Description
TitleData-enabled process monitoring and predictive analytics for smart manufacturing
Date Created2021
Other Date2021-05 (degree)
Extent1 online resource (x, 156 pages) : illustrations
DescriptionRecent developments in smart manufacturing and advanced sensing techniques have allowed researchers to collect voluminous, rich data continuously during manufacturing, demanding new methods to effectively harness the power of data science to enable process transparency and optimization. Valuable information regarding process condition and product quality needs to be unveiled to understand how manufacturing performance is affected by intertwined factors and uncertainties, potentially providing insights to improve operational decision-making. The goal of this Ph.D. research is to contribute to the knowledge base of smart manufacturing by effectively harnessing the power of data science. Specifically, this thesis presents four methods to address (1) the detection of process deteriorations with univariate time series, (2) the recognition of fault causes using multivariate time series (MTS), (3) the in-situ monitoring of additive manufacturing using thermal images, and (4) the augmentation of thermal image data in additive manufacturing to facilitate process analytics.
The first part of this thesis focuses on handling univariate time series whose data profile is complex (nonstationary, time-inconsistent and hardly modeled by common probability distributions) for identifying monotonic increasing/decreasing tendency, i.e., trends, associated with process deteriorations. A nonparametric, online change-point detection method is developed based on Parsimonious Smoothing to precisely characterize trend variations and enable accurate and timely detection of process deteriorations. Motivated by the wide adoption of multi-sensor monitoring in manufacturing, the second part of the thesis focuses on analyzing MTS. A feature extraction-fault prediction framework is developed based on Bag-of-Words models to reveal patterns related to the root causes of faults and improve the predictive power of fault recognition. The third part of the thesis considers data of higher dimensions such as images and videos, as commonly seen in many advanced manufacturing applications. A spatial-temporal model is developed to characterize the thermal behavior of melt pool in laser-based additive manufacturing (AM); model parameters are then used to monitor the evolution of melt pool thermal activities with respect to time, detecting anomalies and potential defects such as porosity. AM’s advantages in producing customized parts and computerized production are prominent, yet they pose unique challenges for data collection and cause low data quantity issue, which prohibits the development of data-driven process monitoring. This motivates the final part of this thesis to develop an age-cGAN to augment and forecast the thermal images of melt pool conditional on the layer index.
The methods proposed in this research are demonstrated and validated with simulated data and real data. The real data in this research include production data in leak test and self-pierce riveting for powertrain assembly from a leading automotive manufacturer, inline measurements from a paper manufacturing plant, and lab experimental data of laser-based additive manufacturing. The proposed methods have proved to achieve robust monitoring, diagnosis, and prognosis in real-time, showing huge potentials of generalizability to various other manufacturing applications.
NotePh.D.
NoteIncludes bibliographical references
Genretheses, ETD doctoral
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Statistical Profile