An Evolutionary Inspired Framework for Grasp Planning Automation on Sheet-Metal Parts with Multi-type Grippers
Ali Tribaldos, Jicmat Andres
MetadataShow full item record
Industrial robot-mounted grippers are commonly used in the automotive industry to position, manipulate, and immobilize sheet-metal parts during several manufacturing and assembly operations. Nowadays, the use of grippers that use two or more different gripping principles such as magnets, suction cups, and fingers to accomplish the gripping task is common. Automation of the steps involved in the design of these multi-type grippers is essential to the timely introduction of new vehicles to the market. The main reason being that production of the vehicles cannot start before designing, testing, and building the grippers for its components, but the design of the latter cannot start before the design of the former is finalized. The objective of this research is to develop a framework for automating the grasp synthesis of multi-type grippers, a step that has historically been commonly accomplished by experienced designers due to the non-trivial nature of selecting the positions of the gripper elements in a viable configuration. By automating the grasp synthesis step in the design of a robotic gripper, the purpose of this research is to alleviate this problem by cutting a significant part of the design of such grippers. A combination of Siemens NX with Siemens Knowledge Fusion and Microsoft Excel with VBA is used to generate the option grasp space and search for optimal grasps using a variation of the genetic algorithm, respectively. A series of 16 tests conducted in 4 sheet-metal parts was conducted to assess the validity of the approach presented here. In all cases, the algorithm was able to produce grasps that had better overall fitness than the respective designer-produced grasp used as the benchmark. Furthermore, although the runs often produced vastly different configurations, these converged to a similar value in overall fitness. This in turn produced a Pareto front of optimized grasps from which the designer could choose. The main limitation of this approach is that some of the grasps produced lacked adequate stability when compared to the designer-produced grasp. This is planned to be addressed in the future by including a set of two fitness criteria that are biased toward stable grasps.