maximized model performance. The final XGBoost model, tuned with Bayesian Optimization, demonstrated exceptional performance on the independent test dataset, achieving a flawless 100% accuracy, precision, recall, and F1-score across all ten classes. The model's robustness and ability to generalize were further validated by a mean 10-fold cross-validation score of 0.997. In a comparative analysis, XGBoost also proved to be the most computationally efficient model, outperforming both RandomForest and Gradient Boosting in training speed while maintaining top-tier accuracy. The results conclusively demonstrate that the integrated RFECV-XGBoost approach provides a highly accurate, robust, and efficient solution for complex fault diagnosis, offering significant potential for enhancing the reliability of industrial condition monitoring systems.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Diagnosa Kegagalan, Mesin Berputar, Analisa Vibrasi, XGBoost, RFECV, Fault diagnosis, Rotating Machinery, Vibration Analysis, XGBoost, RFECV
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA935 Vibration T Technology > TJ Mechanical engineering and machinery > TJ174 Maintenance and repair of machinery
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User:	Daniswara Arkananta Aryasatya
Date Deposited:	04 Aug 2025 06:23
Last Modified:	04 Aug 2025 06:23
URI:	http://repository.its.ac.id/id/eprint/125872

Actions (login required)

View Item