Application of machine learning for asset revenue optimization through equipment performance prediction using inline instrument data : a surface condenser case study/ Mohamad Firdaus bin Mohammed Basheer
Material type:
TextPublication details: Bandar Seri Begawan : Universiti Teknologi Brunei, 2024. Description: xii, 236 pages : color illustrations ; 30 cmSubject(s): Machine learning -- Industrial applications | Predictive maintenance -- Mathematical models | Asset management, Physical -- Automation | Operations research -- Data processing | Surface condensers -- Performance -- Prediction -- Data processingOther classification: UTB 120 REPORT, THESIS & DISSERTATION Dissertation note: Dissertation (Doctor of Philosophy) - Universiti Teknologi Brunei (2024)
| Item type | Current library | Call number | Status | Notes | Date due | Barcode |
|---|---|---|---|---|---|---|
Reports, Thesis & Dissertation Students
|
Universiti Teknologi Brunei Library - at level 2 | UTB 120 REPORT, THESIS & DISSERTATION, RTDS 404 (Browse shelf(Opens below)) | Not for loan | Reg. no. 002448_UTB [RTDS 404] | 850576 |
Submitted in PHD fulfilment of the requirements for the degree of Doctor of Philosophy.
Abstract
Equipment performance assessment or prediction has usually been derived using the conventional approach and heavily relies on expert opinion, where the methods used often vary from one person to another depending on the knowledge they possess. Many assets in the manufacturing industries are fitted with sensors on their equipment and this sensor could deliver values from both technical and business aspects. In this research work, a surface condenser was used as a case study to investigate machine learning methods that could be applied to the inline instrument data to predict the performance of the surface condenser and subsequently develop a cost-effective model for maintenance intervention decision-making. To date, most studies only focus on the technical improvement and prediction of specific technical parameters of surface condensers, and none of the research explores the holistic approach in attempting to link technical prediction to operating cost savings.
In this study, the excess consumption (steam, condensate and water) due to the surface condenser inefficiencies were predicted through the comparison of Deep Learning method, Multi-Layer Perceptron Regression method, Support Vector Regression method and Multivariate Time Series – LSTM method. The Multivariate Time Series -LSTM method responds well to one year training time span with MAPE of 25% compared to MAPE of 43% for four year training time span. The deep learning method was seen as the most promising method, with an average EVS score of 88% for the multi-output prediction of steam, condensate, and water compared to EVS score of 74% and 50% for NNMLPR and SVR respectively. Based on the outcome of the economics assessment in justifying the performance of the surface condenser predicted effectiveness from 0.37 to 0.68, about 4%, 7% and 17% cost savings can be realised from the removal of wastes from steam, condensate and cooling water which equates to a total of $3,014,513.36 based on the same production capacity and period of test data.
Dissertation (Doctor of Philosophy) - Universiti Teknologi Brunei (2024)
Includes bibliographical references and index.
There are no comments on this title.