MAKİNE ÖĞRENİMİ VE DERİN ÖĞRENME TEMELLİ HATA TAHMİN MODELLERİNİN GELİŞTİRİLMESİ
Project no. 22750 · Status: Kapanmış · 2021
Summary
<p>This study addresses the fault diagnosis problem based on vibration data. Various models have been proposed for the problem based on Decision Trees (DT), Random Forest (RF), Support Vector Machines (SVM), and Multilayer Perceptron (MLP) machine learning methods. In the models, various time-based statistical features extracted from the vibration data have been used both directly and after the application of various dimension reduction approaches to these features. To reduce the dimension of the data set, both linear and non-linear dimension reduction techniques have been used. As the linear dimension reduction technique, one of the most widely applied techniques, Principal Component Analysis (PCA) has been applied. On the non-linear side, Kernel Principal Component Analysis (kPCA), Isometric Feature Mapping (Isomap), and Local Linear Embedding (LLE) have been implemented. Before implementing all the developed model variations to the real-life problem, the model variations have been tested on the widely used benchmark test data of bearing data of Case Western Reserve University. The model variations with successful prediction performances on test data have been used in the data of the real-life problem and their performances have been evaluated.</p>