Visual Inspection Combined with Artificial Intelligence: A True Dynamic Duo
The applications of artificial intelligence (AI) are becoming increasingly tangible and profoundly affecting various aspects of our daily lives. Can the potential of neural networks be applied to quality control for pharmaceutical products? What prospects does AI hold in this sector? These are some of the questions Davide Luisari, Design Manager for Visual Inspection at Bonfiglioli Engineering, answered for us.
Artificial intelligence has great potential, but is it truly superior to human capability? In many fields, human abilities still cannot be surpassed. There is a heated debate about the possibility of using neural networks to provide services in complex areas such as education, elder care, and healthcare. Although various experiments are currently underway, these applications are still quite far from becoming a reality. However, in other areas, AI has already proven to be faster, more accurate, and more reliable than humans in performing specific tasks.
For example, the human eye and mind can be deceived by optical illusions that distort our interpretation of what we see. A machine would not make such mistakes and can provide reliable, objective, and repeatable results, even in cases where a human operator might have doubts or uncertainties. The human eye also has anatomical limitations, making it difficult for even skilled operators to detect irregularities measuring 100 microns in diameter. Bonfiglioli Engineering's automatic visual inspection machines can detect much smaller particles, making them significantly more sensitive and accurate.
While artificial intelligence is not "smarter" than humans in absolute terms, it can certainly perform visual inspections more quickly, reliably, and objectively.
Visual Inspection with AI: A Cutting-Edge Technology—How Expensive Is It?
Automatic visual inspection is less expensive than you might think. It is truly advantageous in terms of cost, as this type of technology allows machines to operate at speeds higher than those possible with traditional methods. Automated systems guarantee higher speed performance, with constant accuracy and repeatability. This enhanced speed results in significant economic returns, as well as improved quality.
Bonfiglioli Engineering provides Automatic Visual Inspection (AVI) solutions that incorporate Artificial Intelligence (AI). Each solution is unique and can be custom-built to suit your needs. The system can also be combined with other testing methods to provide a single solution for multiple testing requirements.
What Is Visual Inspection, and How Might Automation Improve It?
Visual inspection assesses the integrity of pharmaceutical packaging, identifying cosmetic and particulate defects. Products that do not comply with expected quality standards are discarded.
Visual inspection utilizes a system of very high-resolution cameras that capture several images of each product for analysis, photographing all the various components of the sample. These images are then inspected for irregularities (such as scratches, foreign body inclusions, or dents) that could compromise packaging integrity and the contents within.
The images can be analyzed by specialized operators or by a computer equipped with a data processing algorithm. Automated control provides significantly better results in terms of testing speed, cost, accuracy, repeatability, and reliability.
领英推荐
All these advantages are further amplified when artificial intelligence is integrated into the capabilities of automated control.
Doesn’t Automated Control Provide Sufficiently Accurate Results? How Could AI Improve the Outcomes?
Automated control uses algorithms to analyze the images collected by the cameras and determine whether a product is compliant. While this method is both efficient and accurate, AI has a unique feature that enhances its competitiveness: machine learning. AI learns from its mistakes and corrects itself, becoming increasingly reliable. Algorithms, in fact, provide results based on parameters established during machine setup. This means that, in unusual situations that don’t match expected case studies, the algorithms may be unable to analyze the problem and simply reject the product, considering it non-compliant since it does not fall within the predefined parameter set.
AI, on the other hand, works from a set of images cataloged as compliant or non-compliant. It can determine whether the product it is analyzing belongs to either category by verifying many parameters in sequence. Artificial intelligence does not rely on a rigid algorithm but instead self-calibrates the parameters it uses to provide results, with no need for operator intervention. AI "learns" and can deliver accurate results even in unexpected situations.
Citations
This article is inspired by information from Bonfiglioli Engineering.