CBE M.Eng. Projects

Faculty Sponsor: Fengqi You

The optimal design of compounds through manipulating properties at the molecular level is often the key to considerable scientific advances and improved process systems performance. In spite of century-long efforts in chemical synthesis and the large set of synthesized molecules (~107), the so-called chemical space is still an unexplored galaxy with an estimated number of small organic molecules populating the space of more than 1060.   This project focuses on Computer-Aided Molecular Design (CAMD) using machine learning techniques, such as variational autoencoders, generative adversarial networks, self-supervised learning, and reinforcement learning.  

Faculty Sponsor: Fengqi You

The concept of smart grid has gained considerable attention from both academia and industries. With the increasing penetration of renewable wind energy in electric power systems, the balance between electricity supply and demand is challenging. The emerging power-to-gas technology effectively converts excessive electricity into natural gas. The manufactured natural gas can be sold to the market or used by gas-fired units. Power-to-gas technology could support a deep penetration of wind energy. However, the uncertainty in wind power generation and dynamics in natural gas system pose a great challenge to the coordination of electric power systems and natural gas systems. 

In this project, we aim to leverage the power of big data analytics and statistical machine learning to extract useful uncertainty information from historical wind data, and organically integrate it into the robust optimization model for scheduling of power generating units, power-to-gas facilities and natural gas production wells. The model formulation includes both the power system part and the natural gas system part. The case study will be based on the IEEE-118 bus power system and 12-node gas system. The electric power system has 7 wind farms, which has a total capacity of 720MW. There are 12 gas nodes, 3 gas wells and 12 pipelines in the natural gas system

Faculty Sponsor: Fengqi You

Cornell University campus aims to Carbon Neutral by 2035. To support this renewable energy transition, hybrid energy systems should be integrated, designed and optimized to satisfy the campus-wide demand on electricity, heat and cooling. This project aims to deliver energy systems analysis and optimization methods and insights to support the renewable energy transition of the campus. The campus energy systems will also provide a living laboratory for this study.

Faculty Sponsor: Jefferson Tester

Objective: In support of Cornell's Climate Action Plan (CAP), complete a comprehensive assessment on the impact and trade-offs of hydrogen supplied steam generation vs Renewable Natural Gas (RNG) supplied Cornell Campus steam generation using the current steam generation process climate footprint as a baseline for comparison. This should include a thorough assessment of current competitive landscape of hydrogen and RNG , available technologies, NYS/federal/local laws and regulations affecting the technology (especially combustion based), impact to current Cornell assets, some level of financial impact (capital and operating cost) of alternatives. The project will start with a tour of the Cornell Combined Heat & Power Plant.

Faculty Sponsor: Fengqi You

Within the PCMM system, the powder formulation components (Active Pharmaceutical Ingredient (API) and excipients) are gravimetrically fed into a vertical, in-line mixer at fixed mass ratios, proportionate to the formulation composition. The homogeneous blends leaving the mixer are then continuously fed into a rotary tablet press and compressed into tablet cores. The tablet cores are vacuum conveyed into a semi-continuous film coating system, where a color film coat is applied to the surface of the tablet core. The film coated tablet are then packaged into drums for subsequent packaging into bottle or blister configurations via a separate operations.

Faculty Sponsor: Fengqi You 

This project aims to apply deep learning models and algorithms to materials related problems. A specific example is developing convolutional neural networks (CNNs) to detect the optical response (micrographs) from liquid crystal interfaced with different proteins. Another example is applying natural language processing (NLP) methods, coupled with 2D representation using Simplified Molecular Input Line Entry System (SMILES), to establish the quantitative structure-property relationship (QSPR) and facilitate polymer design.

