Robotic-based processing is a very attractive option for manufacturers due to the flexibility and reconfigurability of the systems as business and market demands change. Within the robotic space, collaborative robots further extend the flexibility of traditional robotized systems by incorporating highly sensitive internal sensors to better perceive the environment around them. These sensors can be used to provide a source of time-series signals representing the process. This project will look to develop a generalized semi-supervised machine learning-based predictive framework to estimate the surface roughness of a machine finished component to better improve the efficiency of robotic finishing applications. With an accurate predictor of the surface roughness derived from the processing data provided by the collaborative robot, requirements for additional external inspection can be reduced or removed. A fundamental component of the project is to ensure generalization of the framework, allowing for extendibility into other collaborative robotic manufacturing processes.
Within this project there are two main activities, the first major activity is the development of the surface roughness predictive model for the robotic finishing application. Within the micromachining and finishing domain, there has been significant research outlining the contributions of key parameters (normal force, tangential velocity, depth of cut, RPM, etc) on surface roughness. The second major activity is the generation of a pseudo-labeling model that is able to accurately predict the appropriate label/class for new samples based on a reduced training set. This system will take all of the input data and try to match it to a similar previously labeled sample and provide the unlabeled sample with an appropriate label. With this system in place, significant time savings can be had in annotating output data for the development of additional models (new parts, materials, etc).
2020-03-15 till 2023-03-31
© 1878 - Western University
