Our research approach is problem motivated -- that is, given a relevant, interesting, real-world research problem in manufacturing and engineering design domain, we develop computational tools for the related problem. During the course of research we also build physical devices and develop user interfaces for validation of the developed computational tools. By combining engineering innovations with methods from machine learning, computational geometry, statistics and optimization, metamodeling and sampling, ontology, and uncertainty quantification, we strive to solve important research problems in manufacturing and engineering design domain.
Current Funded Projects
Physics LEArning (PLEA): A Hybrid Physics Guided Machine Learning Approach for Predictive Modeling of Complex Systems.
In this research project, we propose to address the lack of success of purely data-driven constructs in predictive modeling of complex systems in the presence of sparse and noisy data beyond their initial set of training data. This research will lead to innovative hybrid methods, especially in studying physically-grounded systems, that integrates systematic understanding about the systems in the form of physics of the systems with data-driven machine learning approaches.
COVIA: Computer Vision base Intelligent Assistant for Mistake Proofing of Complex Maintenance Tasks on Navy Ships
The main goal of the proposed project is to investigate advanced deep learning based computer vision methods and algorithms to enable next generation Handheld Augmented Reality (HAR) based complex maintenance tasks.
Coordinated Holistic Alignment of Manufacturing Processes
Small- and medium-sized organizations spend an exorbitant amount of time and money stitching together disparate data from stove-piped legacy systems in an effort to match the value of insight provided by these dedicated systems. The objective of this research project is development of a low cost solution that will enable manufacturing organizations to overcome the issues associated with data heterogeneity present within and across various platforms, in an effort to provide an alternative to costly enterprise systems. The solution provides semantic alignment of the organization’s data, leveraging a tiered and systematic ontology, to effectively characterize the process and data associated with the product development process (PDP).
Uncertainty Propagation Methods for Networked Complex Systems
The objective of this research is development of a novel class of uncertainty quantification method for networked complex systems. The main challenge that lies at the core of analyzing and synthesizing the dynamic networks at the crux of modern day complex systems is: How do a collection of dynamical systems coupled through a dense wiring topology behave as a unit in the presence of uncertainty? We are developing a suite of novel computational uncertainty quantification methods to tackle this challenge.
IDEA: Intelligent Decision Enabled Application for additive manufacturing
The term 'rapid' in rapid manufacturing is quite misleading as the manufacturing time for even small-medium sized parts using additive manufacturing processes is found in the range of 6 hrs-12 hrs. Even, the algorithms to produce the optimal results are also computationally expensive. IDEA aims to develop an integrated data mining and optimization based routine focusing on (1) minimizing the energy consumption and material waste of the process, and (2) developing a computationally efficient technique to determine the process parameters using manufacturing by analogy.
Knowledge Representation and Design for Managing Product Obsolescence
The research objective of this project is to investigate two novel research approaches to understanding and managing technology obsolescence challenges: (1) a knowledge representation scheme and management system that can facilitate information sharing and collaboration for obsolescence management and mitigation efforts between existing tools and across different organizations, and (2) fundamental principles, teachable methods, and guidelines for designing product architectures that can evolve with changing requirements, enabling proactive obsolescence management across the entire product life cycle.
Distributed Computational Design Environment (DiCoDE)
Communication of a design concept is difficult among a multidisciplinary team, especially if the team is distributed geographically. Conceptual design in distributed environment is collaborative in nature. This research explores issues related to conceptual design in a distributed environment within a computational framework. The focus is on investigating the use tablet PCs for conceptual design communication in geographically dispersed team setting and classroom setting.
Wastewater aeration process optimization
Large scale process industries such as wastewater are typically designed with due consideration of environmental regulations. Often such designs are based on intuitions and leave a large scope for optimization, specifically at the process energy consumption level. The aim of this project was to perform data driven study of Buffalo Sewer Authority and develop energy minimization schemes while maintaining the acceptable water quality. Data driven models capturing the non-linear behavior of the aeration process is extracted and optimized with an aim to minimize the process energy consumption and material waste. Different energy savings scenarios are analyzed.
Conceptual design is perhaps the most crucial task of engineering design process. The automated generation of conceptual designs is a non-trivial task. This project explored the use of graph grammar as a concept generator for complex systems. In particular, graph grammar rules were developed for generating conceptual designs of NASA ADAPTS electrical power system and integrated with modeling packages like Modellica to simulate the performance of each design.
Given a CAD file of a part, graph grammar algorithms were used for automated process plannning, machine selection, and tool selection.