How a paper manufacturer prevents waste and downtime with an Edge AI Solution
Together with hardware partner Advantech and Value Added Reseller HPS Industrial, we developed a valuable edge AI solution for Sappi. Sappi, South African Pulp and Paper Industries Limited, founded in 1936, produces and sells high quality paper products. Sappi is one of the largest paper manufacturers world-wide and operates paper mills across the globe. For this project Scailable, Advantech, and HPS worked together with the skilled Sappi Europe Data Science Squad focused on getting value through computer vision.
High quality paper making consist of several challenging steps. The paper making process at the Sappi Alfeld site—the site we jointly selected to start with our solution—starts at the wood yard. Various types of trees are brought in, carefully inspected, and subsequently “loaded” into the paper making process.
- The solution should run locally (“on the edge”). Most sites are, for security and privacy reasons, not connected to the world wide web. Sending the image streams of the process to a central cloud for processing leads to high networking costs (and risks) involved.
- The solutions should be easy to improve / iterate. The Sappi data science team is actively working on new AI models to identify bark and blockages; as these models evolve the team is looking for a solution that makes it easy to ensure that new models are update on-site and run reliably and effectively.
- The solution should be manageable across sites. The debarking and chipping are integral steps of the paper making process in all mills. However, not all Sappi mills have the same machines, so the solution should scale across sites, with the ability to make minor changes to the models / edge AI setup on specific sites to adapt to the context at hand.
In the end, Sappi is looking for a solution that the Data Science team can independently maintain and improve, which is effective in reducing waste and blockages, that can easily be scaled across all of Sappi’s sites.
For this solution, at the Alfeld site, HPS Industrial supplied an Advantech Industrial PC and a camera. The Advantech IPC has the Scailable AI manager installed, which makes it possible to simply configure the input cameras and run advanced AI models extremely efficiently on the edge device. By virtue of the connection to the Scailable platform the models running on the MIC are easily managed and iterated: thus, the Sappi data science team can deploy their own models to the site, running the actual inferences on the edge. Furthermore, the model output is easily integrated on site to provide feedback to the operator, and the Scailable platform enables effortless scaling across devices on different sites.
We focused on the first problem (the debarking), and jointly installed the selected camera, the IPC, and the Scailable Software. The Sappi data science team trained an AI model specifically for bark detection and used the Scailable platform to configure and deploy the model to the edge device. The solution now provides, at each point in time, a measure of the amount of bark that is left on the trees that leave the debarking: this signal will be communicated to the operator to improve the handling of the machine to reduce waste, and will be logged to Sappi’s data science cloud to gain insights about the debarking process and reduce the waste.
- The solution provides an accurate estimate of the bark on the tree, which allows for optimization of the debarking process by adjusting the debarking speed and throughput. Sappi aims to iteratively reduce the 12% waste to at least 10% waste; the edge AI solution that enables reliable measurements of the amount of bark is an essential building block in this process.
- The solution—in time—should recognize the potential for blockage. Interestingly, given the hardware setup and the Scailable software, the functionality of the solution can be extended independently by the Sappi data science team.
- A reduction in blockage will also improve worker safety. Worker safety is a primary focus for Sappi throughout the paper making process and reducing irregularities in the process inherently improves worker safety.
Benefits with Scailable
Edge AI benefits:
- Edge execution of the AI model reduces latency
- Edge execution of the AI model saves networking costs
- Edge execution of the AI model increases data privacy and security
- Edge execution of the AI model reduces the energy footprint of the solution
- The Scailable AI manager saves valuable on-device engineering time; the edge AI solution can simply be configured.
- The fact that the Scailable AI manager comes pre-installed on Advantech Hardware makes that there is no additional time needed for software staging / installation; this is especially helpfull when scaling accross multiple sites.
- The Scailable platform ensures that the AI models remain Sappi’s IP.