Faculty Sponsor: Fengqi You 

In recent years, deep learning has attracted tremendous attention from both academia and industries, due to its state-of-the-art performance in many areas, such as speech recognition and image processing. Deep learning, one of the most rapidly growing machine learning subfields, demonstrates remarkable power in deciphering multiple layers of representations from raw data without any domain expertise in designing feature extractors. Since it requires very little feature engineering by hand, deep learning can easily take advantage of big data in chemical engineering. With growing amount of data in chemical processes and great advances in computational infrastructure, deep learning holds the potential to revolutionize numerous domains in chemical engineering. In this project, we will apply deep learning techniques to various chemical engineering problems. Among these, deep convolutional neural network (CNN) will be employed for chemical process monitoring to ensure the safety of process operations. By leveraging the natural gas consumption profiles, we will develop an accurate natural gas consumption forecasting model with long short term memory (LSTM) neural networks. Additionally, material discovery typically runs with a trial-and-error approach. By exploiting patterns in massive datasets, deep generative models characterize salient features of molecules and greatly facilitate material discovery.

Faculty Sponsor: Syed S.H. Rizvi

Modern technologies are needed to become more energy efficient in order to reduce their environmental footprint and lower energy costs. Our food engineering lab has developed an energy efficient, novel system for flash-freezing liquid foods that promotes fresh and clean-label food products, especially ice cream. Furthermore, this process relies on economy of scale to be profitable and therefore large, single product batches are common while customized products can be difficult or even impossible to produce economically. The on-demand freezing/cooling system, based on the combined Joule-Thompson and Bernoulli principles, requires expertise and interest in modeling and simulation to provide a framework for scale up and coefficient of performance evaluation compared to the vapor compression refrigeration systems.

Faculty Sponsor: Jeff Varner

Data driven stock and currency trading approaches require access to high quality financial data streams. Unfortunately, there are only a limited number of free or low cost vendors for this type of data. One such vendor is Alpha Vantage, a leading provider of free APIs for realtime and historical data on stocks, forex (FX), and digital/crypto currencies that provides intraday, daily streams along with a large number of technical indicators. However, while Alpha Vantage is free it does require fairly detailed knowledge of web-based programming approaches. Toward this challenge, in this project, we will develop a wrapper around the Alpha Vantage API that allows users to access the data stream without having to directly make the web-based API calls. In addition, this wrapper should transform the raw data produced by the Alpha Vantage into a common format that is easier to work with. This wrapper will be written in the Julia Programming Language, a modern high performance computing language developed at MIT [1] that is gaining popularity in the financial community, including the Federal Reserve Bank of New York [2]. The Julia framework developed on this project will be published as a Julia package under an MIT software license, and released to the financial technology community via GitHub and released on arXiv.org. Taken together, this project offers students training in the development of application programming interfaces (APIs), the Julia programming language, realtime financial data streams and financial modeling approaches.

1. Julia: A Fresh Approach to Numerical Computing. Jeff Bezanson, Alan Edelman, Stefan Karpinski and Viral B. Shah (2017) SIAM Review, 59: 65–98. doi: 10.1137/141000671.
2. Michael Cai, (2108). Estimating Non-Linear Macroeconomic Models at the New York Fed.

Faculty Sponsor: Alex Woltornist

Faculty Sponsor: Jefferson Tester

Cornell is planning to demonstrate the direct use of low-carbon geothermal heat (Earth Source Heat, or ESH) at its Ithaca campus to provide heat for the campus as a part of implementing its Climate Action Plan. This demonstration focuses on job creation in energy and agriculture. ESH provides jobs for energy workers. Agricultural jobs are linked to both the energy supply (supplemental biomass heating) and demand (controlled environment agriculture, or CEA) components. ESH will enable new year-round biomass and CEA markets, replace fossil fuels, and create a broad range of short- and intermediate-term development jobs and long-term operational jobs. Cornell is attractive for demonstration due to 1) representative regional geology with sufficient geothermal resources; 2) heating demand representative of NY State ; 3) existing district energy infrastructure to buildings and laboratories serving 30,000 people (including State-supported colleges); 4) opportunities for multiple cascading uses/applications; and 5) active collaborations with academic, industrial, and government partners interested in this technology. Geothermal resources will be used to meet seasonal base-load/average heat demands. The overall system will be optimized using thermal storage and heat pumps; and identify cascading uses for the heat. The integration of biomass for peak heating with cascading end-use to maximize the economic benefit to NY while promoting sustainability on campus. The US DOE has funded a 2-yr ESH Feasibility Study. At the present time Cornell is seeking funding for the next, critical phase: drilling and development of a demonstration deep geothermal well pair, which is key to attracting private funding for market development (see attached support letters). The MEng project will deal with the system analysis of various geothermal – biomass design options. The work will involve a collaboration with CBE and EAS faculty as well as with engineering staff from Cornell’s facilities management group.

