AI and Machine Learning Enabling Manufacturing Process and Supply Chain Transformation

Dr. Larry Sweet, Advanced Robotics for Manufacturing (ARM) Institute

Abstract:

AI and Machine Learning (ML) are accelerating gains in advanced manufacturing, enabling levels of productivity accessible to the diverse eco-system of large to small-sized manufacturers. Combined with advanced robotics, “point of need” manufacturing has the potential to transform the entire logistics supply chain, reduce the size and of vast inventories, and prioritize transportation and skill field technicians for critical items that cannot be manufactured in the field. Advanced robotic manufacturing will provide capabilities for on-site production, repair, refurbishment of parts and consumables, and inspect for defects to capture potential failures before they occur. This presentation highlights two current ARM ((Advanced Robotics for Manufacturing) Institute collaborative team projects with high impact potential using advanced AI and ML as critical enablers.

Ohio State University is leading a team including CapSen Robotics, and Yaskawa to produce heat treated and forged parts with superior strength and durability for replacement of critical structural components, eliminating manufacturing and supply chain lead-times from distant production sites and inventory storage. The robotic process uses incremental steps to 3D scan, heat treat, forge, using computer models to guide each repetition from raw material to final shape. AI-based in-process adaptability adjusts for manufacturing variations associated with low volume, high mix production. This allows the system to learn and improve its process planning, plan tasks and algorithms for generating gripping, path, and process plans. Using lessons learned from feasibility tests at end user site, work is underway to shorten end-to-end cycle times.

GKN Aerospace, Gray Matter Robotics, University of Washington, and EWI are developing uniform work robotic sanding with intra-stage inspection, overcoming limitations in existing systems that have irregular material removal, deep scratches, rework, and unacceptable optical distortion. Uniform work is controlled by robot velocity, spindle RPM, pad pressure distribution, overlap strategy, and lap optimization. Advanced optics measure fine detail scratches in transparent surfaces. Machine Learning using artificial neural networks is critical at each process step, automatically labeling thousands of images across the surface as Good, Marginal, or Bad, for workpieces that are representative of end user production geometries and materials.