ESE Senior Design,
Ken Laker, Vukan Vuchic, Phil Farnum
Authors: Deger Atay, Gizem Goryakin, Peter Eschenbrenner, Edward Levinson
Advisor: Scott and Sobkiw (L-3 Com.)
Until the past 15 years, Ghana was plagued with corrupt and militant governments. As a result, the nation suffered economically. Recently, Ghana has begun to rebuild its infrastructure in the hope of becoming a more stable and peaceful nation. Although Ghana has many natural resources and an abundant labor force, much of its nation remains in poverty.
In completing this project we plan to provide the means and resources to achieve economic stability in the Volta Region. In order to do so, we developed a realistic model of a village in Ghana. This model village is able to show the means for a given region in Ghana to prosper, including access to energy/electricity, clean water, food production, a health care center, security, education centers, effective waste management, housing, and transportation. This model has been constructed and verified using Agent-based Modeling. From this, it is possible to demonstrate both positive and sustainable influences that our village model will have in Volta region.
According to the United Nations agency, “over the next 25 years, over 2
billion people in the world will add to the growing demand for housing, water
supply, sanitation, and other urban infrastructure services. The ultimate
goal of our project is to implement actual versions of our model throughout the
world to help meet this demand.
Authors: Sean O'Hara, Brandon Park, Aaron Jacobson
Modern LED walls are used in a variety of applications, from store-front displays to large billboards, in order to present high quality images. Tri-color LEDs are used in such applications because they are more efficient than traditional incandescent bulbs, are able to display a full spectrum of color, and have long useful lives. While traditional LED products provide high quality images, these images must be programmed statically and the LED walls are not able to respond to user inputs. If a traditional LED display is combined with an array of sensors, an interactive LED wall can be created. By using sound, touch, and motion sensors to respond to the user, an LED wall can create dynamic images that change based on the user’s activities. Such systems are able to reach their end user in a more profound way, and display information that is more stimulating to the end user. The iLED system combines high-quality visualizations with interactivity while being low cost, low power, and extremely modular, allowing the system to be ported to a variety of applications. The primary application for this project would be in learning applications for museums or children's hospitals.
View Poster Movie
Authors: Zachary Zwarenstein, Xiao Ling, John Gillette, Ravi Patna
In recent years, increasingly more focus has been directed towards renewable energy production for the sake of the environment, as well as political and economic stability. Consequently, the goal with this project was to develop a sustainable and environmentally compatible solution to a pressing issue on Penn’s campus. A brief survey of Penn’s facilities led us to the pond in King's Court English House, also known as the Biosphere Pond, which is overcome with algae bloom. The project team sought to develop and implement a solar powered water maintenance system to keep the Biosphere Pond clean.
The system’s design is based on the pond’s size as well as historical weather patterns, which were used to simulate power generation for the Biosphere Pond. A prototype has been developed which circulates water throughout the pond and also pumps water through a UV sterilizer, which kills off unsightly bacteria. In addition, the system contains a single-axis sun tracker to improve energy conversion as well as a temperature sensor to prevent operation during freezing temperatures and to maintain system integrity. All of the system’s processes are managed by algorithms installed on a microcontroller.
Results for the Biosphere Pond include a successful implementation of a prototype which is able to keep the pond clean on a continuous basis during the spring, summer, and fall seasons. Ultimately, the design process proved that the Biosphere Pond concept is viable and also revealed that the design has the potential for alternative applications including larger pools, fountains, and perhaps even drinking water.
Authors: Raman Gupta, Avinash Rajput, Cullen Talbot
One of the most important characteristics of an effective factory is reliable performance. Currently, factory automation systems employ centralized control systems that are prone to catastrophic system failure as a result of having single points of failure. This means that the failure of one device in the system could potentially cause the entire system to fail. Distributed control systems increase the reliability of a factory control system by reducing the number of single points of failure that can cause factory operations to stop. This project demonstrates the promise of distributed control systems via a proof-of-concept model. The distributed control system is created using wireless communications, as distributed control systems are prohibitively expensive to implement using wired communications systems.
