2021 – Collaborative Research: Theory-Grounded Guidelines for Solver-Aware System Architecting (SASA)
Project funded by National Science Foundation, The Engineering Design and Systems Engineering (EDSE) Program under Grant No: CMMI 2129574
PI: Panhal, J. (Purdue) & PI: Szajnfarber Z. Co‑PI: Topcu T.G.
The objective of this proposal is to advance the scientific understanding of Solver-Aware System Architecting (SASA), leading to theory-grounded guidelines and analytic tools that can be used by practitioners. SASA is a new paradigm of system architecting that seamlessly integrates talent and expertise from outside the organizational boundaries into the systems design process right from the initial stages. The central hypothesis is that the joint consideration of architecture, solvers, and contracts, can significantly improve systems design outcomes. From a methodology standpoint, the hypothesis is that it is feasible to develop theory-grounded guidelines for SASA through computational modeling, in conjunction with machine learning. The research plan consists of developing a computational modeling framework for SASA, and using it in a multi-agent reinforcement learning environment to extract guidelines for architecting and design of contracts. Empirical validation of the framework will be carried out using a rich dataset from NASA autonomous robotic arm. Dissemination of the outcomes to the practitioners and the broader community is an important part of the plan.
The proposal will extend the frontier of knowledge in systems engineering research and practice. It will fill the knowledge gap in traditional approaches to system architecting which assume traditional players bounded by organizational boundaries, and address the limitations of the open innovation literature has not focused on complex systems design. Specifically, the outcomes of this project will formalize the linkages among innovation processes, designer knowledge, systems architecture and contractual structures. The expected outcomes from this project include: (i) a new socio-technical theory for architecting complex systems, (ii) computational models of open innovation processes for systems design, and computational methods for designing contractual mechanisms for SASA, (iii) empirically-validated design guidelines to support practitioners of SASA, and (iv) a set of educational board games to educate students on SASA.
2021 – Integrating Qualitative and Machine Learning Methods to Understand Go-Around Behavior
Supported by MITRE and the FAA
PI: Topcu T.G. & Co‑PI: Szajnfarber Z.
Text reports are the FAA & NTSB standard for collecting information from safety-critical decision-makers and offer a rich source of information about why safety incidents and accidents happen, including the rationale of, and context for, pilots for choosing to follow or deviate from Air Traffic Controller (ATC) guidance. Particularly in the case of go arounds, an exhaustive multi-faceted analysis of text data may shed new light on the large number of cases where pilots appear to be out of compliance with stated policies but otherwise safe or vice versa. Despite their promise and abundance, text reports have been relatively untapped as a data source because of the challenges associated with systematically extracting meaning from them. To elaborate, text reports are stated in the own words of the actors and often captured in the “heat of the moment”; thus, the information could be subject to a wide array of confounding influences such as the sentiments of the actors, peer pressure, cognitive biases, work environment related factors such as fatigue, and so on. This nuanced and messy nature of data renders any one-dimensional disciplinary approach ineffective and requires a reconsideration of existing learning and text analysis methods.
This project will develop new insights about go arounds in particular, and more broadly will develop a synergistic methodology for extracting meaning from text reports by leveraging the strengths of qualitative inductive coding on the one hand and machine learning on the other. Qualitative coding, a rigorous social science technique that is often disregarded by engineers, excels at inducing insights from unstructured data, leveraging the strength of human coders to identify unexpected and counterintuitive patterns. However, it is a labor intensive process that holds limited potential to scale to large data sets. In contrast, Machine Learning excels at processing large data sets and identifying generalizable patterns across variables. However, it is sensitive to initial data formats and is limited by verbatim representation, which often manifests itself by misidentifying unique events or other overgeneralization issues. In this project, we will focus on go around events as a case study, to compare the patterns identified by each approach and then outline a novel analytical method for combining them. For example, the output of qualitative coding may be useful in training machine learning algorithms to examine deeper features of the data. The hybrid approach we will formulate will be generalizable to other contexts than go around events with certain modifications for extracting enhanced information from text data.
Past – as a graduate student
2021 – Safety and the Role of Near-misses in the Socio-Technical Supervision of Autonomous Infrastructures
Project funded by National Science Foundation, LEAP-HI Program under Grant No: CMMI 2051685
PI: Triantis K., Co‑PI: Dillon‑Merril, R. (Georgetown), Co‑PI: Madsen, P. (BYU), Co‑PI: Srinivasan, D. (Clemson)
There is potential for significant benefits from increased autonomy in infrastructure systems. It is assumed that the most significant benefit is the economic efficiency that can be created by operating these systems with fewer human resources. While the expected increase in economic efficiency is clearly a benefit, additional research considering entire infrastructure systems as complex socio-technical systems is needed, to understand the impact of autonomous technologies on the economics, workload, and safety factors and their tradeoffs. Our proposed work on modeling and understanding the associations among these factors will enable policy makers to obtain transparent systems that they can trust. This understanding can significantly assist infrastructure providers and regulators to provide safe and efficient services. We aim to understand the operational, organizational and human conditions that influence the safety of autonomous sociotechnical infrastructure systems, and propose a systems-level theoretical approach that provides a computable safe area of operation for these systems.
This research originated from my dissertation and I made significant contributions to the proposal.
(2017 – 2020) Collaborative Research: Multi-Perspective Evacuation Performance Measurement
Project funded by National Science Foundation under Grant No: CMMI 1536809
Co-PIs: Konstantinos Triantis, Pamela Murray-Tuite, and Joseph Trainor
This research extends dynamic network efficiency measurement approaches to model the relationships between transportation agencies and households. The research integrates deductive (efficiency modeling) with inductive (on-site surveys) methods.
I supported this project as a graduate student.
(2016) Workshop on Theoretical Foundations of Systems Engineering: the Use of Abstraction and Elaboration
NSF Grant NO: CMMI 1548480
PI: Triantis K.
The purpose of this workshop was to demonstrate the process of theory building for complex socio-technical phenomena that occur during a systems engineering project.
I assisted with the organization of this workshop and authored a follow-up journal article.