(2dy) Decision-Making and Learning Under Uncertainty for Complex Systems

Complex systems of interest in engineering applications typically encounter uncertainty which can stem from a variety of sources such as environmental factors, fitted parameters, forecasting, and small length-scales [2, 4, 7, P1]. Motivating applications include wildfire mitigation, high-throughput material synthesis, sustainable energy production, rare earth element recovery, autonomous vehicles, chemical process design/control, and more [2, 5, 7, P1]. Decision-making (optimization) under uncertainty provides rigorous mathematical strategies to guide the optimal design, effective operation, and enhanced understanding of systems in these challenging application areas, while explicitly incorporating the risk incurred by candidate choices in the decision-making process [3, 6, 8, 9, P2]. Generally, this paradigm involves propagating randomly distributed parameters (characterized by data) through a system model (developed via first principles and/or machine learning) to characterize the output distribution which we manipulate via the system variables (constituting our decisions) to optimize the risk incurred (determined by the output distribution) [7, P1, P2]. My research has focused on developing effective optimization and data science strategies alongside innovative software solutions to tackle this challenging class of problems in support of these fundamental engineering challenges. My past and current research pursuits can be classified as follows.

Design and Assessment of System Flexibility and Reliability- Using the motivating application of mitigating the effects of rare, high-impact events (e.g., extreme weather) on power grids, I developed a suite of methods to assess more accurately the ability of a system to operate in response to random fluctuations (e.g., wind, temperature, renewable generation) and random component failures [1-4]. These metrics for flexibility and reliability, respectively, (implemented in the Julia software package FlexibilityAnalysis.jl) provided tools to optimally design power grids to hedge against the risks posed by these uncertainties.

Unifying Abstraction for Infinite-Dimensional Decision-Making- Infinite-dimensional decision-making problems entail decisions indexed over continuous spaces (e.g., time, spatial position, and/or random events). These capture a wide breadth of optimization disciplines (e.g., dynamic, PDE-constrained, stochastic, and robust optimization) which have been largely studied and applied independently. Drawing inspiration from decision-making under uncertainty, I developed a unifying modeling abstraction (implemented in the Julia package InfiniteOpt.jl) to commonly express these in the same space to promote novel modeling and solution paradigms via theoretical transfer. This has enabled a multitude of contributions detailed in [5-7, 9, P1, P2] that can drive-forward emergent engineering applications in critical material recovery, sustainable energy production, experimental design, and material discovery. A contribution of this work is the invention of random field optimization (decision-making with uncertainty correlated of space-time fields) which will be a key enabling factor in my future work as described further below [7].

Decision-Making with Data-Driven Surrogate Modeling- The outburst of effective methods to build data-driven methods using machine learning provides a powerful avenue to conduct decision-making for systems that cannot effectively be characterized using first principles alone. I am working on incorporating state-of-the-art data-driven models (e.g., neural operator networks) into rigorous decision-making tools (implemented using Python and Julia) to enable advancement in the development of rare earth element recovery networks, carbon capture systems, and epidemiology mitigation policy [10].

Robust Computer Vision Sensing in Industry 4.0- The emergence of advanced photogrammetry sensors in modern process systems have enabled significant improvements in automation, profitability, and consistency. However, these introduce significant vulnerabilities to visual disturbances and cyber-attacks. To combat this, I developed an effective framework to detect and respond to these visual uncertainties [8].

Future Direction

My future research group will develop innovative strategies and tools in decision-making and data-science under uncertainty to solve real-world engineering challenges. Two initial motivating application areas will include the operation/design of rare earth element (REE) and critical material (CM) networks and the modeling/mitigation of wildfires. These exhibit nontrivial properties due to the complex nature of the underlying spatio-temporal physics which are subject to severe uncertainties stemming from the environment and the interactions of stakeholders.

Establishing a sustainable domestic supply-chain of REE-CMs is an engineering challenge of high priority to the U.S. Department of Energy and the Department of Defense. REE-CMs are critical to manufacturing electronic components needed for an array of technologies such as lithium batteries, mobile devices, and renewable energy generators (e.g., wind-mills and solar cells). My group will build upon my ongoing research efforts in this area to develop methods that optimally design/operate REE-CM recovery networks across multiple enterprises. The foundation of methods I have pioneered in random field optimization and neural operator surrogate modeling will provide my group unique expertise to innovate in this application space.

