Condition Based Maintenance Menggunakan Deep Learning Dalam Sistem Raw Mill Produksi Semen

Yusuf, Fauzi Irfandi (2023) Condition Based Maintenance Menggunakan Deep Learning Dalam Sistem Raw Mill Produksi Semen. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 02411940000085-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 September 2025. Download (3MB) | Request a copy

Abstract

Sistem Raw Mill merupakan bagian krusial dalam proses produksi semen di pabrik Tuban PT XYZ. Sejak tahun 2021 hingga Februari 2023, Raw Mill System mengalami downtime yang berkisar antara 16,05% hingga 42,79%. Saat ini, perusahaan menerapkan Corrective Maintenance dan Preventive Maintenance. Komponen mesin pada sistem telah dilengkapi dengan beberapa sensor namun tidak digunakan dalam konteks pemeliharaan. Strategi Condition Based Maintenance (CBM) dapat dikembangkan untuk mengoptimalkan strategi yang sedang berjalan saat ini. Penilaian prognostik sebagai bagian dari CBM berfokus pada prediksi Remaining Useful Life (RUL) sebelum terjadinya kerusakan pada peralatan. Tujuan dari penelitian ini adalah untuk membangun dan menentukan model regresi Machine Learning (ML) terbaik untuk prognostik sistem raw mill. Penelitian ini membandingkan performa antara model ML dangkal dan Deep Learning (DL). Algoritma Random Forest (RF) dipilih sebagai model ML dangkal karena fleksibilitasnya yang telah terbukti dan kinerja yang baik berdasarkan penelitian sebelumnya. Jaringan Long-Short Term Memory (LSTM) dipilih sebagai model DL karena kemampuannya untuk menyimpan memori dari prediksi sebelumnya yang sangat cocok untuk time-series forecasting. Input rentang waktu dibandingkan serta perbedaan dalam kinerja prediksi antara rentang prediksi yang berbeda. Kinerja model dievaluasi dengan menggunakan RMSE, MAE, MAPE dan R2. Dasbor dibangun untuk pemantauan dan visualisasi pengetahuan data mining secara real time. Model LSTM didesain dengan menggunakan LSTM layer dan dua Dense layer. Namun, performa dari model deep learning yang dirancang masih inferior dibandingkan dengan model RF yang digunakan dalam tes. Dasbor dibuat dengan basis software MATLAB dan dapat memuat data mentah, model training terhadap data baru, dan memprediksi kondisi motor yang dipilih.
==================================================================================================================================
Raw mill system is a crucial part of the cement production process of PT.XYZ Tuban production plant. From 2021 until February 2023, the Raw Mill System encountered downtime ranging between 16,05% to 42,79%. Currently, the company implements corrective and preventive maintenance. Machine components of the system have been equipped with multiple sensors, however not used in the context of maintenance. Condition Based Maintenance (CBM) strategy could be developed to optimize current ongoing strategies. Prognostics assessment as a part of CBM focuses on predicting Remaining Useful Life (RUL) before equipment failure. The purpose of this research is to build and determine the best Machine Learning (ML) regression model for raw mill system prognostics. This research compares the performance between shallow ML and Deep Learning (DL) models. Random Forest (RF) algorithm is chosen as the shallow ML model for its proven flexibility and good performance based on prior research. Long-Short Term Memory (LSTM) networks is chosen as the DL model for its ability to retain memories of previous prediction that is highly suitable for time-series forecasting. The time-span input is compared as well as difference in prediction performance between prediction horizons. Performance of models is evaluated using RMSE, MAE, MAPE and R2. A dashboard is constructed for real time monitoring and visualization of data mining knowledge. The displayed information is then used to support decision making process regarding maintenance and related actions. The LSTM deep learning model are designed with an LSTM layer and two dense layers. However, the performance of the deep learning model is inferior to the tested RF model. The dashboard is made based on MATLAB software and can load raw data, train models on new data, and predict the selected motor condition.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Raw Mill, Condition Based Maintenance (CBM), Prognostics, Random Forest (RF), Long Short-Term Memory (LSTM)
Subjects:	T Technology > TS Manufactures > TS174 Maintainability (Engineering) . Reliability (Engineering)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26201-(S1) Undergraduate Thesis
Depositing User:	Fauzi Irfandi Yusuf
Date Deposited:	25 Sep 2023 01:50
Last Modified:	25 Sep 2023 01:50
URI:	http://repository.its.ac.id/id/eprint/102175

Actions (login required)

View Item