This project resulted in a wireless system infrastructure that is capable of supporting a distributed control system. This wireless system has been interfaced with a factory model using the requisite hardware interfaces. Finally, the distributed control system exemplifies the design goal of reliability. This project focuses on reliability in the context of control node failure, which is an unavoidable condition of all control systems. Reliability was demonstrated by maintaining factory operations in the event of a benchmark percentage (25%) of control node failures.
View Poster Movie
Authors: Divya Krishnan, Danish Munir (CSE)
Advisor: Wyner (Stat. Dept)
Due to the rapid maturity of trading engines and the level sophistication of
communications technology, high-frequency trading represents the ability to
place trades that are only feasible by automated computer-based systems on
high-speed electronic networks. The profitability of such trading strategies
arises in the intra-day spikes of stock values. The trading system created in
this project analyzed the profitability of trading frequently throughout the day
and comparing the results with overall market performance. With a user-defined
minimum time interval, the algorithm monitors the movement of prices of the
stocks; with each price movement, it is determined if the assets diverge from
its equal-weighted position, and profitable trades are identified using the next
price. Given a user-defined portfolio and threshold levels for stop-loss, it was
shown that the returns of a long-short strategy of two portfolios—one
high-frequency traded portfolio and the second a low-frequency traded
portfolio—outperform the overall market. Furthermore, this strategy takes
advantage of the scaling of portfolio statistics. It was shown that correlation
between assets decrease to close to 0 as the minimum time interval between
trades decreases. Thus, the returns of this strategy, or the mean, are as
substantial as market returns but with a lower variance, or risk level.
View Poster Movie
Authors: Tim McKenna, Anil Venkatesh, Neeraj Wahi, Nicholas Annetta
Advisor: Van der Spiegel
A single camera can be a useful surveillance tool, but video captured from a single point of reference becomes ineffective when objects of interest are blocked by occlusions. Furthermore, a traditional camera’s position and focal length cannot be modified after the video is captured.
Our system circumvents these shortcomings by processing data from an array of eighteen cameras to synthesize a single virtual camera whose position and focal length can be adjusted entirely through software. The effective aperture of the virtual camera is larger than any physically realizable aperture, which provides an extremely shallow depth of field. Such a shallow depth of field allows for the reconstruction of partially occluded objects lying on an arbitrary, user-specified plane.
The system can be used in real-time or with pre-recorded video to penetrate occlusions such as foliage and crowds, making it a flexible tool for military and civilian surveillance applications where preserving a line of sight is important, even when objects are partially obscured.
Technical Analysis is the study of historical price trends in order to predict future price movements. Traders use various forms of Technical Analysis to improve their trading profitability. Japanese Candlestick Charting, one discipline of Technical Analysis, has a long history of use in Foreign Exchange markets. Traditionally, traders would study price movements in the form of candlestick charts to visually search for well-known patterns that lead to buy or sell signals. However, this method is tedious and inefficient, as traders manually comb through charts to find indicators and make trading decisions.
The model created in this project converts the visual signals perceived by
traders from candlestick charts into computer logic, effectively automating the
technique. Price data is imported for any currency pair, and the model outputs
when to buy and sell based on the automated candlestick indicators. This model
contains a master indicator, which is refined to only include candle patterns
that are historically profitable. The master indicator also takes into account
other technical indicators that are used for confirmation. After testing over
seven currency pairs and three decades of data, the model has shown an ability
to increase trading profits on average. The results are promising and merit more
research for Candlestick Charting, as well as Technical Analysis as a whole.
View Poster Movie
Authors: Thomas Macrina, Jason Halpern (MEAM), Derek Ondrusek (MEAM), Nicholas Araujo (MEAM), Matthew Sylvester (MEAM), Alexander Greer (MEAM)
Advisor: Jonathan Fiene
The PowerFlower is a portable solar generator based on concentrated photovoltaic technology. The prototype proves the feasibility of two provisionally-patented concepts: folding optical elements (referred to as petals) for portability and protection, and a tripod actuation system for solar tracking and petal deployment. The six parabolic polished aluminum petals focus sunlight onto concentrated photovoltaic solar cells, which convert the solar energy to electricity. Any energy not converted to electricity is converted to heat, which is dissipated using an active water cooling system. The device is capable of tracking the sun via an open-loop positioning algorithm that incorporates a variety of sensors, including a GPS, magnetometer, and accelerometer. The prototype is rated to produce up to 108 Watts of electric power, with an overall efficiency of over 30%
View Poster Movie
Authors: Stephen Khou, Fang (Jack) Gao, Sungmin Lee
Many students and faculties of the UPenn community trash their used items when many of these items are still in “more than desirable” conditions. The owners of these items would most likely rather sell these items to make some additional income than simply toss them away. However, because of reasons such as the lack of experience or the time commitments associated with these sales, members of the UPenn community have continued to discard quality items.
