Abstract: This paper utilizes a new generation of computer-aided design methods to conduct research on the design methodology and CAD system of modular robots, aiming to propose ideas and frameworks for intelligent computer-aided design software to solve flexible manufacturing systems. Taking the design of modular robots as a breakthrough, this paper proposes the application of a task-oriented, case-based design method in mechanical conceptual design. The paper introduces the standard modules and basic topological relationships of modular robots, which have developed rapidly in recent years. Based on the characteristics of modular robot conceptual design and combined with the case-based reasoning mechanism in artificial intelligence applications, it proposes a framework for task-oriented and case-based computer-aided design methods and application software, as well as the process of realizing top-down computer reasoning. The paper also introduces the representation of user-oriented robot tasks and working environments. Keywords : robot, module, case-based reasoning, intelligent CAD 1 Introduction Modular thinking has received increasing attention in flexible manufacturing systems. Research institutions in Europe and the United States began research on modular robots in the late 1980s. In the early stages, the focus was mainly on the development of the modules themselves, while recently the focus has shifted to the expansion of application areas of modular robots [1-5]. Research on modular robots can be divided into three different fields: research on modular robot hardware, research on control, and computer-aided design for different applications. Most research to date has focused on the first two fields. Currently, commercially available standard modules (modular joints and modular links) are available. The emergence of modular robots undoubtedly provides more choices for flexible manufacturing systems, but the problems that follow are the ever-changing task objects, different working environments, and the arbitrary combination of modular robots—that is, the infinite combinations of modular robot topology, modular joints, and modular links. Modular robot design has become a challenging topic. The topic of computer-aided design of robots has always attracted attention. In 1986, BO Nnaji published a monograph entitled "Computer-Aided Design, Selection, and Evaluation of Robots" [6]. He indexed and coded the range of motion and speed of the four joints that may make up a robot, and qualitatively or quantitatively (16-indexed) specified 89 parameters, including actuators, joint drive units, joint control units, and design parameters. Nnaji also provided a program flow on how to determine the relevant code according to the design requirements, which pioneered computer-aided design of robots. KH Wurst, while developing modular robots, also provided general principles for selecting modules [1]. The former's research mainly focuses on how general robots determine the code based on design parameters, thereby determining the robot's topological relationships and structural parameters that meet the design requirements. This has certain guiding significance when designing new robots. As a conceptual computer design of modular robots, its guiding ideology differs from Nnaji's design in the following ways: First, the composition of modular robots is subject to certain limitations, namely, the selectivity of finite joint modules and infinite links; second, Nnaji's design scheme is developed for robot design professionals, which requires designers to have in-depth knowledge of mechanism kinematics, dynamics, computer control, and robots. However, the user target of our computer-aided design system is robot users, not robot experts. In other words, our system is user-oriented, not robot designers. From the user's perspective, they do not need to understand the detailed internal structure and operation of the modular robot. They only need to understand and describe the tasks the robot performs and the performance it should have. In this sense, the system is task-driven, or task-oriented. A new difference arising from the above differences is that the computer-aided design system and data structure are different. The computer-aided design system must have sufficient intelligence to perform top-down design. This requires the system to have sufficient depth of knowledge to describe the function, performance, and structure (FBS) of modules and modular robots, the tasks the robot should undertake and its environment, and the knowledge in the task-function-structure mapping process. This requirement for intelligent design places more stringent conditions on the system's data structure. The data structure of general relational databases can no longer meet its design needs. For robot knowledge modeling characterized by object orientation, please refer to reference [7]. 2 Modular Robots The high efficiency, accuracy, and low application cost of dedicated robots have been fully demonstrated in large-scale industrial production. However, in the face of the future demand for flexible production with multiple changes and small batches, the design cycle and manufacturing cost of dedicated robots have become urgent problems to be solved. The introduction of the modular concept into robot design has injected new vitality into flexible manufacturing systems. Selecting appropriate modular robot topology relationships and standard modules to quickly assemble modular robots is an effective way to shorten the robot design cycle and reduce manufacturing costs. Modular robots will become one of the most important devices in future flexible manufacturing systems. 