Faculty Sponsor: Syed S.H. Rizvi

Conventional steam-based extrusion (SBX) process is a commercially practiced technology to produce a large variety of expanded food products. During SBX, a heterogeneous melt of starch and other ingredients undergoes a high-temperature (120-180o C), high-shear cooking where water acts both as a plasticizer for melt formation and a blowing agent for expansion. The harsh operating conditions of the SBX process often prevent effective utilization of formulations containing heat and shear sensitive ingredients. Steam-expanded products usually show non-uniform cellular structures and cell sizes. Supercritical fluid extrusion (SCFX) is a novel technology that uses supercritical carbon dioxide (SC-CO2) as a blowing agent, and hence formulations containing heat-sensitive ingredients can be employed to make expanded products at temperatures below 100o C. A higher moisture content (i.e., 30-45 wt.%) in the extruder barrel is utilized to keep the product temperature low via reduction of viscous dissipation of mechanical energy and to maximize SC-CO2 solubilization in the melt. The SCFX results in a more homogenous nucleation and uniform microporous structure. However, the advantage of maintaining low temperatures to minimize thermal degradation of ingredients also poses a major engineering challenge in the process scale-up. When a supercritical extruder is scaled-up, the extruder volume is tripled, whereas the cooling surface area is just doubled resulting in an inefficient cooling and high temperatures. The objective of this project is to perform energy studies on a supercritical twin-screw extruder as a function of its operating parameters to quantify and optimize the cooling efficiency needed for scale-up of the process.

Faculty Sponsor: Alex Woltornist

Faculty Sponsor: Jefferson Tester

Experimental and theoretical research is our group is focused on using thermochemical processing involving pyrolysis and/or hydrothermal liquifaction to convert a range biomass feedstocks from dairy and food wastes to algae containing carbohydrates, proteins and triglycerides and fatty acids in varying amounts to drop-in transportation fuels, sugars, syngas and biochar. In this MEng project students would assist group team members in conducting laboratory-scale experimental conversion studies with kinetic, phase equilibria and mass transport issues considered. The main objectives are to characterize the effects of operating temperature, pressure, and residence time on product yield and to identify the main reaction pathways.
 

Faculty Sponsor: Fengqi You 

Data-driven optimization has received immense attentions nowadays due to its close integrations with both machine learning and operation research. Decision-making under uncertainty poses significant challenges for industrial process control and operations. Machine learning provide powerful tools for uncertainty quantification and modeling, and in turn facilitates the advances of decision-making tools. Production scheduling is a general problem in process industries that involves various types of uncertainties, such as random processing times, fluctuating demands, etc. With the development of sensor and data storage technologies, a significant amount of uncertainty data are being collected online, which provide possibilities to leverage the power of big data analytics to address the challenge of decision-making under uncertainties. In this project, we will use kernel learning-based robust optimization approach to formulate the decision-making problem. Support vector clustering (SVC) as a powerful machine learning technique will be adopted to extract meaningful statistical information from historical data first, which is integrated into the robust optimization model. Then the robust counterpart in the form of mixed-integer linear program will be derived and solved to inform the process control and operations decisions. Comparative studies will be performed to illustrate the exclusive benefits that massive historical data could provide in decision-making in process industries. This project will be helpful for enhancing one’s knowledge and programming skills in both machine learning and optimization.