Thus, the objective is to model a potential establishment that will effectively mitigate the entire selling process by eliminating the seller’s need to research and monitor transactions.
Based on our vision of how this establishment will operate, a mathematical model was created to simulate yearly operations. Data of demand, supply, costs, and other variables associated with maintaining a business was also gathered and analyzed. These empirical data were processed by our model to determine if such a system was sustainable. The data was also altered to find conditions where the business reached its sustainability thresholds.
Should the system operate as a typical buy-warehouse-sell establishment, the system is sustainable under current real world conditions. However, this is highly contingent on the choice of items to sell. Profit is quite sensitive to fluctuations in item pricing and quantities. Employee costs are the primary costs associated with long term sustainability. Item cost in actuality is greater; however, it is unique in that it is not an attribute we can actively reduce as it is tied directly tied with revenue.
View Poster Movie
Authors: Dan Lindholm, Kyle Wang, Jennie Xue, Aaron Jungstein
There is great interest in the development of electric power generation using renewable sources of energy given the monetary and environmental costs associated with fossil fuels. Hot Dry Rock (HDR) technology generates electricity by pumping high-pressure water deep into the earth which is then extracted at very hot temperatures and converted into electricity. HDR technology provides a novel approach to geothermal energy production because, rather than using a pre-existing underground reservoir, the high-pressure water that is injected into the earth is able to fracture rocks and create new reservoirs. This technology provides promise in meeting the growing energy needs of the world while having zero emissions and drawing on a nearly ubiquitous resource.
The team worked with a group of investors and experts in the field to model and assess the technical and financial feasibility of a five megawatt, $35 million HDR power plant. The plant will be installed in Fenton Hill, New Mexico and will utilize three wells, each at over 12,000 feet deep.
After modeling the technical factors of the plant, the team built a complex financial model and performed Monte Carlo simulations to compute feasibility using repeated random variable sampling. Two feasibility measures are outputted, including internal rate of return and net present value of both the cash flows and the net income. The results indicate that a commercial HDR plant at Fenton Hill may be technically feasible, but the financial returns may not be attractive, given today’s environment.
Authors: Mike Rovito, Nick Perkins, Jason Bowlsby, Shana Hoffman
Advisor: Keenan (Scott and Sobkiw)
The United States’ electrical energy sector faces a set of challenges that, if they go unaddressed, could undermine national security and destabilize the Earth’s ecosystem. There is a clear need for a national energy system that is independent of foreign inputs and sustainable in nature.
Many questions, such as when and where electrical energy is needed and how the resources that fuel its generation should be harnessed, underlie the development of a national electrical energy system. Furthermore, the answers to these questions are mutually dependent and highly interrelated.
EESOM utilizes linear optimization to assess whether an independent and sustainable energy system is achievable and determines what the lowest-cost system would look like. The model ensures that electrical energy demand does not exceed supply and that resources-used do not exceed resources available while minimizing the total cost of the system. Issues addressed include: timing of demand and supply, location of natural resources, and energy transportation costs. Finally, the requirements and capacity of various power generation technologies have been assessed and included.
EESOM’s output is the lowest-cost mix of power plants, including their
general location, necessary to meet demand given the resources available. The
model is capable of being run under a variety of scenarios, including carbon
caps, enabling its use as a policy analysis and investment assessment tool. The
most relevant finding is that the domestically available natural sustainable
resources – sun, wind, and subterranean heat – are sufficient to meet double the
current United States’ energy demand.
