School of Mechanical Engineering Seminar
Monday, May 15, 2018 at 14:00
Wolfson Building of Mechanical Engineering, Room 206

HIGH-FIDELITY SIMULATION OF BRITTLE FRACTURE PROBLEMS

WITH UNIVERSAL MESHES

Prof. Adrian J. Lew1

1 Department of Mechanical Engineering, Stanford University

Durand 207, Stanford, USA

e-mail: lewa@stanford.edu

We describe our approach to simulating curvilinear brittle fractures in two-dimensions based on the use of Universal Meshes. A Universal Mesh is one that can be used to mesh a class of geometries by slightly perturbing some nodes in the mesh, and hence the name universal. In this way, as the crack evolves, the Universal Mesh is always deformed so as to exactly mesh the crack surface. The advantages of such an approach are: (a) no elements are cut by the crack, (b) new meshes are automatically obtained as the crack evolves, (c) the crack faces are exactly meshed with a conforming mesh at all times, and the quality of the surface meshing guaranteed to be good, and (d) apart from duplicating degrees of freedom when the crack grows, the connectivity of the mesh and the sparsity of the associated stiffness matrix remains unaltered. In addition to the mesh, we are now able to compute stress intensity factors with any order of convergence, which gives us unprecedented accuracy in computing the crack evolution. As a result, we observe first-order convergence of the crack path as well as the tangent to the crack path in a number of different examples. In the presentation I will introduce the notion of a Universal Mesh, illustrate the progress we have made so far with some examples, and then focus on the simulation of curvilinear fractures, and on the tools we created to compute stress intensity factors. In particular, showing examples in which the computed crack path converges to the exact crack path, regardless of the mesh. If time permits, simulation of thermally induced fracture and hydraulic fractures will be discussed.

Adrian J. Lew is an Associate Professor of Mechanical Engineering at Stanford University, and the Lee Otterson Faculty Scholar. He graduated with the degree of Nuclear Engineer from the Instituto Balseiro in Argentina and received his M.Sc. and doctoral degrees in Aeronautics from the CalTech. He has been awarded Young Investigator Award by the International Association for Computational Mechanics, the ONR Young Investigator Award, the NSF Career Award, and the Ferdinand P. Beer & Russel Johnston, Jr., Outstanding New Mechanics Educator Award from the American Society of Engineering Education. He has also received an honorable mention by the Federal Communication Commission for the creation of the Virtual Braille Keyboard. He served as the North-American co-chair of the XII Pan American Congress in Applied Mechanics, which took place in January 2012, and in the organizing committee of the first Pan American Congress on Computational Mechanics, which took place in Buenos Aires in April 2015. He has also co-founded iBrailler, which produces iBrailler Notes, an app to type Braille in an iPad.

Speaker: Matan Goldman

M.Sc. student under the supervision of Prof. Tal Hassner and Prof. Shai Avidan

Wednesday, Oct 17^th 2018 at 15:30

Room 206, Wolfson Mechanical Eng. Bldg., Faculty of Engineering

Learn Stereo, Infer Mono: Siamese Networks for Unsupervised, Monocular, Depth Estimation

Abstract

The field of unsupervised monocular depth estimation has seen huge advancements in recent years. Most methods assume stereo data is available during training but usually under-utilize it and only treat it as a reference signal. We propose a novel unsupervised approach which uses both left and right images equally during training, but can still be used with a single input image input at test time, for monocular depth estimation. Our Siamese network architecture consists of two, twin networks, each learns to predict a disparity map from a single image. At test time, however, only one of these networks is used in order to infer depth. We show state-of-the-art results on the standard KITTI Eigen split benchmark as well as being the highest scoring unsupervised method on the new KITTI single view benchmark. To demonstrate the ability of our method to generalize to new data sets, we further provide results on the Make3D benchmark, which was not used during training.

(The talk will be given in English)

Speaker:     Dr. Jordan Hashemi and Dr. Anish Simhal
                   Duke University

Monday, October 15^th, 2018
15:00 - 16:00

Room 206, Wolfson Mechanical Eng. Bldg., Faculty of Engineering

Each lecture will be 30 min.

The first (15:00-15:30) will be given by Jordan Hashemi

Computer Vision and Machine Learning for Computational Psychiatry

Observational assessments are critical for screening, diagnosing, and monitoring developmental disorders. However, current tools for objectively measuring young children’s observed behaviors are expensive, time-consuming, and require extensive training and professional administration. This lack of scalable, reliable, and validated tools impacts access to evidence-based knowledge and limits our capacity to collect population-level data in non-clinical, naturalistic settings. To address this gap, we developed a mobile paradigm to collect videos of young children while they watched movies designed to elicit autism-related behaviors and then used computer vision to automatically code behavioral responses of these videos. This mobile paradigm has allowed for one of the largest video data collections for autism and the discovery of novel biomarkers. In this talk, we will go into detail about our paradigm and demonstrate its impact towards scalable, objective behavioral coding for screening and monitoring children's development.

The second (15:30-16:00) will be given by Anish Simhal

A Computational Synaptic Antibody Characterization Tool for Array Tomography

Application-specific validation of antibodies is a critical prerequisite for their successful use. Here we introduce an automated framework for characterization and screening of antibodies against synaptic molecules for high-resolution immunofluorescence array tomography (AT). The proposed Synaptic Antibody Characterization Tool (SACT) is designed to provide an automatic, robust, flexible, and efficient tool for antibody characterization at scale. SACT automatically detects puncta of immunofluorescence labeling from candidate antibodies and determines whether a punctum belongs to a synapse. The molecular composition and size of the target synapses expected to contain the antigen is determined by the user, based on biological knowledge. Operationally, the presence of a synapse is defined by the colocalization or adjacency of the candidate antibody punctum to one or more reference antibody puncta. The outputs of SACT are automatically computed measurements such as target synapse density and target specificity ratio that reflect the sensitivity and specificity of immunolabeling with a given candidate antibody. These measurements provide an objective way to characterize and compare the performance of different antibodies against the same target, and can be used to objectively select the antibodies best suited for AT and potentially for other immunolabeling applications.