Sistem Prediksi Kualitas Berdasarkan Analisis Profil Torsi Motor Servo Pada Auto Bellow Machine Menggunakan Metode LSTM Autoencoder

Maulana, Kevin Safrisal (2026) Sistem Prediksi Kualitas Berdasarkan Analisis Profil Torsi Motor Servo Pada Auto Bellow Machine Menggunakan Metode LSTM Autoencoder. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 2040221125-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Pemantauan kualitas perakitan bodi mesin cuci pada Auto Bellow Machine saat ini masih mengandalkan inspeksi visual manual yang rentan memicu escaped defect (keretakan struktural) dan pemborosan material. Penelitian ini bertujuan mengembangkan sistem prediksi kualitas unit menggunakan arsitektur Long Short-Term Memory (LSTM) Autoencoder untuk memprediksi kegagalan mekanis secara proaktif. Metode yang diusulkan meliputi fusi data Torsi dan Droop Pulses dari aktuator motor servo, penerapan strategi multi-model untuk empat tipe produk (MW, NT, CR, 10KG), serta penentuan ambang batas berdasarkan aturan tiga sigma (3σ) pada prediction error. Inferensi AI difilter menggunakan Hybrid Debounce dan Extreme Bypass Filter via Node-RED sebelum memicu Safety Interlock pada PLC Mitsubishi FX5U. Evaluasi menggunakan metrik MAE dan RMSE menunjukkan model stabil memprediksi profil torsi normal dengan MAE 0,0055 (MW), 0,0062 (NT), 0,0106 (CR), dan 0,0161 (10KG), yang seluruhnya berada di bawah batas kritis. Sebaliknya, prediksi pada data anomali menghasilkan lonjakan galat tajam melampaui threshold. Sistem ini terbukti merepresentasikan penurunan kualitas pergerakan secara kuantitatif dan sukses mengeksekusi preventive interlock dengan latensi di bawah 100 milidetik, sehingga mesin otomatis berhenti sebelum material patah demi menjamin kualitas zero defect.
===========================================================================================================================================
Quality monitoring of washing machine body assembly at the Auto Bellow Machine currently relies on manual visual inspection, which is prone to causing escaped defects (structural cracks) and material waste. This study aims to develop a unit quality prediction system using the Long Short-Term Memory (LSTM) Autoencoder architecture to proactively predict mechanical failures. The proposed method includes the data fusion of Torque and Droop Pulses from servo motor actuators, the implementation of a Multi-Model strategy for four product types (MW, NT, CR, 10KG), and the determination of a threshold based on the Three-Sigma (3σ) rule on the Prediction Error. The AI inference is filtered using Hybrid Debounce and Extreme Bypass Filter via Node-RED before triggering the Safety Interlock on the Mitsubishi FX5U PLC. Evaluation using MAE and RMSE metrics indicates the model stably predicts normal torque profiles with MAEs of 0.0055 (MW), 0.0062 (NT), 0.0106 (CR), and 0.0161 (10KG), all remaining below the critical threshold. Conversely, predictions on anomaly data produce sharp error spikes exceeding the threshold. The system is proven to quantitatively represent movement quality degradation and successfully executes a preventive interlock with a latency of under 100 milliseconds, automatically halting the machine before material breakage to guarantee zero defect quality.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Prediksi Kualitas, LSTM Autoencoder, Programmable Logic Controller, Profil Torsi, Motor Servo, Quality Prediction, Torque Profile, Servo Motor.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK3070 Automatic control T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Vocational > 36304-Automation Electronic Engineering
Depositing User:	Kevin Safrisal Maulana
Date Deposited:	05 Aug 2026 03:58
Last Modified:	05 Aug 2026 03:58
URI:	http://repository.its.ac.id/id/eprint/144012

Actions (login required)

View Item