Close
Search
Searching...
No results
    Reference

    Deep learning inspection in the brewery

    You need to accept cookies to use this functionality.
    Gräfliches Hofbrauhaus Freising operates one of the first Linatronic AI empty-bottle inspectors, thus substantially reducing material wastage.
    • The brewery was founded way back in 1160 and has been enjoying cult status to this very day.
    • Tradition and craftsmanship are the hallmarks of the beers brewed by Gräfliches Hofbrauhaus Freising.

    Gräfliches Hofbrauhaus Freising is among the oldest breweries in the world, looking back on more than 850 years of history. Its roots date right back to the Middle Ages, but as far as technology is concerned, the tradition-steeped Bavarian company lives in the here and now, with an eye trained on the future. Case in point: The brewery is among the sector’s early adopters of the deep learning inspection software.

    “A bit doubtful at first, I thought: Is that really necessary?” recalls bottling hall manager Johannes Kagerbauer with a wry grin. When Gräfliches Hofbrauhaus Freising was looking for a new empty-bottle inspector in late 2020, it got a quotation for the Linatronic AI from Krones. Kagerbauer admits it was interesting that this was the first machine of its kind to employ deep learning software, but nothing to get excited about – well, not at first glance anyway. “We’d been using a Linatronic before and had always been satisfied with its performance,” says the bottling hall manager. “And well, to put it simply the inspector takes photos of the bottles, compares brightness levels of pixels and rejects any nonconforming bottles. I did not really see loads of potential for improvement there at first.”

    Fast turnaround tipped the scales

    That impression was to change soon – especially once the machine had been installed. But initially another cogent reason tipped the scales in favour of Krones: Gräfliches Hofbrauhaus Freising planned to also replace the conveyors between bottle washer and inspector and set quite a tight delivery deadline for that. “We conducted the first talks in November 2020. For us, it was important that installation be completed before the start of the peak season in May – and Krones was the only vendor who was able to promise this,” says Kagerbauer.

    Deep learning technology makes it possible: The Linatronic AI can even distinguish water droplets from genuine defects.

    Deep learning technology makes it possible: The Linatronic AI can even distinguish water  droplets from genuine defects.

    Ecopush rejection unit

    3. Sidewall inspection module, in each case with sidewall inspection and foil detection units, plus lateral neck finish and screw thread inspection units

    4. Sidewall inspection module, in each case with sidewall inspection and foil detection units, plus lateral neck finish and screw thread inspection units

    2D-code test bottle detection unit

    Infrared residual liquid detection unit, base inspection unit, base-chipping and foil detection units

    Screw thread inspection unit

    High-frequency residual liquid detection unit

    Sealing surface inspection unit

    2. Sidewall inspection module, in each case with sidewall inspection and foil detection units, plus lateral neck finish and screw thread inspection units

    1. Sidewall inspection module, in each case with sidewall inspection and foil detection units, plus lateral neck finish and screw thread inspection units

    Ecopush rejection unit

    Camera for contour, height and colour (incl. scuffing detection unit)

    He goes on to say it did not deter the brewery that the deep-learning inspector was still a prototype at the time, quite the contrary: “Our old Linatronic was also a prototype back in 1994,” says Kagerbauer. “So we had no qualms about the machine not functioning properly.” But when the Krones service team arrived in Freising with the new equipment, there was nonetheless a brief moment of irritation. “The electronics engineer and the technician tasked with setting the inspector up for operation were both less than 30 years old. That was the first time I had to deal with an on-site team in which everyone was younger than I was,” says Kagerbauer, laughing, and promptly adds: “Cooperation went swimmingly. To tell the truth, we were always in agreement straight away, without endless discussions. And if I had any questions, they were immediately answered with professional expertise.”

    Lower material losses

    The answer to the question of how deep learning is able to improve inspection is given by the machine itself: It ensures a significantly lower proportion of false rejects, which in turn minimises bottle wastage. This is because, unlike conventional inspectors, the Linatronic AI can distinguish objects like water droplets from genuine defects. “That was what really won me over. You see, I can view the error photos at the machine. And so far, I haven’t found a single one where I’d have said: This bottle should actually not have been rejected,” says an appreciative Johannes Kagerbauer. “And when it comes to identifying base chippings, this inspector is also much more accurate and thus instrumental in minimising burst bottles at the filler.”

     

    Empty-bottle inspection employing deep learning: Here’s how it works

    Please activate JavaScript to play this video.

    The empties tagged as defective are rejected onto two conveyor belts, one of which takes damaged or coarsely soiled bottles and the other one bottles that still contain liquids. Since the bottles are rejected standing upright, they can then be checked visually and if necessary fed back into the bottle washer – yet another new measure making for reduced material losses.

    Article 27585
    Thanks to the new inspector, the proportion of false rejects has been significantly reduced.

    Efficiency upgraded

    The Linatronic AI has been integrated into an existing glass line that had so far handled up to 28,000 bottles per hour. Thanks to the new conveyors, the output was increased to 34,000 bottles an hour. Whereas the conveyor belts always used to run at the same speed, they are now automatically controlled to match ongoing demand. Thus, when there is an interruption in the bottle flow between bottle washer and filler caused, for example, by a malfunction or bottle starvation at the bottle washer, the accumulation table is now filled up again in half the time previously required although its length has remained the same. “When everything has been installed and you see this for the first time, that’s most impressive. The conveyors have increased efficiency levels for the entire line,” says a gratified Kagerbauer.

     

    Feedback from the shopfloor: What bottling hall manager Johannes Kagerbauer thinks about the Linatronic AI today

    Please activate JavaScript to play this video.

    Pure enjoyment with an impressive past

    Gräfliches Hofbrauhaus Freising was founded in 1160 and can thus look back on a rich heritage. The tradition-steeped company selects the raw materials for its beers with meticulous care. The water for brewing comes from the brewery’s own spring, the yeast is propagated in-house, and the hops are purchased from the nearby Hallertau region.

    The broad product portfolio reflects not only the company’s love for brewing craftsmanship but also its rich past. The namesake for the Graf Ignaz Premium Pilsner beer, for example, is a forebear of present-day owner Count of Toerring-Jettenbach. And no less than two historical references are associated with the Moy beer – firstly, it remembers Count Moy, who stood at the brewery’s helm in the 19th century, and secondly it pays homage to the 1970s and 1980s, a period during which the brewery sold an eponymous ‘Helles’ that enjoyed cult status far beyond the borders of Freising. The Moy brand was left unused for several decades, but Gräfliches Hofbrauhaus Freising has now breathed new life into it, in – as the brewery’s website describes it – a nod to a time “when everything was much more relaxed and more down-to-earth than it is today.”

    You can easily send a request for a non-binding quotation in our Krones.shop. 

    Request new machine
    kronesEN
    kronesEN
    0
    10
    1