Part variations in geometry and surface finishes in manufacturing are a common phenomenon and can pose a challenge in assemblies of these parts with bigger than acceptable variations in traditional manufacturing. To derive a generic, smart and integrated solution in dealing with these variations in autonomous assembling and smart manufacturing is the challenge to be addressed in this research. This is also a desired solution in many manufacturing operations. The common treatment to variations is either reduce the tolerance of variations or to pair those parts within the tolerance limits. In this project, AI techniques and robotic solutions are proposed to advance the state-of-the-art by looking at each individual variations and coming up with a smart manufacturing strategy for the parts concerned. These variations are studied using these techniques for a better performance of assembled systems. The novelty of the proposed research is to use manufacturing context knowledge in deep machine learning using systems such as deep networks R-CNN, VGGNet16, VGGNet19, ResNet, InceptionV3, InceptionV5, Xception, to guide the learning to be more suitable for manufacturing.
The proposed project aims to derive a generic, smart and integrated system in which the variations in the parts manufactured and used in an automobile assembly process can be detected by in-house vision based system and a more tolerant assembly strategy will be generated by the AI algorithm to be developed, and therefore help establish an optimised and intelligent automobile assembly line. The proposed project objectives are: 1) recognise parts of automobiles by deep learning networks; 2) classify the parts according to variations by deep learning networks; 3) generate an optimised algorithm for the best match of variations; 4) make the robotic control programme to drive the cooperative robots in autonomous assembly.
For example, a batch of 100 car doors manufactured for the assembly can be detected and classified into 10 variation groups by the machine learning algorithm. The doors of the 10 variation groups are autonomously assembled by robots with their best matching car frames as calculated by the AI software. Similar applications can be made into any assembly pairs in an automobile system enabling the optimised performances of the cars manufactured without spending any extra budget.
Funding available for Home/RestofUK/EU students which will cover Tuition Fees and Stipend.