Strategi Pemeliharaan Preskriptif: Optimalisasi Keandalan Mesin Berbasis Machine Learning Guna Mencegah Terjadinya Downtime Pada Mesin Industri
DOI:
https://doi.org/10.24002/prosidingkonstelasi.v2i1.11199Keywords:
Prescriptive Maintenance, Machine Learning, XGBoost, Zero DowntimeAbstract
Abstrak. Downtime pada mesin industri dapat menyebabkan kerugian yang signifikan dalam produktivitas dan efisiensi operasional. Berbagai metode pemeliharaan seperti Corrective, Descriptive, Diagnostic, dan Predictive Maintenance memiliki keterbatasan dalam mengoptimalkan strategi mitigasi downtime. Oleh karena itu, penelitian ini mengimplementasikan Prescriptive Maintenance berbasis Machine Learning (XGBoost) untuk tidak hanya memprediksi kegagalan mesin tetapi juga memberikan rekomendasi langkah korektif guna mencapai zero downtime. Dataset AI4I 2020 Predictive Maintenance digunakan sebagai sumber data, dengan menerapkan berbagai teknik preprocessing, seperti Min-Max Scaling, SMOTE, Heatmap Korelasi, VIF, serta deteksi outlier menggunakan Z-Score dan IQR. Model XGBoost dilatih untuk memprediksi probabilitas kegagalan mesin, yang kemudian dianalisis menggunakan Feature Importance untuk mengidentifikasi penyebab utama kegagalan. Evaluasi model menunjukkan akurasi 98.12%, precision 97.95%, recall 98.23%, dan AUC-Score 99.59%, membuktikan keandalan sistem dalam mendeteksi dan mengklasifikasikan kegagalan mesin. Dengan implementasi strategi ini yang didukung oleh IoT dan database real-time, industri dapat mengoptimalkan efisiensi pemeliharaan, mengurangi downtime tak terduga, serta meningkatkan keandalan operasional.
Kata Kunci: Prescriptive Maintenance; Machine Learning; XGBoost; Zero Downtime.
Abstract. Downtime in industrial machinery can cause significant losses in productivity and operational efficiency. Various maintenance methods such as Corrective, Descriptive, Diagnostic, and Predictive Maintenance have limitations in optimizing downtime mitigation strategies. Therefore, this study implements Prescriptive Maintenance using Machine Learning (XGBoost) to not only predict machine failures but also provide corrective recommendations to achieve zero downtime. The AI4I 2020 Predictive Maintenance Dataset is utilized, incorporating several preprocessing techniques, including Min-Max Scaling, SMOTE, Correlation Heatmap, VIF, and outlier detection using Z-Score and IQR. The XGBoost model is trained to predict the probability of machine failure, which is further analyzed using Feature Importance to identify the root cause of failures. Model evaluation results demonstrate 98.12% accuracy, 97.95% precision, 98.23% recall, and a 99.59% AUC-Score, proving the system’s reliability in detecting and classifying machine failures. With the implementation of this strategy, supported by IoT and real-time databases, industries can optimize maintenance efficiency, reduce unexpected downtime, and enhance operational reliability.
Keywords: Prescriptive Maintenance; Machine Learning; XGBoost; Zero Downtime.