Sistem Predictive Maintenance untuk Deteksi Anomali Kinerja Pompa Looping Menggunakan Klasifikasi One-Class SVM Terintegrasi IoT

Abstract

FAWAZ ALFATAMA PUTRA. Predictive Maintenance System for Detecting Performance Anomalies in Looping Pumps using Integrated IoT One-Class SVM Classification. Supervised by SOFIYANTI INDRIASARI. Operational monitoring of looping pumps in industrial settings currently relies on manual methods that are prone to human error and unable to provide early warnings. This study aims to develop an Internet of Things (IoT)-integrated predictive maintenance system to identify operational anomalies in Water for Injection (WFI) looping pumps. The data acquisition system was built using an ESP32-S3 microcontroller and a W5500 Ethernet module to ensure transmission stability in industrial areas, supported by a validated PZEM-004T sensor (MAPE 0.54%) to read voltage and current parameters. The anomaly detection process was designed using a hybrid architecture that combines a rule-based filter (current threshold < 2A) to mitigate indications of dry running, along with the One-Class Support Vector Machine (OC-SVM) classification algorithm. The OC-SVM algorithm has proven effective at mapping normal operational load profiles without requiring data normalization, thereby preserving the integrity of the original physical values in forming a precise decision boundary. Based on field operational testing of 2,812 test data samples, the system demonstrated a very high level of reliability with a Precision of 95.76% and a Recall of 97.41%. When the system identifies an anomaly indication, the early warning mechanism is automatically triggered via WhatsApp notifications, the monitoring dashboard, and on-site buzzer alarms. Synergy of the OC-SVM algorithm has proven effective at mapping normal operational load profiles without requiring data normalization, thereby preserving the integrity of the original physical values to form precise decision boundaries. Based on field operational testing of 2,812 test data samples, the system demonstrated a very high level of reliability, achieving a Precision of 95.76% and a Recall of 97.41%. When the system identifies indications of an anomaly, the early warning mechanism is automatically triggered via WhatsApp notifications, the monitoring dashboard, and on-site buzzer alarms. This synergy between hardware and software has proven capable of minimizing the risk of catastrophic mechanical failure and preventing unplanned downtime in plant operations.

FAWAZ ALFATAMA PUTRA. Sistem Predictive Maintenance untuk Deteksi Anomali Kinerja Pompa Looping Menggunakan Klasifikasi One-Class SVM Terintegrasi IoT. Dibimbing oleh SOFIYANTI INDRIASARI. Pemantauan operasional pompa looping di lingkungan industri saat ini masih bergantung pada metode manual yang rentan terhadap human error dan tidak mampu memberikan peringatan dini. Penelitian ini bertujuan untuk mengembangkan sistem predictive maintenance terintegrasi Internet of Things (IoT) guna mengidentifikasi penyimpangan operasional pada pompa looping Water for Injection (WFI). Sistem akuisisi data dibangun menggunakan mikrokontroler ESP32-S3 dan modul Ethernet W5500 untuk menjamin stabilitas transmisi di area industri, didukung oleh sensor PZEM-004T yang telah divalidasi (MAPE 0,54%) untuk membaca parameter tegangan dan arus. Proses deteksi anomali dirancang menggunakan arsitektur hybrid yang mengkombinasikan filter rule-based (ambang batas arus < 2A) sebagai mitigasi indikasi dry running, serta algoritma klasifikasi One-Class Support Vector Machine (OC-SVM). Algoritma OC-SVM terbukti efektif memetakan profil beban operasional normal tanpa memerlukan proses normalisasi data, sehingga integritas nilai fisika asli tetap terjaga dalam membentuk decision boundary yang presisi. Berdasarkan pengujian operasional lapangan terhadap 2.812 sampel data uji, sistem menunjukkan tingkat keandalan yang sangat tinggi dengan capaian Precision sebesar 95,76% dan Recall sebesar 97,41%. Ketika sistem mengidentifikasi adanya indikasi anomali, mekanisme peringatan dini akan terpicu secara otomatis melalui notifikasi WhatsApp, dashboard pemantauan, serta alarm buzzer di lokasi. Sinergi perangkat keras dan perangkat lunak ini terbukti mampu meminimalkan risiko kerusakan mekanis fatal dan mencegah terjadinya unplanned downtime pada operasional pabrik.