Zhenghui Sha

Behavioral Experimentation and Game Theory in Engineering Systems Design

Game-theoretic models have been used to analyze design problems ranging from multi-objective design optimization to decentralized design and from design for market systems (DFMS) to policy design. However, existing studies are primarily analytical in nature, which start with a number of assumptions about the individual decisions, the information available to the players, and the solution concept (generally, the Nash equilibrium). There is a lack of studies related to engineering design, which rigorously evaluate the validity of these assumptions or that of the predictions from the models. Hence, the usefulness of these models to realistic engineering systems design has been severely limited. In this paper, we take a step toward addressing this gap. Using an example of crowdsourcing for engineering design, we illustrate how the analytical game-theoretic models and behavioral experimentation can be synergistically used to gain a better understanding of design situations. Analytical models describe what players with assumed behaviors and cognitive capabilities would do under specified conditions, and the behavioral experiments shed light on how individuals actually behave. The paper contributes to the design literature in multiple ways. First, to the best of our knowledge, it is a first attempt at integrated theoretical and experimental game-theoretic analysis in design. We illustrate how the analytical models can be used to design behavioral experiments, which, in turn, can be used to estimate parameters, refine models, and inform further development of the theory. Second, we present a simple experiment to understand behaviors of individuals in a design crowdsourcing problem. The results of the experiment show new insights on using crowdsourcing contests for design.

Towards the design of complex evolving networks with high robustness and resilience

Abstract Network design and optimization research has traditionally been focused on networks where the designers have direct control over the nodes and their connectivity. However there is increasing importance of social, economic and technical networks whose structures are not under the direct control of the designers, but evolve as a result of decisions and behaviors of individual self- directed entities. These networks are endogenous in nature, where the local characteristics and behaviors of nodes affect the overall structures. The structure of a network affects its properties, and the properties affect the systems performance. Hence, the problem of designing such endogenously evolving networks involves determining the node-level characteristics and behaviors through appropriate incentives to achieve the desired system-level performance. In this paper, our goal is to illustrate the problem of designing endogenously evolving networks, and to present a specific illustrative example. We perform a conceptual exploration of the problem, present the current state of the art and identify research gaps. The illustrative example involves designing an endogenous network with two objectives, robustness to random node failure and resilience to targeted attack, considering specific node-level characteristics, additional attractiveness, as the design variables. The impact of the design variables on the performance of the network, and potential applications are discussed.

Modeling Airline Decisions on Route Planning Using Discrete Choice Models

We propose a model for the airlines’ decisions on route planning, i.e., the decision on selecting which route to add and delete, using discrete choice random-utility theory. The central hypothesis is that a discrete choice model can effectively model the airlines’ decisions on route selection , and thereby help model the evolution of the air transportation network. We first model the airlines’ utility function as a linear function of decision variables with interaction effects. The decision of route selection is then modeled using a binary choice model derived from the utility function. The preferences for each variable in the utility function are estimated using historical datasets. Advantages of this approach include the ability to use statistical techniques to quantitatively construct decision models as well as to account for the uncertainty in unobserved attributes of the decision model. The proposed model helps predict the airlines’ decisions on routes addition and deletion which affect the network topology of air transportation and its future evolution. This capability can be beneficial to other stakeholders, such as Federal Aviation Administration, who may need to make their decisions in response to those made by the airlines, but do not have access to the airlines’ true decision models.

Estimating Local Decision-Making Behavior in Complex Evolutionary Systems

Research in systems engineering and design is increasingly focused on complex socio-technical systems whose structures are not directly controlled by the designers, but evolve endogenously as a result of decisions and behaviors of self-directed entities. Examples of such systems include smart electric grids, Internet, smart transportation networks, and open source product development communities. To influence the structure and performance of such systems, it is crucial to understand the local decisions that result in observed system structures. This paper presents three approaches to estimate the local behaviors and preferences in complex evolutionary systems, modeled as networks, from its structure at different time steps. The first approach is based on the generalized preferential attachment model of network evolution. In the second approach, statistical regression-based models are used to estimate the local decision making behaviors from consecutive snapshots of the system structure. In the third approach, the entities are modeled as rational decision-making agents who make linking decisions based on the maximization of their payoffs. Within the decision-centric framework, the multinomial logit choice model is adopted to estimate the preferences of decision-making nodes. The approaches are illustrated and compared using an example of the autonomous system (AS) level Internet. The approaches are generally applicable to a variety of complex systems that can be modeled as networks. The insights gained are expected to direct researchers in choosing the most applicable estimation approach to get the node-level behaviors in the context of different scenarios.

Modeling Airlines’ Decisions on City-Pair Route Selection Using Discrete Choice Models

An approach based on the discrete choice random-utility theory is presented to model airlines’ decisions of strategically adding or deleting city-pair routes. The approach consists of methods for i...

