Page 58 - Undergraduate Research Opportunities Program Proceedings 2017-18

School of Engineering
Department of Civil & Environmental Engineering

Geysers are explosive eruptions of air-water mixtures from manholes or drop shaft of underground drainage system.
During rainstorm, the capacity of drainage systems may be exceeded with a vast amount of storm water inflow. With
rapid continuous intake of rainwater and large fluctuation on water level, air can be easily trapped in the enclosed
drainage system and move along the pipe until releasing through another manhole. Geysers are therefore formed due
to the rapid release of air, pushing water up to the surface drastically to bring perilous bursting effects. This physical
phenomenon of air-water interaction is the main topic for discussion, observation and analysis. To simulate this
situation, an idealized urban drainage system is scaled down to a simple model made of valves, horizontal pipes, vertical
risers and pressure transducer. The formation of geyser is observed by filling the system originally with water, and
release an air pocket into the horizontal pipe. The trajectories of the air pockets in horizontal pipe and vertical riser are
measured by a high speed camera, while pressure is measured at four specific locations of the vertical risers by pressure
transducers.

A Big Data Landslide Early Warning System with Apache Spark and Scala
Supervisor: WANG Yu-Hsing / CIVL
Student: CHANG Bing An / MATH-CS Course: UROP2100, Fall
WONG Wen Yan/ COMP UROP2100, Fall

This report attempts to improve existing monitoring system and enable predictive maintenance by classifying unlabelled
audio data. To achieve this, we propose an unconventional approach which involves developing autoencoders and
applying transfer learning. This report provides a description of the transfer learning architecture that was chosen, as
well as the implementation and proposed evaluation methods. Furthermore, this report compares the performances
and bottlenecks of various open-source data bases based on research done by other groups in the past. The volume and
structure of data involved, as well as the querying algorithms used by the research to evaluate database performance, is
in accordance with the Social Network Benchmark developed by the Linked Data Benchmark Council. Therefore, the
results are reproducible and consistent, so we would use them directly in our analysis. But first, we want to reproduce
their results to ensure correctness. Besides that, this report also aims to provide suggestions for extending existing
querying languages to perform complex queries quickly, easily and intuitively. Finally, we discuss lessons learned along
the way and potential future developments.

Supervisor: WANG Yu-Hsing / CIVL
Student: LIU Chengzhong / RMBI Course: UROP1000, Summer

This report introduces two ways in setting up local Spark with Zeppelin and Jupyter notebook on two systems: Windows
and Ubuntu. Apart from this, the report also examines failed attempts in setting up local Spark to provide some insights
for the future students in this UROP topic. A mini project in implementing a machine learning model with Spark MLlib
will be introduced in accordance with Jupyter notebook on Windows. I do hope this project can be helpful to future
students in this UROP project. The project is available on GitHub at here. You can find more detailed instructions and
more references in this repository, in case I missed out any importance. Also, due to the limitation inSpace and format,
my report is only are presentation of my work, and more information is store in this GitHub repository.

Supervisor: WANG Yu-Hsing / CIVL
Student: SUN Ji / CIVL Course: UROP2100, Fall

We have entered an age of images and videos because of the explosion of Internet which is a carrier of data as well as
sensors. In order to process and analyze these data intelligently, technicians try to teach the machines to see things just

48