444903 Harnessing Big Data for Smart Manufacturing: A Novel Optimization Framework  

Monday, April 11, 2016: 1:30 PM
335A (Hilton Americas - Houston)
Thomas F. Edgar, McKetta Department of Chemical Engineering, The University of Texas at Austin, Austin, TX and Christodoulos A. Floudas, Department of Chemical Engineering, Texas A&M University, College Station, TX

Manufacturing planning and management practices traditionally feature vertically optimized operational units, where enterprise segments are optimized in a vacuum, planning and scheduling rely on conservative forecasting, and reactive decision making is slow and hindered by incomplete information and inherently static infrastructure.  Advanced smart manufacturing frameworks are characterized by enterprises which are fully-integrated and able to harness and optimize vital operational, business, and supply chain information in real time.  Current and developing capabilities in the rapid and decentralized acquisition of data both enables and complicates the development of smart manufacturing methods. Existing and new methods that are to be applied to Big Data for smart manufacturing must be able to accept and prioritize highly variable data sources, including factory sensors, historical and real-time information, images, market and demand data, and customer-side product performance collected via the cloud.  Applicable methods must also be efficient to draw rapid conclusions such as for fault detection and root cause analysis while still capturing the highly nonlinear and interconnected nature of the data inputs.

In this plenary talk, we will present and discuss  optimization- and statistics-based frameworks for data-driven decision making which can identify the key components of an input high-dimensional dataset.  The methodology can be applied to a variety of situations under the domain of smart manufacturing by modularly incorporating machine learning approaches, including those for nonlinear regression, classification, outlier detection, or clustering.  Optimization models were formulated which cast the high-dimensional feature selection problem into implicit nonlinear feature space, resulting in minimum-maximum mixed-integer nonlinear optimization problems..  Extensive computational studies over numerous benchmark datasets show that new algorithms outperform existing state-of-the-art methods in the machine learning literature and show great promise for accurate, real-time decision making from Big Data in smart manufacturing operations.

Extended Abstract: File Not Uploaded
See more of this Session: Big Data Analytics Plenary II
See more of this Group/Topical: Topical A: 2nd Big Data Analytics