Faculty Sponsor: Jeff Varner

Background: An option contract is an equity derivative whose value is based on the price movements of an underlying asset, for example a single company stock or an equity basket such as an exchange traded fund (ETF). Equity options provide the right, but not the obligation, to buy (call option) or sell (put option) a quantity of stock or ETF, where one option contract controls 100 shares of an individual company stock or ETF, at a set price (strike price), within a certain period of time (prior to the expiration date). European option contracts can be exercised only on the expiration date, while American style option contracts can be exercised at any time between the time the option was purchased and the expiration date. Financial engineers, traders and investors use option contracts to control the risk associated with taking a position (buying or selling) an underlying asset, or to speculate on the price movements of an underlying asset. They also use option contracts to take positions in an underlying asset without actually buying or selling the asset. This is advantageous because taking an option position allows significant leverage; the amount of capital needed is much less with an option compared to a similar position in the underlying asset directly. However, this leverage comes at a price. Option buyers are charged (by option sellers) a premium for the rights and privileges associated with the option contract. This premium is a function of the price of the underlying asset, the volatility of the asset price, and the time to expiration.

Objectives: The objective of this project is to develop a monte carlo approach to simulate the price of combinations of European and American style option contracts in the presence of dividend payments from the underlying asset. Toward this objective, geometric brownian motion models of the price variation of the underlying asset will be developed and used in combination with discrete, and continuous option pricing algorithms to produce an option price prediction as a function of market factors and contract parameters. Option price predictions will be backtested using historical data, and forward tested by comparing predicted and current option prices for different underlying asset classes, both collections of single companies and a broad range of ETFs. Approach. The option pricing algorithms and asset pricing models will be implemented in the Julia programming language. Lattice based approaches will be used to provide discrete alternatives to the Black–Scholes–Merton model modified to include dividend payments, and American style option contracts. The latter approach for valuing American style option contracts involves using the Black–Scholes–Merton model as a constraint in an optimal stopping problem calculation.

Deliverables: All project codes, documentation and testing data will be made available under an MIT software license on GitHub. The student will submit a project report, and present the project findings at department, college and university symposia.

Faculty Sponsor: Fengqi You 

The development of new pharmaceuticals is a long and expensive process. It takes an average of 15 years for a new drug to move from the discovery into the marketplace, and the average cost for the development of a new drug is more than $900 million. Out of 5,000 compounds that emerge from discovery, only five perform well enough to move into human testing, and only one of these five compounds is approved by the Food and Drug Administration (FDA). At the same time, due to changing circumstances in the managed-health-care environment, the profit margins of US pharmaceutical companies and the productivity of their Research and Development (R&D) pipelines (in terms of new entities registered per dollar of investment) have started to decline; effective patent lives have been shortened, and patents provide lower barriers to entry even while active. Therefore, it is imperative for pharmaceutical companies to manage their R&D pipelines more effectively to reduce the cost of developing new drugs. This is a challenging task due to the highly stochastic nature of the R&D process: if a drug fails a clinical trial, its development stops and all prior investment is lost; if it passes all trials, it enters the marketplace and profits are typically significantly larger than development costs. To address this problem, we will develop a multi-stage stochastic programming strategy that accounts simultaneously for the selection of drugs, the scheduling of clinical trials, and the resource planning and out-licensing decisions. We will also develop new theoretical results and solution methods to solve large-scale instances. The planning of R&D activities falls into a broader class of less-studied optimization problems, namely, stochastic optimization problems under endogenous observation of uncertainty. The challenge in addressing these problems is due to the fact that the decision-maker alters the underlying stochastic process by changing the timing of uncertainty observation. We study this class of problems and generalize the results and solution methods we developed for R&D pipeline planning.

Faculty Sponsor: Fengqi You

When a battery or electronics reach the end of their primary life, manufacturers have two options if the packs are still functional: 1) recycling the precious metals or other energy-intensive materials embedded within the packs; 2) repurposing the packs for less-demanding second-life applications, such as stationary energy storage. Given a sufficient gap between the procurement and recycling cost, recycling can make sense and drastically reduce the use of virgin resources toward a circular economy. In this way, sustainable manufacturing of battery and electronics is achieved without sacrificing the quality of life for consumers. On the other hand, second-life applications provide the most market value where there is demand for batteries or electronics for stationary energy storage applications that require less-frequent battery cycling. The economic performances and environmental implications of second-life of batteries and electronics vary significantly depending on the applications, requirements, and market conditions. Therefore, it is necessary to conduct systematic analyses from energy systems engineering and sustainability perspectives to identify the most promising recycling strategies. To address this problem, we will systematically perform comparative techno-economic and environmental life cycle analyses (LCA) of the second-life pathway and recycling pathway, given a wide spectrum of device types, application requirements, and market conditions. We will also use life cycle optimization that integrates the tenets of techno-economic analysis and LCA through a multi-objective optimization framework to determine the optimal pathway and shed light on the EEE performance of the batteries and electronic products.

