If you want to ride an electric Citibike there is a small surcharge, even if you are a Citibike member. However, there is a catch! If you take an electric bike out from a station that only has electric bikes parked at it, the fee is waived. Freeebike is a web app that shows all of the free electric bikes in the city, and if you enter an address in Brooklyn will make predictions on the likelihood of nearby free electric bikes remaining free at different points in the future. The app relies on a logistic regression model which was trained on data scraped from the Citibike API at 10 minute intervals and historic weather data. It has an area under the curve (AUC) and accuracy of 0.86 and 0.96 for predictions 10 minutes out and those drops to 0.72 and 0.86 for predictions an hour out. I used a logistic regression model because it performed as well as a random forest model but was better calibrated and is easier to interpret.
Inspired by welded frames from MoMA, I decided to try to make my own. They were quite simple, as they are four pieces of steel angle cut to the desired lengths and welded along the 45 degree angle. I then used an angle grinder to get the finish and treated them with a rust proof coating. The support pieces that hold the images are laser cut acrylic and held in place by pressfit magnets. The photos, which were taken by Eric Baylen, are images of ‘cats’.
Automated Alignment Stage
The automated alignment stage was designed to control the angle between a probe (flat punch) and an adhesive surface for probe tack tests. The stage has two orthogonal motors allowing it full versatility of the angle space. It utilizes a Nelder–Mead (downhill simplex) method to find the alignment where the force required to separate the adhesive and probe is maximized. It is run by an arduino and interfaces with a ZwickRoell tensile tester and controlled through software written in Python. The automated alignment stage was designed and built in collaboration with the excellent Carleton students: Lydia Fick, Ben Hafner, and Henry Sotrel.
Support Vector Regression to Understand Nanoscale Topology and Adhesion
We used machine learning to predict the adhesion of surfaces with nanoscale topological features. By inputting the roughness of the surfaces, found via atomic force microscope, as the input features in a support vector regression model we were able to derive data-driven traction-separation relations, which in previous studies had pre-assumed forms. From the traction-separation relations we could quantify the adhesion. Over 200,000 data sets were analyzed, and we were able to predict the adhesion with an R2 of 0.98. The results were validated via finite element analysis. This work was done in collaboration with Nava Raj Khatri, Xin Ji, and Yijie Jiang.
During the pandemic I worked to make my classes accessible to students in various time zones, with different health concerns, and unique learning styles. Part of this effort was producing a full set of video lectures for Electricity & Magnetism, Thermodynamics & Statistical Mechanics, and Introductory Physics. These videos were made to supplement and/or replace textbook reading and allowed the course to be taught via a flipped classroom model.
Knitting
I knit a lot and have for a long time. It is something I can do while doing other things to keep my hands busy. This is a detail from a sweater I made for my nephew who has a black cat as a pet.
Tracking Bubbles
Bubbles form in pressure sensitive adhesives (tapes) when they are loaded in shear. The shape and evolution of these bubbles depends on the crosslinking ratio of the polymers (which can be thought of as how soft and sticky the tape is). If the load is applied and then removed the material is allowed to relax and the shape of the bubbles morph over time. If there is a high crosslinking ratio teardrop shaped bubbles form and just get larger over time, while for lower crosslinking ratios initially the bubbles are elongated, but over time they evolve into thermometer shapes. I discovered this phenomena while testing shear failure mechanisms, but became interested in it and wrote MATLAB code to track the size and perimeter of the bubbles over time. This work was done under the supervision of Matteo Ciccotti and Costantino Creton.
Shear Failure of Pressure Sensitive Adhesives
Pressure Sensitive Adhesives (colloquially referred to as tapes) perform incredibly well under shear loading, meaning that they can stick for a long time. However when failure finally does occur it can be catastrophic. The long wait time followed by the very quick failure makes understanding the mechanisms that cause the failure hard to predict and understand. I built an experimental set-up with a fast camera and a triggering system to capture the final failure and classified different types of failure depending on the amount of overlap for the adhesive and the loading. I also studied the strain field in the adhesive backing through digital image correlation. We also produced analytical models that capture the behaviors seen experimentally. This work was done under the supervision of Matteo Ciccotti and Costantino Creton and in collaboration with Chung-Yuen Hui and Zezhou Liu.
