Steel manufacturer in Latin America
Reinforcing safety with machine learning and computer vision.
To ensure safe and efficient production across their steelworks, a leading steel manufacturer in Latin America had set up an extensive video surveillance network, using it to assess and reduce the risk of bottlenecks and dangerous events. However, their system was highly labor-intensive, requiring each camera to be continuously monitored by an individual operator to determine when and where a safety rule was being breached. That amounted to over 700 hours of effort for each camera, each month. What’s more, the inherent challenge of ensuring total focus on the video stream made the process highly prone to human error. The company wanted a more effective solution for preventing avoidable risks and injuries in the mill.
We worked together to develop a first-of-its-kind solution for Latin America: an intelligent plant surveillance framework that uses AI in the form of computer vision and deep learning to automate the process of monitoring and protecting plant operations. With the help of open source computer vision libraries, we created algorithms to detect and characterize any objects moving within forbidden zones in the company’s plant. Coordinating each camera’s field of view also enabled us to determine more safety risks as they were occurring, such as whether a load was above head height. By cropping and categorizing examples of safe behavior, then running iterative training and manually validating the results through the deep learning library, our solution acquired the ability to spot a much broader array of potential safety infringements and case-specific rule breaches, and then generate automatic alerts to the company’s safety system.
Thanks to the power of computer vision and deep learning, together with their employees’ know-how, this steel manufacturer is making a real difference to the safety of mill operations. Our AI surveillance solution is detecting potential infringements with an accuracy of over 80 percent. Within several days of going live, the number of high-risk events was already on a clear downward trend. Thanks to the solution’s data collection and analytics capabilities, the company is now able to target their safety awareness initiatives far more effectively, using data-driven insights from actual operations. It’s a major step forward for AI-powered safety in the industrial workplace.