Authors: Thomas Lumpkin, Jack Chen, Mathew Gatto
The equity markets offer an enormous opportunity for the creation of wealth. Two main schools of thought exist with regards to investment decisions: fundamental analysis and technical analysis. The former bases investment decisions on variables that will affect a firm’s value, such as operations and economic trends. The latter believes that future price movements can be predicted using past price movement data.
By combining these two methods, this project attempts to utilize technical analysis to predict changes in the fundamental value drivers of companies. The model attempts to capitalize on market inefficiencies to capture statistically significant, risk-adjusted excess returns.
The model utilizes momentum focused financial indicators to predict the movements of commodity prices. Because commodities are a key value driver for companies that are producers of commodities, the price movements change the fundamental profitability outlook for such companies. Once a prediction of commodity price movements is attained, financial indicators are used to determine the predicted price movements of equities with exposure to those commodities. After the most attractive equities have been chosen, a pairs trading strategy is implemented to create a portfolio with zero systematic risk.
The model produced a return of 18.44% over a nearly six-year period. This is
equivalent to a 2.9% annually compounded return. Over the same period of time,
the S&P 500 returned -10.2%, a -1.8% annually compounded return. The Beta of
the strategy, which is the measure of the systematic risk, was -0.2375. The 95%
confidence interval of this estimate is bounded by -0.37 and -0.01797, which is
slightly statistically different than zero.
View Poster Movie
Authors: David Wang, Cem Dedeaga, Titus Chew
With the recent turmoil in the global capital markets, risk management has once again become a relevant topic. Dictated substantially by modern portfolio theory, many assets relevant to the public are characterized as being fully described by an expected return-risk pair based on historic observations, where return is calculated through an average and risk through standard deviation.
Mean variance optimization is arguably one of the building blocks of portfolio management, suggesting that collections of assets which exhibit stationary return processes can be assembled into portfolios based on some optimization of given risk-return trade-off parameters.
We approach the problem of portfolio optimization first by performing the classical mean-variance optimization via a convex optimization package. We then build upon this idea with techniques including resampling, shrinkage, and time series approaches to the modeling of returns and covariance. We demonstrate sample returns based on these strategies, and find that in all applicable cases, the relative outcomes are largely in-line with existing literature.
This project culminates in a MATLAB-based software package with a graphical
user interface targeted towards a moderately sophisticated user. The tool allows
for the input of historic data, a personal utility function and variety of model
parameter, executes a range of optimization programs, and provides an overview
of the user’s investment possibilities and their associated risk-reward
View Poster Movie
Authors: Daniel Hillman, Ethan Keller, Emily Mc Grath, Thalia Shamash
There is a growing concern in the medical and environmental communities regarding the current disposal methods for pharmaceutical waste. The practice of discarding pills and intravenous solutions in sink drains of hospitals compromises the safety of our water and environment.
RxPunge has worked closely with the inpatient pharmacy at the Hospital of the University of Pennsylvania (HUP) to study the IV waste system in place and to assist the pharmacy in pioneering a new system to properly classify, manage, and dispose of these dangerous chemicals. RxPunge recorded quantities and types of expired IV drug solutions returned to the pharmacy to assess the current system in place. Over 100 drugs were classified and evaluated for hazard, reactivity, biodegradation, and potential disposal methods. Statistical distributions were identified to simulate the drug disposal, and these were used to develop a stochastic simulation to predict the daily mix of pharmaceutical waste.
This model, combined with implementable recommendations, depicts a method to better manage pharmacy operations and reduce hazardous waste. Inventory practices, drug-preparation times and distribution frequencies can be modified to reduce drug waste from the hospital and cut costs.
Waste disposal and treatment methods were analyzed, thereby providing
actionable options with which HUP create positive environmental impact.
View Poster Movie
Authors: Sam Russem, Cameron Finucane, Helen Kim
Advisor: Mackenzie White and Berkay Deniz Ilhan
EduBot is a six-legged, bio-inspired robot that specializes in dynamic locomotion over varied terrains. While EduBot is physically capable of running through a variety of environments, the robot’s lack of sensory input has limited it to operator-controlled movement only.
In order to decouple EduBot from its human operator, additional sensors are necessary provide the robot with more information about its environment. With this new data, localization and high-level navigation algorithms may be implemented to achieve semi-autonomous locomotion.