Wildfire modeling and mitigation present important endeavors with the rise of frequent large wildfires across the U.S. However, rigorous decision-making strategies have seen little application in this area due to the underlying complexity and severe stochasticity associated in capturing the behavior of wildfires. Again, the unique skillset of my group in decision-making and surrogate modeling under uncertainty for complex spatio-temporal systems provides an avenue to innovate and make significant progress in this space. Moreover, this effort will provide the means to fundamentally improve these optimization strategies and drive the way toward widespread use in process systems engineering.

Teaching Experience and Interests

My teaching philosophy focuses on building a flipped classroom environment that features active learning that promotes a growth mindset inclusive of students from diverse backgrounds, experience, and confidence levels. I also believe that a professor’s research should actively influence and improve the courses they teach.

During my undergraduate and graduate education, I taught a several of recitation sections in organic chemistry, engineering computing methods, and process control. Moreover, I co-instructed a course on engineering computing methods at UW-Madison where I received a 94% approval rating following applying the above teaching strategy. More recently, I have taught short courses on programming in Julia with an emphasis of infinite-dimensional optimization at UW-Madison, Carnegie Mellon University, and in South Korea with several more planned in the coming months throughout South America and Mexico.

As a professor, I am well-qualified to instruct undergraduate courses which include (but are not limited to) computational techniques, process design, process control, unit operations, and statistics. Here, my expertise in considering and optimizing system risk can help innovate courses such as process design to help students actively consider the influence of uncertainty on process operation. Moreover, my technical expertise enables me to lead/create graduate courses in optimization and data-science.

Presentations at the current AICHE Annual Meeting

[P1] “Event Constrained Optimization.” J.L. Pulsipher, C.D. Laird, and I.E. Grossmann. Session: CAST Division Plenary Invited Talks (Nov. 14 @ 9:10am)

[P2] “New Measures for Shaping Trajectories in Dynamic Optimization.” J.L. Pulsipher, B.R. Davidson, and V.M. Zavala. Session: Advances in Nonlinear and Surrogate Optimization (Nov. 15 @ 8:00am)

[P3] “Computer Vision Aided Process Control: Methods for Enhanced Autonomy and Robustness.” J.L. Pulsipher, L.D.J. Coutinho, T.A. Soderstrom, and V.M. Zavala. Session: Advances in Process Control I (Nov. 15 @ 9:54am)

Select Publications (In Chronological Order)

[1] J.L. Pulsipher and V.M. Zavala. “A Mixed-Integer Conic Programming Formulation for Scalable Computation of the Flexibility Index under Multivariate Gaussian Uncertainty.” Computers & Chemical Engineering. 2018

[2] J.L. Pulsipher, D. Rios, and V.M. Zavala. “A Computational Framework for Quantifying and Analyzing System Flexibility.” Computers & Chemical Engineering. 2019

[3] J.L. Pulsipher and V.M. Zavala. “A Scalable Stochastic Programming Approach for the Design of Flexible Systems.” Computers & Chemical Engineering. 2019

[4] J.L. Pulsipher and V.M. Zavala. “Measuring and Optimizing System Reliability: A Stochastic Programming Approach.” TOP. 2020

[5] J.L. Pulsipher, W. Zhang, T.J. Hongisto, and V.M. Zavala. “A Unifying Modeling Abstraction for Infinite-Dimensional Optimization.” Computers & Chemical Engineering. 2021

[6] J.L. Pulsipher, B.R. Davidson, and V.M. Zavala. “New Measures for Shaping Trajectories in Dynamic Optimization.” DYCOPS. 2022

[7] J.L. Pulsipher, B.R. Davidson, and V.M. Zavala. “Random Field Optimization.” Computers & Chemical Engineering. 2022

[8] J.L. Pulsipher, L.D.J. Countinho, T.A. Soderstrom, and V.M. Zavala. “SAFE-OCC: A Novelty Detection Framework for Convolutional Neural Network Sensors and its Application in Process Control.” Under Review. 2022

[9] J.L. Pulsipher, D. Ovalle, H.D. Perez, C.D. Laird, and I.E. Grossmann. “Characterizing Event Constraints with Generalized Disjunctive Programming.” Under Review. 2022

[10] J. Liu, J.L. Pulsipher, D.A.T. Cummings, J.D. Siirola, and C.D. Laird. “Computationally Efficient Global Optimization Approaches for Estimation of Seasonal Transmission Parameters in Infectious Disease Models.” Under Preparation. 2022