"Stevanato Group is now using artificial intelligence (AI) in visual inspection to address the problem of false rejects. In time - says pharma visual inspection product development manager Andrea Sardella - the technology could be used even more widely to classify different categories of rejects or identify foreign particles, providing critical information to improve the production process".
AI enables systems to correlate the actual image of the product under inspection with the content of a huge library of similar images for accepted and rejected products and assign the likelihood of belonging to one of the categories in the library. The pharma industry has been actively studying this technology in the last two years and Stevanato Group is among the forerunners.
“There are two problems,” says Sardella. “First, there are not as many images available related to pharma as there are to everyday objects like cars or dogs - which accelerated the development of software for face recognition, image captioning and autonomous driving. Secondly, the images are part of customers’ process data. We have to guarantee data integrity and security and that they are available on customer permission to the US FDA, EMA and other regulatory agencies for validation purposes.”
With standard visual technology, operators use classical parameters of measurement, using their own experience and taking samples from production. ‘False rejects’ are the inevitable trade-off from the fact that we only have access to a limited set of samples. Sometimes we end up rejecting good product to stay on the conservative side.
AI enables to submit hundreds of thousands of images automatically. In this way, the system gives a richer representation of the defects and a better knowledge base from which to start parameterisation. “We can greatly reduce the false reject rate, without sacrificing the detection rate.”
Stevanato Group is actively supporting a customer to implement a prototype system in-line, while others are comparing the results of AI with specific visual inspection issues on some part of the container or drug. The company itself is experimenting AI in its own production line to reduce false rejects.
Would you like to learn more about the implementation of Artificial Intelligence in visual inspection systems? Then meet Dr Andrea Sardella at PDA Universe of Pre-filled Syringes and Injection Devices (22-23 October, Göteborg) or at 2019 PDA Ireland Visual Inspection Event (22 November, Dublin).