Deep Learning

A lesson for Fill machines

How Manuel Hofbauer, as part of his university thesis, taught our Fill camera systems to intuitively look for production faults by themselves.

There are some things where you cannot accept any compromises. Truck brakes are a good example. They must brake, but never break. Or mica panels which protect batteries in the event of an accident and give drivers of electric vehicles valuable seconds to get out of their vehicle before it explodes. Or steering columns. Because it must always be possible to steer vehicles. 

But what if a system basically works properly, yet produces faulty – safety-relevant – components for inexplicable reasons? What if the results are unsatisfactory in spite of all efforts? Well, somehow you have to teach the machine the difference between right and wrong.

And that was the subject of the university thesis of Manuel Hofbauer, who used this as a case study and developed a Deep Learning technology for Fill.

“In my university thesis, I was confronted with a genuine customer problem and had to find a solution. I solved the problem by developing a Deep Learning technology, and the customer was delighted.”
//Manuel Hofbauer, software engineer for industrial image processing at Fill

Manuel Hofbauer applied plenty of creativity to solving this challenge, alongside the courage to try something new. Fill, the university and other technology partners gave Manuel lots of freedom – conventional paths had previously not led to the desired result, so it was time for new methods. Manual also greatly appreciated the active support provided by his colleagues in the software and sales teams at Fill while he was preparing his university thesis.  

The aim was for the machine to learn how to find faults and to distinguish between different types of fault. The decisive feature was the quality of the relevant component: The inspection technology is trained to detect the relevant criteria so that the machine can “learn more” step by step.

Another project involved a two-week trial phase during which components which simultaneously checked by people and by a machine equipped with Deep Learning technology. The result was impressive: 99.99% concordance. Moreover, the automatic inspection technology even identified four faults that were missed by the human inspectors.

The technology gets better with every experience and after a short time can detect even the slightest displacements of a component and classify characteristics, such as scratches or dents, by itself.

The superiority of the machine over humans is clear: Its camera systems are faster, more precise, and above all more robust in detecting faults – it never has a bad day. A human checking the same component day in, day out, while never finding a fault, runs the risk of becoming careless – which can have fatal consequences.

The inspection technology for safety-critical components developed by Manuel Hofbauer makes a major contribution to product safety. Production faults can be identified and rejected, thereby saving human lives in extreme cases. This ensures that truck brakes do not break, steering columns can make the vehicle controllable, and mica panels are able to save lives in an emergency. Reliable components can therefore be produced, which also means security for our customers. Because rejected batches can harbor significant economic risks.

“Currently customers are very curious about our developments. As regards technology, we need to start from scratch with every customer and every new project. But thanks to the university thesis, our company already has in-house knowledge of how to detect faults and repair them. Now we are integrating this knowledge into the various projects.”

//Manuel Hofbauer, software engineer for industrial image processing at Fill

Brave people who are willing to go off the beaten track for their ideas can achieve great things, if they are given the space to experiment. Make space to find solutions and you will find ideas.

“My goal is to integrate robots into a system, which are able to perceive, communicate, and work autonomously – like WALL-E, the robot who cleans away the world’s waste. That's my idea of Industry 4.0.“

//Manuel Hofbauer, software engineer for industrial image processing at Fill

More Fill-Stories