In today’s highly competitive business environment, every company seeks to work at their full potential to keep up with competitors and stay in the market. Manager and engineers, therefore, constantly try to develop technologies to improve their product quality. Advancements in online sensing technologies and communication networks have reshaped the competitive landscape of manufacturing
systems, leading to exponential growth of Condition Monitoring (CM) data. High-dimensional data sources can be particularly important in process monitoring and their efficient utilization can help systems reach high accuracy in fault diagnosis and prognosis. While researches in Statistical Process Control (SPC) tools and Condition-Based Maintenance (CBM) are tremendous, their applications
considering high-dimensional data sets has received less attention due to the complexity and challenging nature of such data and its analysis. This thesis adds to this field by designing a
Deep Learning (DL) based survival analysis model towards CBM in the prognostic context and a DL and SPC based hybrid model for process diagnosis, both using high dimensional data. In the
first part, we a design support system for maintenance decision making by considering degradation signals obtained from CM data. The decision support system in place can predict system’s failure
probability in a smart way. In the second part, a Fast Region-based Convolutional Network (Fast R-CNN) model is applied to monitor the video input data. Then, specific statistical features are derived from the resulting bounding boxes and plotted on the multivariate Exponentially Weighted Moving Average (EWMA) control chart to monitor the process.