This paper addresses failure detection in automated parts assembly, using the force signature captured during the contact phase of the assembly process. We use a supervised learning approach, specifically a Support Vector Machine (SVM), to distinguish between successful and failed assemblies. This paper describes our implementation and experimental results obtained with an electronic assembly application. We also analyze the tradeoff between system accuracy and number of training examples. We show that a less expensive sensor (a single-axis load cell instead of a six-axis force/torque sensor) provides enough information to detect failure. Finally, we use Principal Component Analysis (PCA) to compress the force signature and as a result reduce the number of examples required to train the system.
In this paper, we describe a generative process planning system for robotic sheet metal bending press-brakes. This process planning system employs a distributed planning architecture. Currently, our system consists of a central operation planner and three specialized domain speci c planners: tooling, grasping, and moving. The central operation planner proposes various alternative partial sequences and each specialized planner evaluates them based on its objective function. The central operation planner uses state-space search techniques to optimize the operation sequence. Once a CAD design is given for a new part, the system automatically determines: the operation sequence, the tools and robot grippers needed, the tool layout, the grasp positions, the gage and the robot motion plans for making the part. The distributed architecture allows us to develop an open-architecture environment for doing generative process planning and encapsulate the specialized knowledge in specialized planners.
The effective planning of a product’s manufacture is critical to both its cost and delivery time. Recognition of this importance has motivated over 30 years of research into automated planning systems and generated a large literature covering many different manufacturing technologies. But complete automation has proved difficult in most manufacturing domains. However, as manufacturing hardware has evolved to become more automated and computer aided design software has been developed to support the creation of complex geometries; planning the physical fabrication of a virtual model is still a task that occupies thousands of engineers around the world, every day. We intend for this paper to be useful to newcomers in this field, who are interested in placing the current state-of-the-art in context and identifying open research problems across a range of manufacturing processes. This paper discusses the capabilities, limitations and challenges of automated planning for four manufacturing technologies: machining, sheet metal bending, injection molding, and mechanical assembly. Rather than presenting an exhaustive survey of research in these areas, we focus on identifying the characteristics of the planning task in different domains, current research directions, and open problems in each area. Our key observations are as following. First, the incorporation of AI techniques, geometric modeling, computational geometry, optimization, and physics-based modeling has led to significant advances in the automated planning area. Second, commercial tools are available to aid the manufacturing planning process in most manufacturing domains. Third, manufacturing planning is computationally challenging and still requires significant human input in most manufacturing domains. Fourth, advancement in several emerging areas has the potential to create, in the near future, a step-change in the capabilities of automated planning systems. Finally, we believe that deploying fully automated planning systems can lead to significant productivity benefits.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.