Festo Makes Smart Factories a Reality With W&B
The True Cost of Downtime
Machines sometimes fail unexpectedly—it’s inevitable.
But that doesn’t make unplanned breakdowns any less frustrating. The costs can spiral to well over $100,000 per hour for large manufacturers in almost all industrial sectors.
Festo, a global leader in industrial automation, is at the forefront of changing that. By pairing hardware expertise with machine learning, the company is helping its customers improve production reliability, optimize maintenance processes and reduce costly downtime.
From Preventive To Predictive
Traditional factory monitoring relies on routine inspections based on usage or time. While this approach helps with catching issues, it can be resource and time-consuming. Plus, more often than not, a problem might be diagnosed too late to prevent damage.
To anticipate equipment failures before they occur – this is where Festo comes in.
Take for example, pneumatic cylinders. These are devices used in production lines across many industries. They handle a variety of tasks like moving parts around in machines, pressing items, opening and closing doors, and many more. Air should never leak out of these cylinders, and if it does, it creates a subtle hissing sound.
Relying on audio signals, the ML team at Festo is developing robust and scalable algorithms that can identify whether the sounds from the cylinders correspond to leakage-related sound patterns. If a leakage is detected, users are alerted with a notification telling them where and how large the leak is, preventing production downtime and ensuring maintenance can be swiftly carried out. What’s worth mentioning is that this use case is part of KI-MUSIK4.0, a publicly funded project of the German BMBF.
“What we’re working on provides a big convenience factor for our customers. It eliminates the need to dig through a very complicated system,” said Daniel Spies, machine learning engineer at Festo.
A rendering of the AI device on top of a Festo cylinder.
Unlocking New Levels of Efficiency
To help its customers achieve maximum uptime, Festo needs a powerful MLOps platform that can deliver unmatched levels of automation and productivity. With W&B, the platform automatically records all of the hyperparameters and metrics, saving the information to a central location. This frees up the team’s time to focus on iteration and tuning – allowing them to try as many things as they need to improve model performance.
“W&B has saved us so much time and effort by streamlining our workflow and making it easy to compare experiments side-by-side and see the impact of different approaches on our results,” said Daniel.
Additionally, to develop high-performing models, Festo uses W&B Sweeps extensively – but that wasn’t always the case. The team had previously relied on open source tools for hyperparameter optimization. When the solutions proved to be cumbersome and tedious to set up, it became an obvious choice to switch to Sweeps.
“Before W&B, it often took me, on average, eight hours to set up a new experiment and tweak the parameters with new data,” said Daniel. “Using W&B, this time has been reduced drastically to twenty to thirty minutes, including Keras callback and plot logging.”
When it comes to maintaining a smooth, efficient, and scalable ML workflow, Festo relies on W&B Launch. Instead of running complex training and hyperparameter tuning jobs on just one local machine, leveraging Launch, the team can easily launch jobs into any target environment with additional compute resources to accelerate model training. And this is all done from a familiar and easy-to-use W&B interface.
“We’ve integrated W&B into our deployment pipeline using tools like Launch,” said Daniel. “This allows us to quickly spin up various training on GPU machines and scale our experiments.
Factories of the Future
Fewer production errors and less machine downtime are just a few of the reasons why manufacturers are turning to AI to streamline the shop floor. With extensive industry knowledge and robust use of machine learning, Festo is making it possible for any factory to transform into a smart factory.
As Festo continues to shape the future of manufacturing, W&B is crucial to streamlining, orchestrating, and automating their ML workflows at scale. Leveraging W&B, Festo can optimize everything related to their ML operations, free up time to focus on building and experimentation, and dramatically scale up model training with ease.
“What I appreciate the most about W&B is that it is truly an all-in-one solution,” said Daniel. “It works out of the box and has been a huge productivity gain for us.”