In this system, two complementary sensors are integrated with EduBot to establish relative position accurate to the meter. A Global Positioning System (GPS) receiver is used to determine the robot’s absolute position, but this reading is only available once per second and is only accurate to a radius of 10 meters. To supplement the GPS, an inertial measurement unit (IMU) is installed as well to provide acceleration and angular velocity data at a high frequency. A Kalman filter is used to maintain a continuous estimate of the robot’s orientation and position based on the data received from the GPS and IMU.
The position and heading estimates produced by the Kalman filter are demonstrated to be significantly more accurate than the raw GPS or IMU readings alone. Using these improved estimates and a unicycle steering waypoint-based navigation algorithm, EduBot is able to successfully navigate between user-specified locations.
View Poster Movie
Authors: Dan Garzarella, Jake Quain, Tim Potens, Kyle Andrews
The SEPTA Regional Rail system serves as an important network for the Philadelphia region, moving many commuters during the peak hours on suburb-to-city or city-to-suburb trips. Although this service fulfills commuter needs well, the opportunity exists for SEPTA to improve service during the entire day and on non-commuter trips.
This project developed a new model for scheduling Regional Rail service with a focus on improving the ability to transfer among lines to encourage suburb-to-suburb trips. The model targeted operational factors, increasing the frequency of service or reducing the scheduled wait time between two given lines, while bypassing the need for expensive capital improvements.
The optimization model minimized the average wait time for passengers traveling between lines according to probabilistic weightings based on the priority of a given transfer. These scheduled transfer times were subject to physical and policy constraints that limited the range of times when a given train could travel. The alternative schedules produced by the optimizations were evaluated upon their cost-performance package.
Authors: Andrew Avrin, Danny Lustig, Brandon Duick
Currently, drivers must utilize a third-party, such as a radio or broadband device, to learn about local traffic conditions. However, this information is often out of date by the time it reaches a driver, and the area covered by such services is often limited. The fastest and most efficient way to transmit information about road conditions to drivers would be to create vehicle-to-vehicle wireless networks. This way, cars can freely share information with each other in real-time, allowing drivers to be more aware of the current conditions.
This project demonstrates the capabilities and potential impact of vehicle-to-vehicle networks. It uses modified versions of existing WiFi technology and the emerging Wireless Access in the Vehicular Environment (W.A.V.E.) family of protocols. The system detects specific events from the existing computer systems in a Toyota Prius and communicates this information wirelessly to nearby drivers. A GPS receiver is also used to provide accurate location and timing synchronization to within 20 μs. Information received from other vehicles is selectively displayed to the user, and in emergency settings a tone is generated to ensure that the driver can react quickly enough to avoid a dangerous situation.
Location Division Multiple Access (LDMA) was implemented to allow for multiple vehicles to transmit without message collision. We were able to successfully demonstrate the generation, retransmission, and reception of event information, triggered from both simulated and real events in the vehicle.
Authors: Oscar Nunez, Pedro Maia (CSE)
Advisor: CJ Taylor
One of the major obstacles to the development of small scale Unmanned Aerial Vehicles (UAV) that could be deployed in indoor environments is the collision avoidance problem. This avoidance problem lies on locating obstacles through real time estimation of either depth or time to collision (TTC). This problem is complicated by the small payload and power available on small UAV platforms.
There are currently two trends for solving the collision avoidance problem, the first is to use Laser Range Finders to map the environment by extracting depth information; the second approach is to use video cameras and compute TTC from Optical Flow. Both approaches have been proven to work; however, they require a powerful processor or power, size, and weights unsuitable for small UAVs.
This project seeks to develop simple, robust vision based obstacle avoidance schemes that are amenable to implementation on low power, low profile hardware. Spatial-temporal derivatives are used to compute TTC, which is used to classify the state of the UAV as whether it is in collision danger or not; the relationship between the derivatives and TTC was gotten by approximation of rectilinear trajectories and by making use of the brightness constancy equation.
The correctness of the proposed approach was proved through unit tests and
ROC curve performance analysis. The true positive detection rate of the system
is currently above 90%. The functionality of the system implementation will be
demonstrated through closed loop simulations.
