Solving the Reconfigurable Design Problem for Multiability with Application to Robotic Systems

Systems that can be reconfigured are valuable in situations where a single artifact must perform several different functions well, and are especially important in cases where system demands are not known a priori. Design of reconfigurable systems present unique challenges compared to fixed system design. Increasing reconfigurable capability improves system utility, but also increases system complexity and cost. In this article a new design strategy is presented for the design of reconfigurable systems for multiability. This study is limited to systems where all system functions are known a priori, and only continuous means of reconfiguration are considered. Designing such a system requires determination of (1) what system features should be reconfigurable, and (2) what should the range of reconfigurability of these features be. The new design strategy is illustrated using a reconfigurable delta robot, which is a parallel manipulator that can be adapted to perform a variety of manufacturing operations. In this case study the tradeoff between end effector stiffness and speed is considered over two separate manipulation tasks. ∗Graduate Student, Department of Industrial and Systems Engineering, ASME Student Member †Assistant Professor, Department of Industrial and Systems Engineering, ASME Member 1 MOTIVATION Reconfigurable systems have been defined in literature as “those systems that can reversibly achieve distinct configurations (or states), through alteration of system form or function, in order to achieve a desired outcome within acceptable reconfiguration time and cost [1].” (Systems that are able to change their states or configurations have also been called “adaptable” [2], or “flexible” [3,4]. We will continue using the term “reconfigurable” for consistency, but note in passing that the use of the adjective “flexible” to describe reconfigurable systems could cause confusion, especially when designing systems with elastic compliance, i.e. mechanical flexibility.) Many practical applications benefit from the implementation of reconfigurability in design. A prime example is manufacturing. According to Koren, high responsiveness is among the three principal goals of modern manufacturing systems [5]. In 1996, the state of Michigan, the National Science Foundation, and several major manufacturing companies provided $47 million in seed money to found the Engineering Research Center for Reconfigurable Manufacturing Systems at the University of Michigan. This funding resulted in over 1700 papers published about reconfigurable manufacturing systems by 2010. Reconfigurable manufacturing systems offer firms the ability to change manufacturing capacity and functionality quickly, allowing fast response to unpredictable changes either externally (e.g. in market demand) or internally (e.g. in necessary machine maintenance) [6]. 1 Copyright c © 2014 by ASME FIGURE 1: POLYBOT ROBOT DEVELOPED AT XEROX PALO ALTO RESEARCH CENTER [8] Reconfigurability also finds application in aircraft design. Morphable airfoils are a reconfigurable system with the potential to increase the efficiency and maneuverability. Because a single aircraft may assume many distinct operating states over a typical mission (loitering, climbing, cruising, etc.), and because a fixed (non-reconfigurable) system is not capable of providing optimal performance over multiple different operating states, the system is a good candidate for reconfigurability. The performance benefits of changeable wing shape have been explored at least since the beginning of manned flight (the Wright Flyer in 1903), and have seen continued application in many other commercial aircraft to date (e.g., the Grumman F-14 Tomcat, and the Rockwell B-1 Lancer) [7]. Another application of reconfigurability is modular robotics. The goal of the work in this domain is to develop a versatile, cheap, and controllable robot made up of self-similar modules. PolyBot (Fig. 1) is one example of research aimed at achieving this goal. Developed at the Xerox Palo Alto Research Center (PARC), Polybot is a composed of many robotic units that function as actuated hinges in the assembled design. By altering the connections between units, the robot reconfigures itself into different structures, each capable of a different gait. Attainable structures include a four-legged walking robot, a snake-like chain capable of “slithering,” and a circular structure capable of moving like the tread on a tank [8]. These examples only scratch the surface of the applications of reconfigurability. Many more systems in industry and academia serve as excellent examples of exploitation of reconfigurability. The interested reader is referred to [1] for a more exhaustive list. 2 LITERATURE REVIEW AND BACKGROUND This section is divided into two parts. First, we focus on a review of studies that target design for reconfigurability. Second, we review the basic problem structure in product family design and the insight it offers into the reconfigurability problem. 2.1 Classification of Reconfigurable Systems In the design research community, reconfigurable system design has received considerable attention in recent years. Common issues that are discussed in the literature include: (1) generation/selection of reconfigurable design concepts, (2) estimation of the cost of reconfigurability, (3) classifications of reconfigurable systems, and (4) selection of reconfigurable variables. Research in reconfigurable design concept generation and selection aims to provide quantitative methods to discount inferior design concepts as early as possible. Mattson and Messac distinguish clearly between the terms “concept generation” and “alternative generation”—the prior referring to generation of topologically different designs, and the latter referring to different embodiment within a particular system architecture. In the same work, Mattson et al. introduced the idea of s-Pareto optimality, a set-level parallel of Pareto optimality. The authors show that a design concept can be represented by an attainable Pareto set in the objective space. When the Pareto sets of multiple concepts are combined in the performance space, the non-dominated points on this combined surface form the s-Pareto frontier [9]. Literman, Cormier, and Lewis expand on existing concept analysis methods to target design for reconfigurability, adapting and applying them to generate promising design concepts for reconfigurable systems early in the design process [10]. Classifying the types reconfigurable systems and the underlying motivations for reconfigurability helps to break down design problems into more manageable parts. For example, the designer’s approach in a modularly reconfigurable system being designed for robustness should be very different from the approach used in a continuously reconfigurable system being designed for adaptability. To sharpen the meanings associated with terms like “robustness”, “adaptability”, and “modular reconfigurability”, there was a need for clear classifications. In this vein, Olewnik et al. introduced a hierarchical classification of reconfigurable systems [3], which is illustrated in Fig. 2. The most general class of systems was termed “open”, which is a generalized term used originally by Simpson to describe “systems of industrial products, services, and/or processes that are readily adaptable to changes in their environment...” [11]. The hierarchy of Olewnik et al. differentiates between “flexible” and “modular” systems; system adaptability versus system robustness; and “passive” versus “active” adaptability. Definitions from the paper are included here for completeness. They are taken directly from [3]. Flexible systems: Systems designed to maintain a high level 2 Copyright c © 2014 by ASME