Microscale Composite Posts for Tunable Adhesion and Transfer Printing
This work focused on shrinking the composite posts (to understand the mechanics behind them check out Composite Posts for Tunable Adhesion) to the microscale so that they could be used for microtransfer printing, a process used to move MEMS devices like integrated circuits (with dimensions on the order of ~100 micrometers x 100 micrometers x 10 micrometers) to new, potentially flexible or biocompatible, substrates. By reducing the dimension of the posts we saw an enhancement in adhesion strength and were able to use these posts to manipulate microscale objects. Depending on the dimensions of the inset the crack propagation pattern would vary (as seen in the images above where each row is a post with a different geometry and each column is a point in time). This work was done under the supervision of Kevin Turner.
The adhesion of high aspect ratio tapered epoxy nanopillars (300 nm in diameter at the base of the pillars and 1.1 μm in height) with different cross-sectional geometries (pencil-like, stepwise, tall cone-shaped, and short cone-shaped) were studied under normal and shear loading. While high aspect ratio fibrils are common in natural dry adhesives, they tend to collapse due to capillary forces, decreasing their effectiveness as adhesives, however tapering the structures reduces this effect. To evaluate the dry adhesion strength of these structures we used nanoindentation. The pencil-like and stepwise arrays had the highest reported adhesion strength for any synthetic gecko-like adhesives, 42 N/cm2, however at larger indentations they fractured. The short cone-shaped pillars maintained a strength of 34 N/cm2 even at depths where the pencil-like and stepwise pillars fractured. This could be explained by the more uniform stress distribution throughout the short cone-shaped nanopillars, thus preventing fracture formation observed in other nanopillar structures. Under shear loading pencil-like nanopillars exhibit the highest adhesion strength at all indentation depths and that the shear adhesion increases with greater indentation depth due to the higher bending stiffness and closer packing of the pencil-like nanopillar array. Finite element simulations were performed to verify the and elucidate observations. This work was done in collaboration with Y Cho, G Kim, Y Cho, SY Lee, Y Jiang, K Yin, KT Turner, DS Gianola, and S Yang.
By designing posts with rigid cores and compliant outer shells we were able to create structures that adhered well under normal loading (when pulled straight off of a surface) and separated easily under shear loading (when slid to the side before retracting). These tunable adhesive structures could be used for numerous applications where temporary attachment and detachment are required, for example when microtransfer printing semiconductor elements, in material handling in manufacturing, and as gripping surfaces on climbing robots. They also have the advantage of simplicity over many previous designs. The key to these structures is that the compliant outer layer allows for conformal contact between the post and the surface it is attaching to and the rigid core focuses the stress distribution to specific locations. Under normal loading all of the stress is under the rigid inset, away from the edge of the post. As the load is distributed over a large area away from the edge of the post where it is most natural for a crack to initiate, this configuration will provide enhanced adhesion. When a shear load is applied the stress field shifts to the trailing edge of the post and the load is concentrated in that smaller region, meaning a high stress at a location where separation will naturally occur, therefore it is trivial to initiate a crack there which propagates across the interface, allowing detachment to occur easily. These experiments were performed on a home built testing apparatus and modeled using finite element analysis. This work was done under the supervision of Kevin Turner.
I built this banjo in a backyard in West Philly. I’d never built an instrument before but my friend Corey Chao (a third generation banjo builder) was making one and I thought it sounded like a fun project. It was! We bent the wood for the drum using a home built steamer, we cut, rasped, and sanded the neck with tools from the West Philly Tool Library, we carefully measured frets and set inlay to make them look good.
Adhesion Testing Apparatus
To measure the pull off force and displacement while visually observing the fracture at the interface of an adhesive post and a rigid surface I design an adhesive testing apparatus. It can measure forces in one direction (z) and displacements in two dimensions (x and z). It was specifically designed to allow for visual access of an interface during probe-tack tests, which required altering a number of off the shelf components and building some parts from scratch. I collaborated with the UPenn machinist to design the physical structures and wrote the software to run it using LabVIEW.