Prediction of Machine Health Stability using Deep Learning

다국어 초록 (Multilingual Abstract)

Estimating Machine Health Stability (MHS) is the best way to understand and get insight into the machine's stability in the manufacturing process. In addition, MHS provides usefulness to the end-user, such as making a report of the diagnostic or sometimes specific task that is essential for the company to help prevent any failure occurring, which saves many resources and helps create a beneficial decision-making process. Many researchers propose predictive maintenance methods using statistical process analysis, machine learning, and deep learning methods. This thesis proposes a methodology to predict machine health stability using deep learning methods. Before proceeding to the prediction, we collected data from car parts manufacturing from 2020 to 2022 in South Korea. We added additional features to the existing research work, including product-making conditions and the improvement of alarm-type features. Then we extracted the essential features by using a data analysis program. Following this process, we could select related features to predict Machine Health Stability using deep learning techniques. We then compare our model performance using evaluation metrics such as MSE, RMSE, MAE, and R-Squared. The result found that LSTM performs better than other methods.

목차 (Table of Contents)

• Chapter 1. INTRODUCTION 1
• 1.1. Motivation 1
• 1.2. Machine Health Stability 2
• 1.3. Problem Definition 3
• 1.4. Contribution 4
• 1.5. Thesis organization 5
• Chapter 2. RELATED RESEARCH 7
• 2.1. Related Works 7
• 2.2. The Comparison of Related Works 10
• 2.3. Summary and Implications 12
• Chapter 3. Proposed method and GLOBAL FLOWCHART OF MHS 13
• Chapter 4. Data preparation and features selection for mhs 16
• 4.1. Data Gathering 16
• 4.2. Product Making Condition 17
• 4.3. Alarm Type 19
• 4.4. Statistical Process Analysis 21
• 4.5. Process capability analysis 23
• 4.6. Fault Machine Data Structure 24
• 4.7. Defective Product Rate 25
• 4.8. Not Good Product Rate 26
• 4.9. Non-active Rate of Machine State 27
• 4.10. Manufacturing Cycle Time 27
• 4.11. Data Analysis program 29
• 4.12. The table structure of preprocessed data 29
• 4.13. Dataset Summarisation 31
• 4.14. Selected features for deep learning 31
• Chapter 5. PREDICTING MHS WITH DEEP LEARNING 34
• 5.1. Deep Learning for Machine Health Assessment 34
• 5.2. Recurrent Neural Network 35
• 5.3. Long Short-Term Memory 35
• 5.4. The framework of Machine Health Stability 36
• 5.5. Evaluation Method 40
• Chapter 6. Experimental result 42
• 6.1. System Environment Setup 42
• 6.2. Result 44
• Chapter 7. CONCLUSION AND FUTURE WORK 50
• BIBLIOGRAPHY 52
• Acknowledgements 56