M.Eng. Students Are Engaged in an Exciting Slate of Projects
From an airline management software company in Turkey, a railroad and an emergency services provider in Canada, and a not-for-profit organization in Washington to some of the most prominent companies in the world, clients are interacting this spring with teams carrying out this year's M. Eng. project slate. (Projects in Financial Engineering, which are carried out in the fall, also have a broad set of clients).
Working on team-based projects on real problems for real clients has been a defining aspect of ORIE's Master of Engineering program from the inception. Last fall, Program director Kathryn Caggiano offered students an exciting slate of projects, students were assigned to project teams based on their preferences, and work on the projects has been underway since the beginning of the year. Client organizations include Walmart.com, MITRE Corporation, Microsoft, Ornge, Hitit Computer Services, Canadian National Railway, and Xerox Corporation.
Like most retail businesses, Walmart experiences the return of some of the merchandise it sells and works to recover the dollar value of this merchandise, some of which is liquidated, some destroyed and some returned to the manufacturer. A team of six students is working on the analysis of returns to the e-commerce division of the world's largest retailer. This area of study is known as "reverse logistics." They are currently implementing a software tool that predicts the rates at which products are returned, the rates at which various forms of disposition occur, and the resulting overall recovery rate.
The team, consisting of Eddie Hsiao, Wenfeng Li, Aukrit Unahalekhaka, Aazam Vishram, Xiaole Wang and Tian Yu, worked during winter break to finish the analysis of sales data, and plan to present their results in early May. Li commented that "it has been a blast to work with some very bright people on this project." Assistant Professor Peter Frazier, the project advisor, said that "it's been great working as a team."
The law enforcement community in the United States is interested in possible uses of small, remotely piloted aircraft to provide real-time information to reduce the danger to officers. Regulations for the use of such vehicles will be issued this year. MITRE, a collection of federally funded research and development centers, is sponsoring a multi-year M.Eng. project to analyze some of the logistical issues involved in the use of these aircraft.
This year's MITRE team, consisting of Kenneth Chu, Mark Fontana, Thomas Petek and Katherine Schoenfelder, is focusing on a scenario employing a fleet of such aircraft to support ground patrol in a major US metropolitan area. While the situation is specific to one area, the team hopes its results will be applicable to local law enforcement in other metropolitan areas.
The team is working to find the optimal profile for a fleet and to identify parts of the metropolitan area where vehicles would be best deployed. The number of patrol cars that can be equipped with the aerial vehicles and the technology that is deployed from them are constrained by cost and regulation. The objective is to maximize the targets that are apprehended, as a function of the regions that are patrolled and the capability of the vehicles and sensors.
"There are many open questions about integrating the vehicles into the national airspace," said Schoenfelder. In coming up with a specific research topic, the team was "forced to think somewhat outside of the box," said team member Fontana. "I had never before appreciated the difficulty of such a feat -- defining a problem that is neither so complex as to render it intractable, nor so simple as to render it trivial," he added.
Using recent crime statistics as input data, the team has formulated a large mixed integer linear programming problem and is working on methods to solve it.
The work is being advised by Associate Professor Mark Lewis. "My teammates and I are excited to tackle a project at the cutting-edge of law enforcement," said Petek. "We hope the results will be useful to both local law enforcement agencies and policy makers," he added.
Microsoft earns more than $18 billion each year by licensing its products to companies and other organizations through what are called Enterprise Agreements. These agreements are sold primarily through partners, typically hardware manufacturers. The agreements are managed by Microsoft Licensing, which annually surveys partners to find out how satisfied they are with the services the Microsoft unit provides.
A team comprised of Abhilash Babu, Min Ki Jeon, Sejun Kim, Ginger Shao and Victor Wu has been asked to determine how well responses to the survey relate to other measures associated with the partners. These measures include such things as the amount of revenue each partner provides to Microsoft and how familiar and accurate the partners are with the procedures entailed by the relationship. The team is using correlation, regression and other statistical techniques to determine how variations in survey scores affect the overall performance of a partner, with the potential for changing the way Microsoft Licensing works with its partners. Professor Paat Rusmevichientong is advising the team.
The team has found that "there is a lot of uncertainty working with real-world data compared with the sheltered setting of an academic environment. Everything doesn't fit perfectly and come out nicely like a simulation problem set," said team member Victor Wu. "Our weekly meetings with our Microsoft contact, Mario Cortes, have been very dynamic and greatly helped us in getting up to speed."
Since 2007, ORIE has carried out a series of M.Eng. projects for a medical transport organization in the province of Ontario, Canada, called Ornge. Last year, a project team built a Flight Planning Optimization Tool to produce daily schedules of patient transfers Ornge staff in preparation for a rollout to routine use.
This year's Ornge project team is enhancing the tool so that it can be used during the scheduled day as conditions arise, such as weather, changes in requirements, and emergency calls, that render the existing schedule inadequate. Doing so requires not only handling the resulting computational issues, but understanding why existing schedules cannot be completed; understanding restrictions regarding planes, airports and staff that constrain changes to the existing schedule; and controlling disruption to the existing schedule.
