Sistem Klasifikasi Pada Kegagalan Stasiun Uji Insulation Withstanding Mesin Cuci

Abhinaya, Reynal Viverio (2026) Sistem Klasifikasi Pada Kegagalan Stasiun Uji Insulation Withstanding Mesin Cuci. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221139-Undergraduate_Thesis .pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Stasiun uji isolasi (insulation test station) pada lini produksi mesin cuci di PT Sharp Electronics Indonesia (SEID) menghadapi permasalahan ambiguitas kualitas, yaitu kondisi di mana sistem mendeteksi status Not Good (NG) pada pallet dan status Fail pada unit secara bersamaan, sehingga menyulitkan proses pengambilan keputusan terkait akar penyebab kegagalan. Sistem inspeksi berbasis logika ambang batas (threshold) pada Programmable Logic Controller (PLC) yang digunakan saat ini rentan terhadap gangguan sensor akibat interferensi elektromagnetik di lingkungan pabrik, sehingga berpotensi menurunkan akurasi klasifikasi kegagalan. Proyek akhir ini mengembangkan sistem pendukung keputusan (Decision Support System) berbasis Machine Learning untuk mengklasifikasikan status kualitas True Fail atau False Fail berdasarkan tiga parameter elektrikal, yaitu Resistansi Isolasi, Tegangan Uji, dan Arus Bocor, yang diperoleh melalui integrasi data (data fusion) antara PLC Mitsubishi FX5U dan sistem RFID menggunakan middleware Node-RED. Penelitian ini melakukan evaluasi komparatif terhadap enam algoritma klasifikasi, yaitu Decision Tree, Logistic Regression, K-Nearest Neighbors, dan Support Vector Machine sebagai representasi model tunggal, serta Random Forest dan Extreme Gradient Boosting (XGBoost) sebagai representasi model ensemble. Ketidakseimbangan kelas yang ekstrem pada dataset awal (rasio 418:1) ditangani melalui kombinasi random undersampling dan penambahan data historis, menghasilkan dataset final yang lebih proporsional. Hasil evaluasi menunjukkan temuan yang berlawanan dengan asumsi umum bahwa model ensemble senantiasa unggul dibandingkan model tunggal: Logistic Regression dan Decision Tree justru menunjukkan performa terbaik (akurasi 91,33% dan 90,17%), sementara XGBoost mencatatkan performa terendah (79,77%). Analisis lebih lanjut mengungkap bahwa fenomena ini berakar dari karakteristik data yang bersifat deterministik dan berbasis ambang batas tegas. Analisis kelayakan komputasi turut membuktikan bahwa seluruh model layak diimplementasikan pada sistem waktu nyata, dengan waktu inferensi jauh di bawah satu persen dari waktu siklus produksi (18 detik). Berdasarkan pertimbangan akurasi dan interpretabilitas, Decision Tree direkomendasikan sebagai arsitektur optimal untuk implementasi praktis di lini produksi.
=====================================================================================================================================
The insulation test station on the washing machine production line at PT Sharp Electronics Indonesia (SEID) faces an issue of quality ambiguity, a condition where the system simultaneously detects a Not Good (NG) status on the pallet and a Fail status on the unit, complicating the decision-making process regarding the root cause of failure. The current threshold-based inspection system on the Programmable Logic Controller (PLC) is susceptible to sensor interference due to electromagnetic interference in the factory environment, potentially reducing failure classification accuracy. This final project develops a Machine Learning-based Decision Support System (DSS) to classify quality status as True Fail or False Fail based on three electrical parameters: Insulation Resistance, Test Voltage, and Leakage Current. These parameters are obtained through data fusion between the Mitsubishi FX5U PLC and the RFID system using Node-RED middleware. This study conducts a comparative evaluation of six classification algorithms: Decision Tree, Logistic Regression, K-Nearest Neighbors, and Support Vector Machine as single models, along with Random Forest and Extreme Gradient Boosting (XGBoost) as ensemble models. The extreme class imbalance in the initial dataset (a ratio of 418:1) was addressed through a combination of random undersampling and the addition of historical data, resulting in a more proportionally balanced final dataset. The evaluation results reveal findings contrary to the common assumption that ensemble models always outperform single models: Logistic Regression and Decision Tree demonstrated the best performance (accuracies of 91.33% and 90.17%, respectively), while XGBoost recorded the lowest performance (79.77%). Further analysis indicates that this phenomenon stems from the deterministic and strictly threshold-based nature of the data. Computational feasibility analysis also confirms that all models are viable for real-time implementation, with inference times well below one percent of the production cycle time (18 seconds). Based on considerations of accuracy and interpretability, the Decision Tree is recommended as the optimal architecture for practical implementation on the production line.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Resistansi isolasi, klasifikasi kegagalan, Machine Learning, Decision Tree, Logistic Regression, model ensemble, sistem pendukung keputusan. ============================================================ Insulation resistance, failure classification, Machine Learning, Decision Tree, Logistic Regression, ensemble models, decision support system.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Reynal Viverio Abhinaya
Date Deposited:	14 Aug 2026 02:40
Last Modified:	14 Aug 2026 02:40
URI:	http://repository.its.ac.id/id/eprint/144144

Actions (login required)

View Item