Zeus is a fluoropolymer extrusion manufacturer whose tubing goes into catheters and other life-saving medical devices. On a mission to provide solutions, enable innovation, and enhance lives, Zeus leverages analytics and AI to empower that story. Before Dataiku, Zeus had limited insights from their manufacturing data — multiple millions of rows generated per day — because they had a limited number of data scientists and were limited to code-only tools.
The analytics and AI team has changed this paradigm with Dataiku Cloud, driving adoption of and value with analytics initiatives. Today, they have enhanced collaboration and visibility on enterprise-wide analytics projects, notably across the entire manufacturing value chain — from process engineers to operators and plant managers.
It’s the first time in Zeus’s history that there has been so much cross-collaboration between the data team and the manufacturing teams.
The Road to Value: Yield Improvement & Inventory Optimization
Zeus’ polymer tubing is used in critical medical applications and undergoes rigorous quality testing. Because of the myriad of variables that can affect the manufacturing process, produce yields can often result in scrap.
With a limited number of data scientists (and, again, millions of rows of manufacturing operations data per day), they knew they needed to find a solid partner to help with a solution.
With Dataiku as their partner, the team set out to reduce scrap and improve their yield. They are using automated data generated from their machines and SCADA system, which will eventually recommend improved machine settings for each product in phase II, thus reducing imperfections.
Further, using Dataiku Cloud means that Dataiku handles the administration and management of the platform, so the data scientist could focus on use case development.
Zeus is committed to involving all relevant stakeholders and making analytics an enterprise-wide activity. This use case impacts many team members in the manufacturing process and involves:
- The operators responsible for making the tubing. They will become more efficient as real-time information is shared between the shop floor and the data team.
- The plant manager will then want to incorporate this new data model into the process to make this part number run more efficiently.
- The automation engineering team who will help provide the proper software that brings the sensor data needed for the model.
- The floor supervisor who will review and advise if they also agree with the optimized model.
- Plant operations, who will sign off on the model to ensure proper process and documentation are completed and become the norm.
Historically, it took Zeus approximately 16-20 weeks (that’s 4-5 months) to get data science models into production. Now, with Dataiku, they have shortened that process to a few short weeks. Before Dataiku, the corporate materials management team at Zeus worked with the inventory planning team on a manual solution for inventory optimization. Thanks to Dataiku, they have built a model that runs in near real time (every 15-30 minutes) — but that’s the easy part.
The more difficult part of the process is when a new order comes in that matches the physical dimensions and characteristics of an item currently in stock, apart from a variable such as length or how the material is packaged. With Dataiku, Zeus can now identify which incoming open jobs could be matched with on-hand inventory across all Zeus sites. This allows materials management to reduce inventory and quickly fill new orders, saving process time and money.
Zeus estimates that pure inventory (meaning just jobs that don’t require the modifications outlined above) is the first phase of improved process efficiencies and speed to customers. These metrics will be even more significant once modification jobs have been quantified.
Maturing the Analytics Practice With Dataiku
For the yield improvement and inventory optimization use cases, Zeus has data inputs and data outputs (which come from blending time series process data and operator data) to get into Dataiku to model, take the model output, and present it to the right stakeholders on the manufacturing floor. The output can’t be a complex data science model — the analytics and AI team needs to be able to present the findings in non-technical terms back to the machine operators, which they were able to do with dashboarding and data visualization in Dataiku Cloud along with their BI platform.
Other Dataiku difference makers for Zeus include:
- What-if analysis enables technical users and business stakeholders to gain a deeper understanding of model behavior. This came into play with Zeus operators and automation engineers (accustomed to legacy processes), helping them understand and trust the model, and enhancing overall explainability.
- Fast project iterations in Dataiku (i.e., starting on one machine and scaling out to different machines and sites) are something Zeus is very keen on exploring in the future.
- Data preparation and cleaning, specifically statistical analysis and distributions.
The Future of Analytics and AI at Zeus
Looking ahead, the analytics and AI team at Zeus is excited about continuing to sharpen analytics fundamentals and optimizing their MLOps process. Getting more buy-in and feedback from end users around the company (such as plant managers) is also a top priority. Further down the line, the team wants to make actual data democratization a reality: they want to be embedded in every single department (from marketing and demand planning to workforce analytics and supply chain), helping them deploy models that improve their overall job efficiency. Finally, once a solid foundation is in place, they will eventually explore Generative AI and Large Language Models.
