Manufacturers use in-line ink jet printers to add variable information to the product package after the label has been applied. This allows for the use of pre-printed labels without the need to switch between lot numbers or expiration dates. Unfortunately, the nature of production means this information does not always come out as clearly as it could, as movement on the production line, uneven package surfaces, and other issues can interfere with the printing process.
Ink jet printers work precisely in the way that the name suggests. A jet of ink is fired through a nozzle (or nozzles) onto the product surface. Most expiration dates and lot numbers printed in-line use a dot matrix font, in which evenly-spaced dots give the appearance of a particular character. The problem with dot matrix fonts is that any missed dot caused by a clogged nozzle, low ink, or movement of the target surface can mean the difference between, for example, a Q, O, G, or 0. For critical information such as a lot code, that confusion can be a real problem.
These readability issues can be easily corrected provided they are quickly spotted and addressed. However, speed of production is still the top priority, so a slower, manual inspection of the printed characters is not a realistic solution. As with other print and label inspection applications, vision inspection is the preferred method for spotting incorrect or defective print. It is unfortunate, then, that ink jet printing is one of the most challenging vision applications to perform reliably. This is in part because of the way vision inspection systems read characters and the way dot matrix characters are formed.
Vision inspection systems rely on the ability to look for contrast in an image and highlight that contrast. This highlighted contrast forms a shape which is compared against either the expected shape of a character or every character in a pre-programmed font until it finds the closest match, depending on the specific character inspection being performed. In both cases, however, the system is “reading” the character by deciding what the shape of the contrast most closely resembles. This is essentially how the human brain reads characters as well, although the brain has a much larger database of shapes to pull from.
What this means is that the worse the print quality is, the more difficult it is for the system to know which character it is supposed to be. Compared to the human brain, a vision system is not particularly powerful when it comes to being able to fill in missing information or make the sort of logical leaps humans make when confronted with a misspelled word. Therefore, the threshold for what is and is not acceptable is much lower than it would be for the human brain. Dot matrix characters are low-resolution, necessitating the system to almost literally connect the dots in order to have a better idea of what the character is supposed to be—again, just like the human brain does, but not as effectively.
Unfortunately, vision software has historically struggled with connecting the dots unless the dots are precisely where they are supposed to be. This is usually not the case, leading to frustration on the part of manufacturers when trying to use vision inspection to verify important information printed using ink jet printers. In many cases, inspection systems are either set to pass all but the most egregious errors or simply not used for these applications. Manufacturers instead choose to rely on random sampling to catch anything out of the ordinary. Finding better ways to have vision systems behave more like the human eye and mind is one of the great challenges of vision inspection.
One possible method being investigated involves the addition of a “good enough” setting. For example, in a five-character string, one of those characters may be poor quality. A “good enough” setting would allow the string to pass, but only a set number of times. For less-critical information this makes sense as humans are smart enough to fill in missing gaps on their own. One bad character once in a while is less a sign of a serious problem and more just the reality of printing on high-speed production lines. It is when multiple bad characters show up, or bad characters show up on every package, that there is a serious problem that merits action. Putting a “good enough” setting keeps the system from unnecessarily rejecting products which are, in fact, readable enough for humans.
Another method would involve creating “confusion pairs.” This is similar to a “good enough” setting, but unlike that method the character would need to be accurate up to a point. The similarities between certain characters are close enough that a system can have trouble determining which is which—the most common example would be “O” and “Q” or “O” and “0.” A human reader would likely be able to figure out what the character was supposed to be based on context (i.e., a string of letters would not have a zero in it, and vice-versa), and identifying those potential confusion pairs and instructing the system to work around them works on the same principle.
There are an enormous number of benefits to ink jet printing—it can keep pace with increasing production speeds, it is inexpensive to implement, and it is relatively accurate. But the realities of the modern-day production environment and uneven product surfaces caused by unique package designs mean that the print quality can fluctuate. The key is ensuring these fluctuations in quality are identified and addressed in order to continue delivering high-quality products to consumers without needlessly rejecting products due to the difficulties inherent in inspecting ink jet printing. By continuing to research ways to make vision systems behave more like a human reader, the effectiveness of vision inspection can increase, saving manufacturers money and time better spent elsewhere.
For more information about METTLER TOLEDO, click here.
Follow us on Twitter and Facebook for updates on the latest pharmaceutical and biopharmaceutical manufacturing news!