2.1 Standard Modules As the name suggests, modular robots consist of modules—that is, modular joints and modular links. Modules should generally have standardized mechanical and electrical interfaces for inter-module connections. Module joints with one to three degrees of freedom are driven by DC or AC motors and integrate reduction mechanisms and controllers. Module links without degrees of freedom are only used for connections between module joints. Different lengths of module links and different orientations of standard interfaces allow the connections between module joints to meet different kinematic and dynamic requirements of the robot. Figure 1 shows a schematic of a standard module developed by Wurst. A one-degree-of-freedom joint module can be oscillating or translating; a two-degree-of-freedom joint can be rotation and oscillation, translation and rotation, and translation and oscillation. The same type of joint can have different drive mechanisms to adapt to different motion and dynamic requirements, but the options are limited. The length of the joint can be manufactured according to actual requirements. Figure 1 Standard Module 2.2 Modular Robot Topology Theoretically, countless robots with different topological relationships can be constructed using the same type of standard modules. However, from a practical application perspective, a serial robot that meets the six-degree-of-freedom spatial motion requirements (the standard module in Figure 1 is limited to serial robots) consists of no more than four multi-degree-of-freedom joint modules and three link modules. If the three degrees of freedom of the end effector itself are taken into account, the degree of freedom requirement of the manipulator will be reduced further. Figure 2 shows several common serial robot topologies composed of standard modules [1]. The six-degree-of-freedom modular robot shown in Figure 2(a) is the most typical industrial robot topology, which can meet the requirements of most industrial applications. The advantage of this type of robot is that it can avoid obstacles in its workspace, but it is not the optimal topology for some applications. For robots with small actuator motion space requirements, such as assembly robots on assembly lines, the robots shown in Figures 2(b), (d), and (e) are more commonly used. The other robots shown are less commonly used. Figure 2. Topological Relationships of Modular Robots 3. Computer-Aided Design of Modular Robots Computer-aided design of modular robots can follow design flows proposed by Nnaji or other experts. However, using these methods requires the user to be an expert in the field of robotics, possessing expertise in robot kinematics, dynamics, robot control, and familiarity with the structure and performance of existing robot products. This is the main obstacle preventing the widespread adoption of most computer-aided design software, and it is also incompatible with modern conceptual design methods and user- and object-oriented software design philosophies. Our research aims to propose a user-oriented, case-based approach and computer-aided design system based on the characteristics of modular robot design, making computer-aided design of modular robots no longer exclusive to domain experts. 3.1 Characteristics of Modular Robot Design In the case of computer-aided design of modular robots, the end-user's design does not involve structural design of all robot joints and links. Instead, it determines the optimal topological relationships, joint and link parameters of the robot based on a given task, thereby identifying standard modules to assemble a modular robot that meets the task requirements. This is a typical example of conceptual design for mechanical systems. The design philosophy of user-oriented modern software determines that the users of computer-aided design software are end users, not experts in the fields of robotics or computers [8]. In fact, users do not need to be experts in robot design, nor do they need to have an in-depth understanding of the details of robot structure and control. The only thing users care about is to correctly determine the task that the robot wants to complete, describe its working environment, input the functions and performance that the modular robot should have and the limiting constraints on the application interface of the computer-aided design software. As the conclusion of the reasoning of the computer-aided design system, the structure of the robot, namely the topological relationship and module parameters, becomes a new technical solution to meet the requirements of the new task. In other words, the conceptual design of modular robots should be a task-driven, top-down design process. The task that the robot performs determines the functions and performance requirements that the robot should have. It is important to emphasize here that the topological relationship of the robot determines the robot's function, while the joint characteristics, link lengths and masses affect the robot's performance. In other words, when the robot's topological relationship is determined, the robot's function is already determined, and different joint and link parameters only affect the robot's performance. This assumption makes the two-way mapping between task-function-structure of modular robots possible. 3.2 Selection of Intelligent Computer-Aided Design Scheme The development trend of modern computer-aided design is towards software intelligence. Intelligent design software, characterized by user-oriented and object-oriented features, relies on a knowledge base and uses computer reasoning as the main thread. A case-based reasoning (CBR) process is applied to the conceptual design of complex systems, which can closely link the search for new technical solutions with existing successful design cases [9]. As a method similar to the human design process, case-based design effectively utilizes existing successful experiences and greatly shortens the time for seeking the final solution. The advantage of adopting the case-based design idea is that it simplifies the knowledge in the intelligent system, filters out many low-level meta-knowledge, highlights the upper-level knowledge related to the task, and makes the expression, storage, and indexing of knowledge more concise and clear, solving the potential hidden danger of "combinatorial explosion" that may occur in meta-rule-based reasoning. The user-oriented feature of intelligent software is not only the