Why poor quality is a risk for manufacturing companies and needs to be fixed — our hypothesis on EthonAI
🦅💰Welcome to the #EBVCgang
Please join us in welcoming Zurich-based ETH spin-off EthonAI to the Earlybird portfolio. We are excited to co-lead their CHF 6.27m Seed funding round alongside our friends at La Famiglia (Judith Dada, Samuel Beyer), as well as existing investors Wingman Ventures and Acequia Capital.
The founders, Dr Julian Senoner and Dr Bernhard Kratzwald, both PhD graduates from ETH Zurich, have assembled a rockstar team that has built a hardware-independent AI-based quality management solution that detects defects and prevents them from happening in the first place through a best-in-class root-cause analysis. Their product is an easy-to-integrate, plug-and-play solution that works across-industries and use cases, is hardware agnostic, super quick to deploy and can be used by domain experts without data science knowledge.
📊👁 Quality management is crucial to a company’s success
Poor quality products can be detrimental for any company. Especially in the context of industrial manufacturing, where hardware products go through a number of production steps and endure a variety of manufacturing methods, defects can sprout at numerous occasions. Even minor, and often invisible defects, can lead to unserviceable products.
In industrial manufacturing, average defect rates often lie in the 5–8% range (Source); for products in critical infrastructure, that’s often far above (e.g. semiconductors). It’s simple math to derive the immense cost this bears for manufacturers. But the impact of poor quality management goes beyond this. Shipping defective products to customers is also detrimental for their reputation on the market and will make good headlines. Minimizing production waste is also a topic gaining momentum as part of companies’ sustainability agendas.
Additionally, to make optimal decisions on the factory floor, process engineers and managers need full visibility on quality levels and sources of defects from across machines.
📽😿 Defect detection and quality monitoring cannot be performed manually at scale
These insights have been long understood and even spurred concepts such as Six Sigma back in the factories of the 1990s in companies such as Motorola or General Electric, where Jack Welch made the concept central to his overall business strategy.
However, the tools to really identify defects and monitor quality are also still stuck in the past and need a makeover.
Even today, many defect identification processes are still done manually, by human workers sitting behind a screen and looking for defects with the naked eye. Finding a hair-thin scratch on a PCB module, however, sounds a lot like looking for a needle in the haystack. It is prone to error, slow, expensive and inaccurate.
🌊👎 First wave of solutions only works in niche, individual use cases and are hard to integrate and use
The above shortcomings are exactly why we observed a first wave of companies trying to tackle this issue in a more digital way, often using Computer Vision modules to more consistently find errors.
Admittedly, most of these solutions could not live up to their promise due to a variety of limitations. Many solutions required purpose-built hardware such as cameras or specific sensors that first needed to be purchased and then installed on the factory floors. The true adoption killer, however, is the need for these models to be trained and used by data scientists, not the process engineers in the factory halls. Lastly, the majority of detection models follow a supervised learning approach which requires at least one, oftentimes more, observations of a specific defect before it can be automatically identified. This results in a very long time-to-value.
🏄♂️🥇 EthonAI is here to help
Set the stage for EthonAI! During their PhDs at ETH Zurich, the EthonAI founders explored a way to make their models not only much more precise, but also much quicker to deploy and easier to use by non data scientists. Customers can connect the software to whatever existing hardware infrastructure they already have in their factory and train the model to any use case across any industry in a matter of minutes. EthonAI’s novel unsupervised learning approach allows the detection models to be trained in a sleek UI, with about 15–30 images of an intact product. That means no more collecting thousands of observations of defective products such as images of scratches at different positions, or in different forms and sizes. All this can easily be done in their polished no-code product that can be used by process engineers and managers alike.
🎬🔮 What’s left to say
In case you see yourself facing some of the above mentioned issues and are looking for a lean and scalable solution, reach out to the EthonAI team. You will be in good company with firms such as Siemens, Roche, and Lindt, that are already using EthonAI’s software in their factories.
Also, if you are as fascinated by EthonAI’s vision as we are and want to work with them, the team is hiring across different teams, in engineering as well as sales. Find open opportunities here.
We are incredibly excited to be working with EthonAI and supporting them on their path to bringing quality management to the next level.
