Implementasi Penggunaan Model Predictive Analytics Pada Sistem Predictive Maintenance Pembangkit Listrik

Budi, Santoso (2022) Implementasi Penggunaan Model Predictive Analytics Pada Sistem Predictive Maintenance Pembangkit Listrik. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6032201032-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2024. Download (6MB) | Request a copy

Abstract

Deteksi anomalyberbasis model machine learningmerupakan pendekatan yang banyak dikembangkan saat ini untuk optimasi dan peningkatan efektivitas predictive maintenance(PDM) serta peningkatan kehandalan pembangkit listrik termal. Penelitian bertujuan untuk mengembangkan model berbasis data untuk diagnostik maupun prognostik peralatan, untuk menghasilkan prediksi yang akurat dengan digunakannya banyak data sensor yang diambil secara real-timedari sistem supervisory control and data acquisition(SCADA), serta merancang frameworkmanajemen PDMdengan penggunaan sistem deteksi anomaly. Induced Draft Fan(IDF) dan Primary Air Fan(PAF) merupakan peralatan kritikal dalam pembangkit listrik tenaga uap yang digunakan sebagai obyek studi kasus dalam penelitian ini.Pada penelitian ini diusulkan penggunaan gabungan pendekatan recurrent neural network(RNN)-autoencodersebagai normal behavior model(NBM) dengan metode statistik Mahalanobis Distance(MD) untuk deteksi anomalyperalatan pembangkit. Model RNN–Autoencoderdigunakan untuk memodelkan perilaku normal peralatan berdasarkan data input timeseriesdari sensor, sedangkan MD digunakan untuk menentukan jarak antara parameter aktual peralatan dengan prediksi perilaku normalnya untuk menentukan kondisi anomaly. Pada penelitian ini dilakukan optimasi hyperparameterpada model long short-term memory(LSTM)dangated recurrentunit(GRU). LSTM dengan hyperparameterterbaik mampu mencapai validation loss 5,690 x 10-4dan akurasi validasi 93,36%, sedangkan GRU dengan menggunakan hyperparameter terbaik mampu mencapai validation accuracysebesar 4,484 x 10-4dan akurasi validasi 93,47%. GRU dapat mencapai performancelebih baik dibandingkan model LSTM. Frameworkyang diusulkan ini dapat mendeteksi kondisi anomaly dengan baik pada beberapa kasus gangguan peralatan IDF dan PAF baik sebagai early warning sebelum breakdown peralatan maupun pada saat terjadinya kondisi downtime.
===========================================================================================================================
Anomalydetection based on machine learningmodels is an approach that is currently being developed for optimization and increasing the effectiveness of predictive maintenance (PDM) as well as increasing the reliability of thermal power plants. The research aims to develop a data-driven model for diagnostic and prognostic equipment, produce accurate predictions using many sensor data taken in real-timefrom the supervisory control and data acquisition (SCADA)system, and design a PDMmanagement frameworkusing an anomalydetection system. Induced Draft Fans (IDF) and Primary Air Fans (PAF) are critical equipment in steam power plants that are used as study case objects in this research. This research proposes to use a combined recurrentneural network (RNN)-autoencoderapproach as a "normal" behavior model (NBM) with the Mahalanobis Distance(MD) statistical method for the detection of anomalies in power plant equipment. Based on time-series input sensor data, the RNN –autoencoderis utilized to predict the behavior of the equipment in health circumstances. In contrast, the MD is used to determine the distance between the actual parameters of the equipment and its "normal" behavior prediction to determine the anomalycondition. This study examined the performanceof the long short-term memory (LSTM) and gated recurrentunit (GRU) models in modeling normal behavior with hyperparameteroptimization. The LSTM with the best hyperparameters had a validation loss of 5,690 x 10-4and a validation accuracy of 93.36 percent, whereas the GRU with the best hyperparameters had a validation loss of 4.484 x 10-4and a validation accuracy of 93.47 percent. GRU can outperform the LSTM model. The proposed frameworkcan detect anomalyconditions in various cases of IDF and PAF equipment disturbances, both in early warningof equipment failure and during downtime conditions.

Item Type:	Thesis (Masters)
Additional Information:	RTMT 658.202 Bud i-1 2022
Uncontrolled Keywords:	Anomalydetection, LSTM, GRU, auto encoder, normal behavior model, Mahalanobis Distance, Predictive Maintenance
Subjects:	Q Science > QA Mathematics > QA76.76.S64 Software maintenance.
Divisions:	Interdisciplinary School of Management and Technology (SIMT) > 61101-Master of Technology Management (MMT)
Depositing User:	Anis Wulandari
Date Deposited:	09 Nov 2022 05:36
Last Modified:	09 Nov 2022 05:36
URI:	http://repository.its.ac.id/id/eprint/95075

Actions (login required)

View Item