Linatronic AI: detection good and proper at Bitburger - Krones

Sherlock Holmes is without a doubt considered the world’s most famous master detective and to this very day adored as the epitome of an analytical-rational mind. He is known for his precise evaluation of data and meticulous inspection of the crime scene. And while it’s true that the new Linatronic AI inspector does not solve any criminal cases, its methods are remarkably similar to those of the celebrated mastermind. Thanks to its intelligence and a comprehensive collection of data, this machine is currently the beverage industry’s most accurate inspection unit.

The Linatronic AI handles its tasks to consummate perfection. It can be relied upon to detect whether bottles are soiled, exhibit chipping or cracks and/or contain any residual liquid or caustic. Thanks to its artificial neural network, the machine accurately detects material damage and contamination and permits only flawless bottles to pass, thus ensuring better quality and reducing resource consumption.

One of Sherlock Holmes’ distinguishing features is his high innate intelligence. The Linatronic AI, by contrast, has been systematically trained to proceed intelligently. Two partners were of crucial importance in that process: Krones and the Bitburger Brewery. But let’s start at the beginning, just as every careful investigation does.

The Linatronic AI has been up and running successfully at Bitburger since 2022.
The brewery fills four million bottles every day, 90 per cent of them returnables.

Rigorous use of the returnable-container system is one of the essential elements in the brewery’s sustainability strategy. Dependable inspection technology is of crucial importance for ensuring flawless quality of the bottles put into circulation. “That is why we were keenly interested when Krones approached us in 2020, offering to collaborate on developing the Linatronic AI inspector with artificial intelligence,” says Dr. Hankes.

Partners in inspection excellence

He goes on to explain: “We’re always receptive to innovative ideas in order to optimise our processes and steadily upgrade our quality levels.” Needless to say, the long-standing, mutually supportive relationship with Krones also played a major role. After all, the two companies have been working together as partners since the 1970s.

The timing was also spot-on. An inspection unit, which had in 2007 been installed in a Krones turnkey line filling 50,000 0.33-litre glass bottles per hour, was past its prime and had anyway been earmarked for replacement. And this is where the Linatronic AI came in.

Rigorous use of the returnable-container system is one of the essential elements in our sustainability strategy. That is why we were keenly interested when Krones approached us in 2020, offering to collaborate on developing the Linatronic AI inspector with artificial intelligence. Erwin Hächl Dr. Johannes HankesHead of Technology and Sustainability at Bitburger

From a rural brewery to one of Germany’s premier privately owned breweries

Founded back in 1817 in Germany’s South Eifel region, the brewery, which is meanwhile managed by the seventh generation, can look back on more than 200 years of family tradition. Today, it ranks among Germany’s most modern and prestigious privately owned breweries. The brewing group owns the brands Bitburger, König Pilsener, Köstritzer and Licher, and has also agreed a sales partnership with Benediktiner Weißbräu GmbH. It has a payroll totalling roughly 1,500 people. The present-day facility was built on a greenfield site in South Bitburg in the early 1970s and has been steadily expanded ever since. The brewery sources its water, which has rested at a depth of 300 metres for over 10,000 years, from a network of deep wells. That precious water is high in minerals.

Better than Sherlock Holmes

Bitburger opted for a twin-track strategy to start with. “That means we initially used the Linatronic with its traditional functions while at the same time preparing it for AI applications,” explains Bernd Köhl, Head of Electrical Maintenance, who has been with Bitburger for 40 years and understands the inspection processes needed in the brewery like no other.

In actual reality, preparing the Linatronic for AI meant to first of all train it with tens of thousands of photos. And just as Sherlock Holmes can always count on his friend John Watson for help in solving his cases, the Linatronic AI likewise got some hands-on support here: The inspection technology team at Krones examined every single photo down to the tiniest detail and assessed their findings. To give you an example: They’ll look at a photo and determine whether what they’re seeing is a bit of foam or a dirt stain? If it’s the first, the bottle can be passed on to the filler, and if it’s the second it must be rejected. In tech talk, assigning a specific information to a photo is called annotation. The aim here was to accurately record each and every potential fault or defect by means of photos.

A high-frequency measuring unit and an infrared sensor are used to check whether any residual liquid is left in the bottles, for example.