requirement of a user-friendly interface, but more importantly, the users of the software are only general engineering technicians in the field, rather than experts in the field. Task-driven, top-down design should be the main thread of intelligent design, but it is not the only strategy for designing systems. When task-function-structure mapping fails, bottom-up forward reasoning based on meta-knowledge can help generate new robot structures to meet new functional requirements and adapt to new task requirements, which increases the complexity of system knowledge and reasoning mechanisms. Bottom-up design is transparent to end users, who are not required to understand the details of the robot's internal structure. Furthermore, the constantly changing tasks performed by modular robots, their working environments, and the ever-growing composition of modular robots lead to continuous changes and expansion of system knowledge. To eliminate data chaos that could cause system crashes, object-oriented data structures are the only option to solve this potential problem. Studying the relationships between the functions, performance, and structure of modular robot objects is the most important aspect of computer-aided design for modular robots. As a data abstraction example, the encapsulation, inheritance, and overloading characteristics of object class member data and methods allow users to effectively define or develop various complex objects, which is crucial for the definition and application of knowledge, data, and methods involved in large-scale engineering problems. The application of object-oriented design principles to intelligent CAD results in a knowledge representation and organization within the system that differs from that in general rule-based reasoning mechanisms. In summary, based on the characteristics of modular robot conceptual design, a task-driven, object-oriented, and case-based reasoning-based computer-aided design system, employing a top-down reasoning strategy, is the optimal choice for modular robot conceptual design. 4. Modular Robot Conceptual Design CAD System Figure 3 shows a schematic diagram of the CAD system for modular robot conceptual design. Figure 4 shows the flow chart of the case-based reasoning CAD system. Domain experts, acting as the system's designers and maintainers, hierarchically index successful modular robot instances (objects) according to function and performance. This tree-like index is directly used to support the knowledge base for reasoning. Users input the task, working environment, and constraints the robot will perform through a human-machine interface. The task compiler maps the input to indicators of robot function and performance, using them as labels for the inference engine's indexing. The inference engine first performs relevant matching candidate searches in the tree-like knowledge base based on functional requirements. Modular robots that meet basic functional requirements and partially meet performance requirements are considered candidates, while the modular robot with the closest performance is selected. Since the selected modular robot may not meet the performance requirements of the new task, appropriate modifications are inevitable. Because the robot topology that determines the modular robot's function is already determined, the adaptive modifications are simply the selection of appropriate joint and link modules. Forward calculations after changing the module parameters easily determine the new robot's performance; this is essentially an optimization process, aiming to minimize the difference between the robot's function and performance and the function and performance required to complete the new task. The optimal modular robot structure confirmed by the user through simulation will be the system's output and added to the case library. Figure 3 shows the case-based reasoning modular robot assisted design system. Figure 4 shows the case-based reasoning. If no suitable candidate is generated during the matching process, the system first asks the user to revise the task description, such as relaxing constraints or lowering performance requirements, to facilitate the callback of relevant robots. When the system cannot callback relevant modular robots, it will consult domain experts for further knowledge to solve the new task. If the relevant case still cannot be callback after a finite number of iterations, the system calls the synthesis process to synthesize new robot topology relationships from the module library. The system task description interface is shown in Figure 5 (omitted). Users can describe the basic tasks, working environment, and constraints of the robot operation from three attribute interfaces. For example, a robot performs arc welding in a planar space. The welding head (Welder) weighs 3.5 kg, and its maximum working range is within a 500 mm × 450 mm area in the plane. The welding head can deflect in the X-plane, and the trajectory type is continuous. The welding head (Welder) starts from a given point PStart, passes through trajectory Path_1, and ends at point PEnd. The working space can be defined and displayed intuitively in a graphical way, and the trajectory description can be in the form of an array or a graph. The working environment description mainly includes the definition of the system coordinate system positioning, related equipment, sensors, etc. Constraints refer to other constraints on the robot's operation, such as considerations for the maximum speed and acceleration of the actuator, robot positioning accuracy, repeatability accuracy, manufacturing cost, and usage cost. 5 Conclusion This paper aims to apply case-based reasoning to the intelligent design of modular robots based on the characteristics of the conceptual design of modular robots. The intelligence of the auxiliary design system lies in the introduction of object-oriented knowledge representation and knowledge- and case-based reasoning mechanisms. This research was supported by the Hong Kong Government Research Foundation from 1996 to 1999 (Project No. 9040222).