Faculty Sponsor: Syed S.H. Rizvi

The need for cheese analogues like plant protein based vegan cheeses has risen sharply in the last few years and the trend is projected to continue over the next several years. A combination of shear, temperature and pressure during extrusion processing creates opportunities for both conformational changes and chemical reactions in many proteins, especially plant proteins, to modify their rheological properties to mimic cheese-like viscoelastic behavior. This project is aimed at high-shear modification of selected plant proteins via supercritical fluid extrusion processing and fingerprinting their rheological and functional properties.

Faculty Sponsor: Fengqi You

Weather forecasts are essential for optimal predictive control of energy systems. However, conventional machine/deep-learning-based weather forecasts are either short-term or long-term so that it cannot be directly used for coordinated control of multiple systems with different response times. The project will develop a novel scheme for short-/long-term stochastic weather forecasts using deep learning techniques for temporal coordinated control of multiple energy systems. The project aims to develop a novel scheme for short-/Long-term stochastic weather forecasts using deep learning techniques for temporal-coordinated energy systems (e.g., buildings, distributed energy systems) control. The project tasks include: (1) Develop a scheme for both short-term and long-term weather forecasts using deep learning, (2) Apply the scheme to building control through simulations.

Faculty Sponsor: Fengqi You

Reinforcement learning and model predictive control are promising model-free and model-based optimal control approaches for dynamic systems. However, there are still many challenges when they are applied to real-world systems (e.g., buildings) due to their model-free/model-based features. A model predictive control guided reinforcement learning scheme will be developed merging their strengths to overcome the challenges. The project aims to develop a novel control and optimization scheme combining deep reinforcement learning and model predictive control for smart control of building systems including cooling, heating, ventilation and lighting. The project tasks include: (1) Setup a live testbed in buildings on Cornell campus, (2) develop a scheme combining model-based and model-free approaches and apply it to the live testbed.

Faculty Sponsor: Jefferson Tester, Max Zhang (MAE)

In this project, you shall investigate the technical, environmental and economic performance of geothermal heat pumps used for cooling of large-scale data centers. Based on energy data from a large scale commercial data center, you shall propose the design of a geothermal heat pump system using industry standard methods to replace currently employed absorption chillers and/or vapor compression refrigeration machines. Specifically, the design includes sizing of heat pumps and boreholes, and estimating capital and operating costs and full life cycle cost. Also a comparison of the environmental impacts and benefits with current cooling systems should be made that includes tradeoffs in terms of energy consumption, costs, and CO2 and other emissions.
 

Faculty Sponsor: Alex Woltornist

Two addition ideas that are really valuable to Air Products, linked to our Green H2/NH3 developing business, require fundamental Chemical Engineering calculations, but I fear may not appear super exciting on the surface are both regarding the large ammonia tanks that we (and others in the industry) are starting to build, own, and operate: o Although there is API guidance for sizing relief valves for these NH3 tanks, the guidance is surprisingly vague for what we call the barometric pressure case (basically what happens to the NH3 tank pressure when a hurricane approaches and passes over the NH3 tank, for example in Texas of Louisiana). This case ends up being the sizing case in many projects. We have therefore come up with a dynamic model solution for determining the relief flows required, but it is complex and it takes time to solve. We would like a student/team to dig into the details, see if they agree with our methodology, and see if a steady state solution is possible for different tank sizes. o In addition, there is something called a thermal overload case where warm NH3 gets added to a tank full of cold NH3. Again, the industrial guidance is surprisingly vague and we need to know when there is a relief case in this situation and what would the relief flows be.