Date(s) - 17 Sep 2018
1:10 PM - 2:00 PM
3043 ECpE Building Addition
Speaker: Adarsh Krishnamurthy, Assistant Professor of Mechanical Engineering at Iowa State University
Title: Geometric Algorithms for Integrated Design and Analysis
Abstract: Computational models are beginning to play a significant role in design and manufacturing as well as in engineering analysis applications in providing designers with tools that aid in their decision making process. Many of these computational models require better geometric algorithms that are computationally efficient and interactive for widespread adoption. The talk will focus on some of the main application areas of geometric algorithms that would integrate design and engineering analysis. Complex computer-aided design (CAD) operations were not previously interactive due to the compute-intensive nature of the geometric computations performed by the CAD system. We have developed geometric algorithms that are accelerated using graphics processing units (GPUs). The first part of the talk will touch upon common strategies to develop new parallel GPU algorithms; illustrating those using examples of CAD operations such as surface-surface intersection and interactive multi-level voxelization. These new GPU-based geometric algorithms achieve tremendous performance gains (>40x) over existing CPU implementations. The second part of the talk will focus on geometric algorithms that are required to build an artificial intelligence enhanced CAD system. Deep 3D convolutional neural networks (3D-CNN) are traditionally used for object recognition, video data analytics, etc. We present an application of 3D-CNNs in understanding difficult-to-manufacture features from models to develop a decision support tool for cyber-enabled manufacturing. We have developed a framework using Deep 3D-CNNs to learn salient features from a CAD model of a mechanical part and determine if the part can be manufactured or not. Using our framework, an interactive decision-support system for DFM can be integrated with current CAD systems. Finally, 3D-CNNs have the disadvantage of high-memory usage while training high resolution 3D voxelized models. We have overcome this challenge by developing a multi-resolution convolutional neural network, which is trained on multi-level voxelized 3D models. This network reduces the training memory requirements on the GPU by as much as 15 times, while achieving similar accuracy as dense voxelization for object recognition.
Bio: Adarsh Krishnamurthy is an assistant professor in the mechanical engineering department at Iowa State University, where he currently leads the Integrated Design and Engineering Analysis (IDEA) lab. Prior to this, he was a post-doctoral researcher in the bioengineering department at UC San Diego. He received his Ph.D. in mechanical engineering from UC Berkeley and his Bachelors and Masters from Indian Institute of Technology, Madras. His research interests include computer-aided design (CAD), GPU and parallel algorithms, biomechanics, patient-specific heart modeling, solid mechanics, computational geometry, and ultrasonic non-destructive testing.
ECpE Seminar Host: Ashfaq Khokhar