Stelco Inc. is a Canadian steel company based in Hamilton, Ontario. Stelco was founded in 1910 from the amalgamation of several smaller firms. They build quality products, and their commitment towards exceeding customer expectations is simply unmatched. Their products are commonly supplied to automotive, construction and energy customers across Canada and the United States.
Pluto7 is an artificial intelligence, Machine learning and Data analytics service company with domain expertise in retail, supply chain, manufacturing, hi-tech, education and healthcare providing innovative and accurate business solutions. Pluto7 possesses unique capabilities for helping organisations grow with the increasing demands on infrastructure, managing growing data archives and helping solve big problems with smart analytics solutions.
Pluto7 is among the top 1% of Google Cloud partners of AI worldwide with the highest number of Google ML specialisations. Having worked with multiple industries, Pluto7 has experience in understanding the needs of the customer leveraging machine learning, artificial intelligence, and analytics on Google Cloud.
Stelco was looking for a ML solution to optimise their manual efforts and minimise wastage. They manufacture steel from iron ore producing the steel sheets of particular thickness. During the production Cambers that are unwanted bends are formed in the steel sheets.
The Pluto7 team recognised two use cases to be solved for the customer. The first one was to identify the severity in the camber occurring in the rolling steel sheets. There could be camber presence or not depending upon the curvature in the steel sheets. The other one was To identify the presence of tail chews that are present at the tail(end) of the steel sheets.
The main goal of this engagement is to demonstrate the capabilities and advantages of Google Cloud Vision API and Video Intelligence API to identify the defects. We built ML models that could identify the severity of camber and tail chew. Every phase went through a validation process by the customer. We were successful in creating an ML model that helped the customer solve their problems.
The data that we collected from the customer was uploaded to the cloud storage bucket. ML is an experimental process, so there were multiple approaches of cleaning and transforming the raw videos. As a result, we built a highly accurate model for use case 1, which is detecting camber severity with an accuracy of 80%. For the second use case (detecting tail chews) we were able to build the ML models with accuracy of 95%.