Machine Learning Toolbox and PCA Visualization for Data-Driven PHM

Abstract

Increased awareness of big data has led to active development of machine learning algorithms for big data analytics. The advent of rapidly emerging data analytics technologies has also brought about considerable changes to the diagnostics and prognostics for smart manufacturing industries. As the importance of managing massive factory data also has grown, many engineers are putting in efforts to implement machine learning algorithms for a PHM (Prognostics and Health Management) purpose in accordance with the type of machinery of interest. In this thesis, research on assisting the quick deployment of supervised and unsupervised learning classification algorithms with data visualization is conducted by building up the GUI software with an emphasis on PHM. It can plot raw data, select hand-crafted features based on an expert knowledge, followed by a dimension reduction step if necessary. The various machine learning algorithms will provide the classification decision boundaries to enable us to diagnose current machine health conditions. Therefore, it can suggest to engineers a guideline to determine suitable features and date-driven PHM algorithms. Principal Component Analysis (PCA) is a widely used dimension reduction algorithm without losing too much information for high-dimensional data analysis. It transforms the high-dimensional data into a meaningful representation of reduced dimensional data. In a machine health monitoring system, a result of dimension reduction via PCA is often utilized. Although eigenvectors and eigenvalues of PCA are important information, it is too difficult for users to interpret where the principal components are coming from. In order to assist the user in better understanding and interpreting PCA, data visualization can be used. We have developed a system that visualizes the eigenvectors and eigenvalues of PCA using JavaScript library, D3 and demonstrated that how the key information and insights of PCA results can be intuitively visualized.