"We went to Toronto to visit the client and learn first hand the issues and problems faced by them," said team member Shriram Subramanian. "We also got to visit their operations team to understand how the current tool is being used and how mid-day requests are being handled," he said, adding that "the need to come up with a new optimization tool while still accommodating all the new requirements is what makes this project exciting and challenging."
The Ornge team also includes Andrew Baldinger, Rami Jawhar, and Abinav Rameesh. It is being advised by Professors Michael Todd and David Shmoys.
One major application of operations research is found in the airline industry, which at all times has a perishable inventory of seats on flights - once the flight has departed the inventory can no longer be sold. Airline ticketing has evolved over many years to price the product on the fly, with regular price changes based on an understanding of customer behavior and close tracking of inventories (a practice that is only now coming to selling over the internet). The science of accomplishing this, based on OR methods of forecasting and optimization, is known as revenue management.
The Turkish company Hitit Computer Services offers an array of data management software to a number of airlines in several countries, but does not yet have a revenue management product. An M.Eng. project team comprised of Christopher Kim, Pourmehr Sarram, Arib Rahman, Lauren Robinson and Kevin Wald, advised by Professor Huseyin Topaloglu, is taking steps to help them develop one. In particular they are building a model that uses past reservation data to forecast customer arrival patterns and estimate reactions to different price levels. The team is working on a model that, unlike models that minimize the statistical error in the estimates, maximizes the expected economic value realized from the pricing strategy based on the estimates it produces.
Once an estimating procedure has been developed and embodied in software, the output can be fed to an optimization routine that establishes prices based on the current inventory of airline seats. The team will also develop the optimization routine.
Doing a project with a client six time zones away has extra challenges. It has not been possible for the team to travel to Turkey. However "Our contact person in Hitit has been very helpful throughout the process," said team member Kim, "and we have been able to schedule multiple meetings over skype with him."
"We have spent a lot of time trying to understand the data given to us, and formalize a process to capture the probabilities that a customer will purchase at a certain fare, so we are excited to have finalized that phase and begin to look at how to build an optimization model," said Wald.
Like Ornge, Canadian National Rail (CN) has been the client for several years. The focus of the activity has been the management of scheduling of train crews, through iCrew, which is proprietary crew management software being developed jointly by CN and Axon-HCL. Cornell M.Eng teams have been invited to influence the evolution of this development effort through a series of projects about different aspects of the overall solution.
This year's team, Ruobing Han, James Hu, Viral Malnika, Giancarlo Soto and Systems Engineering M.Eng. student Tushar Madaan, has been given access to detailed snapshots of the crew forecasting and dispatching process for the entire CN network at four hour intervals. They are using statistical analysis to study how the forecasts of demand for and supply of crew members evolve over many weeks of activity. The primary objective of the project is to recommend how far ahead iCrew can and should look in establishing dispatch plans. To do so they are using simulation techniques to test various time intervals ranging from several hours to a day or two. According to their advisor, Professor Peter Jackson, "their recommendation will have a significant impact on the architecture of the iCrew solution."
Team member James Hu commented that "this project is best described as a problem-solving process, major and minor setbacks occurred all along the way, however they are a great way to become familiar with how to work with clients to solve real industry problems. Our real discovery was the challenging nature of the data, which has been extracted from CN's crew planning system for an in-depth analysis for the first time." However the team was able to visit CN's facilities in Montreal, and found that "people at CN were very helpful and easy to interact with. They were willing to share all relevant information and were very patient in answering all our questions," according to team member Malnika.
For their part, CN Crew Management System Officers Jean-Philippe Provencher and Marcel Dupere said that "the students dedication and energy are outstanding. Their motivation to execute and achieve their objectives in terms of the project requirements is very impressive."
Xerox dispatches technicians to fix customers' printers, photocopiers, and other products. Technicians differ in their familiarity with specific models, in their base locations, and in other attributes. An M.Eng. project completed in 2010 developed a workforce planning tool to help Xerox determine how many technicians of each type should be assigned to an area in order most effectively meet demand and satisfy other constraints. This year's follow-on project starts with technicians already assigned to areas. The team, consisting of Christine Barnett, Nate Holley, Andy Sheung and Yu Zhang, is working on optimizing which technician should respond to specific clients and machines, and when. This depends on travel time, technician familiarity with client and machine, overtime considerations, and the like.
The team has formulated an integer program that will provide the basis for an algorithm to be employed in real-time dispatching. They will test this algorithm using a simulator that employs it against data from a certain month in the past. The project is advised by Professor Jack Muckstadt.
According to team-member Andy Sheung, the project has already learned some valuable lessons. Collecting data is not easy, particularly where qualitative aspects such as the relationship between a technician and a client is taken into account. Establishing the scope of the problem, formulating a real world problem with complex logic into a mathematical program, and designing an easy-to-use interface for the product are all challenging. They have also learned, though day-long rides on the job with some technicians, that the dispatching process must balance the needs of customers, management, and the workers themselves.