View Poster Movie
Authors: Ayesha Hussain, Yinwei Liang, Yexiang Tan, Shutong Zhang
Advisors: Keenan, Carchidi,
The Perioperative Services Unit at the Hospital of the University of Pennsylvania is responsible for processing instruments to be used in surgeries. Currently, the fixed staffing schedule at the instrument-processing unit is inefficient, resulting in times of understaffing and over-capacity at other times.
The goal of this project is to design a Quality Staffing Model that will optimize the staffing needs of the instrument processing unit and thus reduce overall system costs. In order to optimize the staffing needs of the department, the inflow of dirty instruments into the department is determined through analyzing historical trends. The processing times are determined through empirical data gathering and a work sampling study. These inputs are utilized in a simulation model that uses both analytical methods based on queuing theory and Monte Carlo techniques to determine optimal labor deployment. Sensitivity analysis on factors such as employee productivity and instrument arrival volatility is conducted to determine overall system cost savings. Finally, an Excel scheduler program is created that will help HUP staff employees based on forecasted labor demand.
The Excel scheduler program is comprised of a base staffing scenario that
minimizes costs while fulfilling employee work-hour constraints using historical
arrival and processing rates. In addition, the program has a Flex component that
allows for an increase or decrease in employee staffing based on the forecasted
arrival of instruments twenty four hours in advance. Results indicate that the
overall instrument inventory level is highly sensitive to both instrument
arrival and processing speed. Thus, significant savings can be achieved through
increasing the processing rate by reducing the occurrence of interruptions.
Authors: Ray Cheng, Emerson Barth
For hundreds of years, one of the major trends of land use has been the deforestation of temperate regions. In countries like Costa Rica, landowners have acres of pasture that are devoid of both grazing and growth. The land held by private owners could have a major impact on the surrounding ecosystem, but local inhabitants do not have the information or resources to initiate the reforestation process.
The approach this system takes to bridge this gap is to use a series of algorithm, based on historical and real-time data such as tree growth factors and weaher parameters to determine what trees can be grown. An economic analysis based on carbon sequestration and timber prices will follow to determine what trees are most viable for reforestation efforts.
The microreforestation aspect comes from the land metrics specific to each area, which includes land size, light intensity, and altitude. Our optimization scheme delivers an interface that is accessible for users all over the world. The model also aim at addressing the misconception that reforestation is a negative investment.
With the system implemented, any individual in Costa Rica or Columbia can
understand what types of trees they can plant, how to plant them and the
positive impact and potential monetary returns associated with such plantation
effort just by inputting their location and land size.
View Poster Movie
Authors: Mohammad Sadiq, John Choi, Alex Wei-Chun Chueh
Modeling the equity system using technical analysis or other mathematical methods to achieve high returns from the market has become prominent in today’s finance. Analysts have tried to simplify the complex market dynamics by using indicators to predict the trend of a market or a stock.
However, with increasing global markets interaction, simple indicators oftentimes fail to signal the trend changes timely to prevent losses. Moreover, since many models have only a limited number of factors incorporated into their systems, they require re-modeling as new factors are found.
The model of this senior design embodies a basic feed-forward, error-back-propagated neural network structure, with modifications to neuron connection weight matrix formation. The inputs for the main system are percentage changes in price data and volume data, and the outputs form signals for trading decisions. Weight matrix is a collection of bipolar, converging neural sub-network, using inputs related to the main network inputs, namely related stocks. With the main network holding two layers of neuron, each matrix node represents interactions between corresponding main system inputs and related factors by volatility or trending average means. Separation of calculations ensures additional factors can be added later. With training over an extensive time period, the system displays convergence and prediction capabilities.
With the above model used, an overall trading software is implemented with an
automatic data update system and a user interface. The resulting system displays
performance superior to returns achieved from commonly used technical
View Poster Movie
ESE Home Page > ESE Undergraduate Labs > Senior Design Home Page >
©University of Pennsylvania, Department of Electrical and Systems Engineering.
This site maintained and administered by Siddharth Deliwala, (deliwala_at_ee.upenn.edu ), Last updated Apr 19, 2009