Krones’ team of AI experts developed an artificial neural network (ANN) based on these annotated photos. An ANN is a computer model that tries to emulate the neuronal structure of the human brain, which consists of billions of interconnected nerve cells that process electrical signals. The artificial neural network aims at detecting patterns in data and deriving conclusions from them. Each artificial neuron (or “node”) constitutes a mathematical formula, which processes inputs and generates outputs. In Bitburger’s case, the inputs were the annotated photos, which were passed through the various layers of the neural network. The Linatronic AI is trained using “Supervised Learning” where the algorithm is given labelled datasets that contain a matching output for every input. For Bitburger’s Linatronic AI, that means that every photo has been assigned to a specific fault or defect.

Defective bottles caught in the act

During the training phase, the artificial neural network was repeatedly uploaded onto the empty-bottle inspector and tested, and appropriately extended and optimised. “Cooperation went without a hitch,” says Bernd Köhl, looking back. However, true to the maxim of “Trust but verify”, the partners had the Research and Teaching Institute for Brewing in Berlin (in German abbreviated to VLB), an independent research body of the brewing, malting and beverage industries, perform the crucial validation and acceptance test on this neural network in November 2022.

After the Linatronic AI had passed this acceptance test, it has been up and running successfully with its AI functions at Bitburger. As every inspection unit, it is located directly upstream of the filler and consists of several inspection areas.

To start with, a camera checks the bottles’ contour, height and colour while also detecting any scuffing.
That is followed by the first round of rejections in the Ecopush unit, which passes any defective bottles to a bottle crusher. Bitburger uses discharge chutes for this purpose, which have been specially sealed to prevent the production flow being contaminated by shards of glass from falling bottles.
The first sidewall inspection unit uses a special mirror assembly to check the front and back (together 180 degrees) of the bottles’ sidewalls, at right angles to the direction of transport
High-frequency and infrared sensors are used to check whether the bottles contain any caustic or any other residual liquid.
The bottles are taken off the conveyor in the belt station and clamped sideways for further transport. It is here that sealing surface and base are inspected, and the bottom of the neck finish checked for traces of rust.
Once the bottles have been turned by 90 degrees, the 180 degrees of their sidewalls that have not yet been checked are now inspected, thus guaranteeing 360-degree sidewall inspection.
Finally, it is the Ecopush rejection unit’s turn again.

Smart Sherlock Holmes is invariably convinced that he’s doing the right thing, as is the team in Bitburg. Still, they rely on additional tests at regular intervals to assure top quality levels. Each time 50,000 bottles have been inspected, the team performs a meticulous check to make sure that all fault/defect patterns have been accurately detected. For this purpose, they set a trap for the Linatronic, so to speak, by preparing special test bottles with certain faults and marking them with a QR code so that the empty-bottle inspector can neatly assign and log the faults or defects involved. Such test bottles are fed into the process in order to check whether the Linatronic AI can be relied upon to detect them. Only if the machine has successfully passed this test by identifying and rejecting these special test bottles can normal production be resumed.

Bernd Köhl, Head of Electrical Maintenance, seen at the display checking whether the Linatronic AI reports any faults or defects. Potential irregularities are recorded on photos and can be assessed at the display.

Chief inspector in bottling technology

Inspection accuracy has been significantly improved ever since the Linatronic AI was put into operation at Bitburger. “The crucial advantage is that we now reject only those bottles that must definitely be rejected,” emphasises Bernd Köhl. Previous inspection units also removed bottles from the production process that were in fact still fit for further use. So the Linatronic AI contributes substantially to boosting sustainability at Bitburger. What’s more, it has reduced the operators’ workload since re-adjustments during inspection are no longer required.

The inspector can be relied upon to reject any defective bottles, which are at Bitburger passed into specially sealed discharge chutes to prevent the production flow being contaminated by shards of glass from falling bottles.

The crucial advantage is that we now reject only those bottles that must definitely be rejected. Erwin Hächl Bernd KöhlHead of Electrical Maintenance at Bitburger

The inspection process is running smoothly and consistently, and the team at Bitburger have become fully aware of the benefits provided by AI-based technology, not least when compared to X-ray-based inspection systems. Linatronic AI inspectors have meanwhile been installed in three of the brewery’s eight bottling lines and are earmarked for future lines, too. So the Sherlock Holmes in inspection technology will yet have to solve many cases in future.