Paper, before being paper, was first paper pulp. But where pulp becomes paper is far from where pulp is generated. For example, the dough may come from Canada or South America. It is transported around the world by ship and often by rail to reach its final destination. On ships, pulp bales are often exposed to the weather.
A global paper manufacturer wanted to improve the quality of its products by optimizing its production process. At the preliminary stage of paper production, pulp bales are transported to a dissolving tank. The paper manufacturer had been unable to successfully identify each bale by the printed lot number in order to match it to system information and clearly detect its composition. This is essential to enable an individually tailored production process and to achieve the best possible quality of the end product. Stephan Strelen, Managing Director, Strelen Control Systems (Büttelborn, Germany; https://info.strelen.net) explains that since the paper balls come from different places, their composition is slightly different. At all times, he says, the papermaker is unsure of the pulp he is using in the dissolving station. “Dissolving is one of the first steps in the papermaking process,” he says. “You box work like that; there is paper coming out of it, but the process is not ideal. The more you know about the current dough type, the better for the process.”
After being exposed to the elements on a ship and transported by rail to the paper mill, the text on the sides of pulp bales is often in various states. It can be very difficult to identify the lot number to match the system information. Although the quality of the text on the bales is poor, the papermaker still wants to read as much as possible because the numbers he box reading helps him determine what’s inside the bale of paper and where it came from – information that helps optimize the process.
One of the advantages of this application is that the papermaker often uses batches of pulp bales, sometimes eight or 16 in a row. If the inscriptions on a ball can be read, the manufacturer can determine the composition of the paste and the origin of several balls. “They expect maybe 200 or 500 code variations,” says Strelen. He explains that if five digits of an eight-digit number on one ball are readable, and four digits are readable on another, then five again on another, “that information might be sufficient because when compared to a base of data, there is a logical combination possible.”
Recognizing the problem, the paper manufacturer turned to Strelen Control Systems to offer a solution. The job involves reading hard-to-read information, using basic information such as codes to wait on the side of the bale, and sharing this information with the plant’s process control unit so that it can optimize the process, resulting in better paper. manufacturing. Batch numbers printed in clear characters on pulp bales are barely legible in many cases due to unstable background, contamination, poor print quality or damaged packaging caused by water or cracks. Thus, batch numbers could not be detected, which prevented production from adjusting the manufacturing process accordingly. Conventional OCR would have been a first step to automating the process. However, it reaches its limits in this use case since the characters are often too damaged to be detected.
Strelen Control Systems provided the papermaker with an integrated system that uses deep learning technology: Safe-Ident OCR. The solution consists of hardware and software components enclosed in a stainless steel cabinet protected against dust and moisture, making it suitable for warehouse and production environments.
Basel (Ahrensburg, Germany; www.baslerweb.com) acA1920-40gm GigE cameras with Sony IMX249 CMOS sensors that deliver 42 frames per second at 2.3 MPixel resolution record incoming dough balls from all four sides to capture the footprint and send their data to the vision unit machine, which detects the batch number and matches it with the database to ultimately report bale quality details to production.
The industrial PC used in this system is a Neousys Technology Inc. (Northbrook, IL, USA; www.neousys-tech.com) Nuvo-7000LP series fanless computer with 6xGbE, MezIO™ interface and low-profile chassis, featuring an Intel® processor (Santa Clara, CA, USA; www.intel.com) 9th/8th Generation Core™ i7/i5/i3 processor. The machine vision system also uses a Siemens (Munich, Germany; www.siemens.com) SIMATIC S7-1200 PLC.
In addition to cameras and lighting, “The heart of our solution is the [MVTec Software (Munich, Germany; www.mvtec.com)] OCR based on HALCON deep learning,” says Strelen, adding that with the help of neural networks created using deep learning methods, plain text can now be read very reliably. With the help of training data, these classifiers can learn to recognize plain text even under difficult conditions like choppy backgrounds, those with poor print quality, or unusual fonts.
The solution uses Deep OCR from HALCON. This holistic approach to OCR based on deep learning can locate characters regardless of their orientation, font type, and polarity. The ability to automatically group characters allows the identification of whole words. This increases recognition performance since, for example, misinterpretation of characters with similar appearances can be avoided.
Mario Bohnacker, Technical Product Manager HALCON, explains: “HALCON is a comprehensive standard software for machine vision. It serves all industries, with a library used in hundreds of thousands of installations across a wide range of industry sectors. Our goal is to provide a software product that can be used and applied in many different applications and industries.”
Figure 3: Representation of the application at the customer showing a ball passing in front of four cameras. (Photo courtesy of Strelen Control Systems.)
Bohnacker explains that Strelen Control Systems’ Safe-Ident uses HALCON’s Deep OCR functionality. “OCR is a very typical example of industrial machine vision,” he says. “The demand from our various customers for this method is very strong. There are very difficult scenarios for reading text. It is not always printed very clearly on the various packaging or objects. Our goal was to find a solution that essentially works as an image input/text output. He adds that with previous approaches, customers had to select the type of print to read and the type of print style: handwritten, different font types, and so on. With Deep OCR, there is one model, which means there is only one classifier. “Just grab your image and get all the text that was written, regardless of orientation, size or print style,” says Bohnacker.
He continues, “What we provide is a deep learning model where we use images of many different print styles and training quality images to get a very robust algorithm against deviations, against different types of print styles, against maybe missing images parts of text letters We teach our Deep OCR not with singular letters but with words so the Deep OCR model can learn the context of the words so that he can, for example, make the difference between a word and a number.
According to Strelen, this application is very different from the usual image processing and OCR projects. “Usually we do OCR in the food or pharmaceutical industry, where we have very fast processes with many packages traveling on a conveyor and difficult surfaces that reflect light where we have to work with very specific light” , explains Strelen. With the papermaking application, the process is very slow, so the demand on the material was not so critical. Although the process is slow, it is still moving, so the cameras had to have global shutters. When it comes to lighting, many OCR applications require very specific lighting. The surface of the pulp ball is not very reflective, so the application only required bright light. “So the critical components of an image processing project – the optics, the camera and the light – are pretty simple here,” says Strelen. “A bit more critical is the computer that needs to be used,” because these factories are running 24/7, and they don’t want to stop the process.
Another difference is that the paper manufacturer would not want bales of pulp with damaged text codes to be rejected. If he could figure out the lot number from other codes, he could still use the bullets. “For a customer in the pharmaceutical industry, if the reading result indicates a defect, they withdraw the product,” explains Strelen. “In the food industry, it’s a bit different. Usually you have very fast processes; products move quickly past the printer. Sometimes it happens that the text is written in a bad state or we simply cannot read a packet. Some customers do not want this product removed. This client wants it gone. This customer uses this input in his process, but the handling process is not disrupted if a defect is detected because he can take information from other bales. »
According to Strelen, when the company launched the project, Deep OCR was not yet available. “We started with conventional OCR,” he says. “We had a success rate well below 10%.” At this point, Strelen expected the papermaker to explore other solutions. But the customer decided to stay with Strelen Control Systems. He adds that this was the first time the company had an OCR project where the success rate was low, but the client still placed the order. During the project, MVTec HALCONs Deep OCR became available, so Strelen Control Systems switched to this technology. Soon after, the results were much better, prompting the papermaker to add a second line.