Pembuatan Sistem Prediksi Degradasi Resistansi Pada Heater Mesin HSM MCB PS Menggunakan Metode XGBoost

Ardiansyah, M. Putra (2026) Pembuatan Sistem Prediksi Degradasi Resistansi Pada Heater Mesin HSM MCB PS Menggunakan Metode XGBoost. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221130-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Mesin heat sealing memiliki peran penting dalam proses produksi baterai karena berpengaruh langsung terhadap kualitas dan keamanan produk, dengan heater sebagai salah satu komponen kritis yang rentan mengalami penurunan kinerja akibat suhu operasi tinggi dan fluktuasi tegangan listrik. Kerusakan heater yang tidak terdeteksi sejak dini menyebabkan downtime produksi yang signifikan pada lini MCB PS PT. GS Battery Semarang, sementara pendekatan pemeliharaan berbasis jadwal (Time-Based Maintenance) dinilai kurang efisien karena tidak mempertimbangkan kondisi aktual komponen. Penelitian ini bertujuan mengembangkan sistem prediksi resistansi heater berbasis Extreme Gradient Boosting (XGBoost) sebagai alat pemantauan kondisi (condition monitoring) sekaligus pemberi indikasi dini (early warning) dan juga sebagai acuan untuk penggantian heater berdasarkan data resistansinya. Data yang digunakan meliputi tegangan, arus, dan resistansi yang dikumpulkan secara berkala, kemudian diolah menggunakan pendekatan rolling one-step-ahead forecast dengan skema walk-forward validation, di mana fitur lag 1 hingga 3 hari kerja beserta turunannya (rolling mean, rolling std, slope, dan first-difference) dipilih berdasarkan hasil analisis autokorelasi. Pada tahap evaluasi data uji, model menghasilkan R² (koefisien determinasi) sebesar 0,233 dengan residual rata-rata +1,475 Ω dan simpangan baku 4,008 Ω. Pada pengujian forecasting empat hari (9–12 Maret 2026), model menghasilkan nilai MAE bervariasi antara 1,5969 Ω hingga 8,2560 Ω dan RMSE antara 1,8520 Ω hingga 9,2946 Ω, dengan performa terbaik pada 10 Maret 2026 (MAE 1,5969 Ω; R² 0,3961) dan performa terlemah pada 12 Maret 2026 (MAE 8,2560 Ω; R² −6,4722). Model XGBoost secara konsisten menghasilkan MAE lebih rendah dibandingkan metode baseline persistence pada seluruh hari pengujian, namun belum mampu mengungguli baseline rata-rata aktual ketika terjadi perubahan resistansi secara mendadak (structural break).
====================================================================================================================================
Heat sealing machines play a crucial role in the battery production process because they directly impact product quality and safety; the heater is one of the critical components that is prone to performance degradation due to high operating temperatures and voltage fluctuations. Heater failures that are not detected early on cause significant production downtime on the MCB PS line at PT. GS Battery Semarang, while a time-based maintenance approach is considered inefficient because it does not take into account the actual condition of the components. This study aims to develop an Extreme Gradient Boosting (XGBoost)-based heater resistance prediction system to serve as a condition monitoring tool, an early warning system, and a reference for heater replacement based on resistance data. The data used includes voltage, current, and resistance collected periodically, which were then processed using a rolling one-step-ahead forecast approach with a walk-forward validation scheme, where features with lags of 1 to 3 business days and their derivatives (rolling mean, rolling standard deviation, slope, and first difference) were selected based on the results of an autocorrelation analysis. During the test data evaluation phase, the model produced an R² (coefficient of determination) of 0.233 with an average residual of +1.475 Ω and a standard deviation of 4.008 Ω. In the four-day forecasting test (March 9–12, 2026), the model produced MAE values ranging from 1.5969 Ω to 8.2560 Ω and an RMSE ranging from 1.8520 Ω to 9.2946 Ω, with the best performance on March 10, 2026 (MAE 1.5969 Ω; R² 0.3961) and the weakest performance on March 12, 2026 (MAE 8.2560 Ω; R² −6.4722). The XGBoost model consistently produced a lower MAE compared to the persistence baseline method across all test days; however, it was unable to outperform the actual average baseline when sudden changes in resistance (structural breaks) occurred.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Mesin Heat Sealing, Prediksi Resistansi Heater, Extreme Gradient Boosting (XGBoost), Predictive Maintenance, Heat Sealing Machine, Heater Resistance Prediction, Extreme Gradient Boosting (XGBoost), Predictive Maintenance.
Subjects:	Q Science > QA Mathematics > QA9.58 Algorithms T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7871.674 Detectors. Sensors
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	M. Putra Ardiansyah
Date Deposited:	14 Aug 2026 08:52
Last Modified:	14 Aug 2026 08:52
URI:	http://repository.its.ac.id/id/eprint/143130

Actions (login required)

View Item