Prediction with machine learning

• Main Author: Soh, Melvin Yong Sheng
• Other Authors: Zhong Zhaowei
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Pattern recognition; Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/139202
• Institution: Nanyang Technological University

The focus of this final year project is on the maintenance of transformers. Maintenance is part of daily tasking for an engineer. It is required for equipment to run smoothly. Something equipment may spoil without any warnings. The conditions of transformers are not easy to check as the machine are normally running 24 hours and seven days a week. Therefore, it is important to come up with a way to predict failures. Using Machine Learning, it is no longer a need for engineering to depend on their intuition or experience to predict malfunctions. There are many different types of machine learning algorithms, hence we need to see how accuracy each of the predictions is. The main objective of this final year project is to develop an application that is designed to do a prediction of the condition of the machines. There will be an unbalance amount of data, as there will be many more working machines compared to a failed machine. This will affect the training of the algorithm models. To increase the accuracy of the model, oversampling techniques, like Synthetic Minority Over-sampling Technique, Borderline Synthetic Minority Over-sampling Technique, and Adaptive Synthetic, will be used. To make a prediction using we need to incorporate the machine learning algorithms into the graphical user interface. In this project, we will be focusing on Random Forest, Isolation Forest, and Deep Learning. Random Forest will provide the basic line for the accuracy of the result, while Isolation Forest and Deep Neural Network are for more advanced users. The application was designed and developed using the Python Language on the Integrated Developer Environment called PyCharm. The user will load the dataset into the application using Microsoft Excel. The results have shown that using the Random Forest Algorithm produces the most stable accurate, while Isolation Forest and Deep Neural Network may not work well in our example, it could still be used in our cases.