Pengembangan Diagram Kontrol Multivariat Support Vector Data Description Berbasis Adaptive Exponentially Weighted Moving Average (D’aewma)

Jaya, Andi Indra (2023) Pengembangan Diagram Kontrol Multivariat Support Vector Data Description Berbasis Adaptive Exponentially Weighted Moving Average (D’aewma). Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6003201028-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 April 2025. Download (4MB) | Request a copy

Abstract

Dalam beberapa tahun terakhir, sistem manufaktur modern mulai diintegrasikan pada metode Statistical Process Control berbasis data mining. Hal ini dikarenakan distribusi data saat ini relatif bervariasi sehingga memungkinkan sulit untuk menemukan sebuah pola dalam proses monitoring. Sejauh ini dalam melakukan proses monitoring masih menggunakan metode konvensional yaitu diagram kontrol T2-Hotelling, Multivariate Cumulative Sum (MCUSUM) dan Multivariate Exponentially Weighted Moving Average (MEWMA). Meskipun demikian, dalam penerapannya metode-metode konvensional tersebut memiliki kekurangan, yakni memiliki performa yang kurang baik saat diterapkan pada data yang berdistribusi non normal. Akibatnya, penggunaan diagram kontrol konvensional menjadi kurang maksimal dikarenakan banyaknya out of control yang terdeteksi meskipun proses sebenarnya masih in control. Oleh karena itu, salah satu metode untuk mengatasi kekurangan tersebut adalah one-class classification (OCC). OCC sendiri pada dasarnya berfungsi untuk meminimumkan terjadinya false alarm. Berdasarkan jenis pendekatan dalam penyelesaian metode OCC, terdapat tiga kategori utama yaitu pendekatan density estimation, reconstruction methods, dan boundary description. Dari ketiga ketegori tersebut, penelitian ini berfokus pada pendekatan boundary description dengan metode Support Vector Data Desription (SVDD) karena pada penerapannya memungkinkan untuk belajar melalui distribusi karakteristik data yang relatif bervariasi atau bahkan tidak diketahui. Dalam penerapannya, SVDD memiliki kelemahan saat diterapkan pada peta kendali ketika mengalami pergeseran proses yang fluktuatif. Diagram kontrol Adaptive Exponentially Weighted Moving Average (AEWMA) dapat diterapkan untuk kasus seperti ini. Oleh sebab itu, dalam penelitian ini diusulkan diagram kontrol SVDD berbasis AEWMA (D’AEWMA) guna meminimumkan terjadinya false alarm (peringatan bahaya palsu) serta memberikan performa terbaik saat terjadi pergeseran proses mean yang relatif kecil maupun besar. Untuk mengetahui diagram kontrol yang diusulkan memberikan hasil yang optimal dalam mendeteksi sinyal out-of-control, dilakukan suatu perbandingan dengan diagram kontrol SVDD berbasis EWMA (D’EWMA). Berdasarkan hasil analisis, diagram kontrol D’AEWMA memberikan performa yang cukup optimal untuk mendeteksi sebanyak mungkin sinyal out-of-control dan meminimalisir terjadinya false alarm yang dibuktikan melalui nilai ARL1 yang lebih kecil. Hasil analisis diperoleh melalui penerapan data sintesis maupun data karakteristik kualitas semen (clinker) di PT XYZ.
==============================================================================================================================
In recent years, modern manufacturing systems have begun to be integrated into the Statistical Process Control method based on data mining. This is because the current distribution of data is relatively varied, making it difficult to find a pattern in the monitoring process. So far, the monitoring process is still using conventional methods, namely the T2-Hotelling control chart, Multivariate Cumulative Sum (MCUSUM) and Multivariate Exponentially Weighted Moving Average (MEWMA). However, in its application, these conventional methods have drawbacks, namely they have poor performance when applied to data that are non-normally distributed. As a result, the use of conventional control charts is less than optimal due to the large number of out of control detected even though the actual process is still in control. Therefore, one of the methods that can be used to overcome these deficiency is one-class classification (OCC). OCC itself basically functions to minimize the occurrence of false alarms. Based on the type of approach in solving the OCC method, there are three main categories: density estimation, reconstruction methods, and boundary description. Of the three categories, this research focuses on the boundary description approach with the Support Vector Data Desription (SVDD) method because its application allows learning through the distribution of relatively variable or even unknown data characteristics. In its application, SVDD has a weakness when applied to control maps when experiencing volatile process shifts. Adaptive Exponentially Weighted Moving Average (AEWMA) control charts can be applied for such cases. Therefore, in this study, an AEWMA-based SVDD control chart (D'AEWMA) is proposed to minimize the occurrence of false alarms and provide the best performance when there is a relatively small or large shift in the process mean. To determine whether the proposed control diagram provides optimal results in detecting out-of-control signals, a comparison is made with the EWMA-based SVDD control diagram (D'EWMA) Based on the analysis, the D'AEWMA control diagram provides optimal performance to detect as many out-of-control signals as possible and minimize the occurrence of false alarms as evidenced by the smaller ARL1 value. The analysis result are obtained through the application of syntesized data and characteristic data of cement quality (clinker) at PT XYZ.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Diagram Kontrol, Multivariat, SVDD, Adaptive EWMA, kualitas semen (clinker), Control Chart, Multivariate, cement quality (clinker).
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA76.6 Computer programming.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49101-(S2) Master Thesis
Depositing User:	Andi Indra Jaya
Date Deposited:	30 Jan 2023 15:22
Last Modified:	30 Jan 2023 15:22
URI:	http://repository.its.ac.id/id/eprint/95821

Actions (login required)

View Item