Scientific Foundations for Systems Engineering: Challenges and Strategies

There is an increasing realization of the need for fundamental research in the science of systems engineering. The International Council on Systems Engineering vision document calls for theoretical foundations for systems architecting, systems design and systems understanding. During a recent NSF workshop, a number of knowledge areas ranging from mathematics, information sciences, physical sciences, systems science to human and social sciences were identified as possible sources from which the scientific foundation of systems engineering can be enhanced. However, the primary challenge facing the community lies in orchestrating the breadth and diversity of the many knowledge areas into a cohesive foundation. This paper briefly surveys systems science-related efforts across multiple application domains. The specific objectives in this paper are to present a classification of initiatives for developing foundations for systems engineering, and to discuss the challenges, and potential strategies forward, associated with systems science research. The classification is discussed using two case examples — the Internet and the air transportation system. Through these examples, some of the key research challenges and strategies are exemplified.Copyright

Estimating linking preferences and behaviors of autonomous systems in the Internet using a discrete choice model

The Internet is a system-of-system composed of devices and networking technologies operated by autonomous systems (AS). The linking decisions made by ASes affect the topology, and thereby the performance of the Internet. The objective in this paper is to estimate AS linking preferences and behaviors in the evolution of the AS-level Internet using discrete choice models (DCM). The proposed approach utilizes the observable Internet topology as the input to estimate preferences and behaviors of ASes by considering not only the network metrics but also the geographic and economic aspects, such as AS geographic location, inter-domain traffic, business role and provider-customer relationships. The approach provides a way to quantify the impact of different variables on the linking probabilities. The results show that a) the geographic distance and the number of customers play a significant role in the AS linking behaviors, and b) heterogeneity in the linking preferences of ASes is important for accurate characterization of AS linking behaviors. The uniqueness of the proposed approach is that it provides an interpretation of AS decisions through the lens of utility-maximization principles. The approach can help in developing improved decision-making models for AS-level Internet topology generators.

A Generative Network Model for Product Evolution

Modeling the structure and evolution of products is important from the standpoint of improving quality and maintainability. With the increasing popularity of open-source processes for developing both software and physical systems, there is a need to develop computational models of product evolution in such dynamic product developments scenarios. Existing studies on the evolution of products involve modeling products as networks, taking snapshots of the structure at different time steps, and comparing the structural characteristics. Such approaches are limited because they do not capture the underlying dynamics through which products evolve. In this paper, we take a step toward addressing this gap by presenting a generative network model for product evolution. The generative model is based on different mechanisms though which networks evolve—addition and removal of nodes, addition and removal of links. The model links local network observations to global network structures. It is utilized for modeling and analyzing the evolution of a software product (Drupal) and a physical product (RepRap) developed by open source processes. For the software product, the generated networks are compared with the actual product structures using various network measures including average degree, density, clustering coefficients, average shortest path, propagation cost, clustered cost, and degree distributions. For the physical product, the product evolution is analyzed in terms of the proposed mechanisms. The proposed model has three general applications: longitudinal studies of a products evolution, cross-sectional studies of evolution of different products, and predictive analyzes.

Estimating the Node-Level Behaviors in Complex Networks From Structural Datasets

There is an emerging class of networks that evolve endogenously based on the local characteristics and behaviors of nodes. Examples of such networks include social, economic, and peer-to-peer communication networks. The node-level behaviors determine the overall structure and performance of these networks. This is in contrast to exogenously designed networks whose structures are directly determined by network designers. To influence the performance of endogenous networks, it is crucial to understand a) what kinds of local behaviors result in the observed network structures and b) how these local behaviors influence the overall performance. The focus in this paper is on the first aspect, where information about the structure of networks is available at different points in time and the goal is to estimate the behavior of nodes that resulted in the observed structures. We use three different approaches to estimate the node-level behaviors. The first approach is based on the generalized preferential attachment model of network evolution. In the second approach, statistical regression-based models are used to estimate the node-level behaviors from consecutive snapshots of the network structure. In the third approach, the nodes are modeled as rational decision-making agents who make linking decisions based on the maximization of their payoffs. Within the decision-making framework, the multinomial logit choice model is adopted to estimate the preferences of decision-making nodes. The autonomous system (AS) level Internet is used as an illustrative example to illustrate and compare the three approaches.Copyright

Integrated Part Classification for Product Cost and Complexity Reduction

The manufacturing industry is moving towards a truly global arena. Organizations are adopting the philosophy of “design anywhere, manufacture anywhere, and sell anywhere”. Global operations with local focus have become the core of an organization’s strategy. Organizations are trying to have a vast product portfolio with mass customization to meet the customers’ increasing demand for personalized products. While expanding the product portfolio and bringing new products to the market the aspect of product sustenance across its life cycle is often missed out. With regulatory standards becoming more stringent, product maintenance and retirement are becoming challenging and costly. The aspect of “circular economy” is extending the life of the product and individual parts beyond the traditional end of life with re-fabrication, reconditioning and recycling of parts. The part-level detailing is becoming very important at the design stage. This provides huge growth opportunities for organizations, but comes with challenges of increased complexity, variety and cost.One of the potential ways to address the challenges listed above is the availability and maintenance of part-level information and dynamic traceability across the lifecycle, enriched with cross functional inputs. This is important for business decision making during product portfolio planning and product design in both proactive and reactive scenarios. Based on the authors’ industry experience across multiple product development organizations, it is evident that there is limited awareness of the potential of classification and its impact beyond basic part search and reuse. In this paper, we discuss the need for an integrated, cross-functional model and a common database for part information management. We present an agent-based simulation to show the benefits of such an integrated modeling strategy. In the process, the approach has the potential to also bring configurability of the product till the end of life. Configurability is from the aspect of making a product that will perform to meet customer needs along with delivering profit for business and being compliant with various regulatory norms.Copyright