Abstract: As the use of connected machinery and sensors on manufacturing floors, combined with cloud and edge computing, smart manufacturing AI eco-systems can be built to send early warnings, optimize processes, predictive maintenance and enforce quality control. By collecting the right data, manufacturers can get ready creative with their AI solutions.
Novelis successfully implemented AI solutions in aluminum rolling and recycling industry in the past few years. Combining with Azure cloud service and open-source Python AI package, a novel eco-system which is equipped with auto advanced analytics, user interaction, health monitoring alerting, and smart decision making is building up and making contributions to Novelis Business. This talk covers the innovative things happening in Novelis right now. Multiple successful use cases including MLOps, computer vision-based aluminum recovery solution, machine learning based auto labeling, and etc. will be shared.
MLOps surface quality defect detection is an operational data analysis project aimed at capturing quality defects by analyzing sensor data. The solution is a pilot leveraging a cloud service (Databricks) to automate model maintenance and enhance the scalability of the solution. We designed a novel MLOps architecture which can benefit metal manufacturing industry.
A computer vision-based aluminum recovery solution was proposed for metal cutting optimization and recovery optimization. The challenge of this project was the low resolution of the existing cameras. Upgrading cameras in plants expense extra cost. The proposed solution is a lightweight computational solution suitable for low-resolution cameras. This challenge exists not only in Novelis facilities, but in other metal manufacturing as well. Our solution is an example that can inspire other metal industries to contribute without increasing hardware costs.
Machine learning based auto labeling was developed for alloy research and design. This is an example of how to process experimental data, leverage the domain knowledge of laboratory scientists and help them label samples in a smarter and faster way.
We will share these successful stories to inspire the manufacturing community to better implement AI in manufacturing.
Background Knowledge:
Beginner of the tools, know the basic concepts of machine learning, manufacturing data scientist
Bio: Shanshan is working for Novelis as Lead Data Scientist. Her field focuses on advanced operation data analytics, and AI implementation in aluminum rolling and recycling. Her team is leading AI Eco system build up in Novelis. She got her PhD degree from Missouri S&T, and worked for Center of Intelligent Maintenance Systems with focusing on fault diagnosis, prognosis and predictive maintenance in IIoT systems.
