/f/239067/1920x1080/d0b1e11229/hero.gif)
Visual inspection
Good, better... deeper?
How we’re using Deep Learning algorithms in a GMP regulated environment. Bringing data scientists, machine learning engineers and others together to increase accuracy and efficiency in the visual inspection process.
Patients first
When it comes to patients, quality control is of the utmost importance. For decades, rule-based computer vision detection algorithms were the standard to catch defects in products. In essence, a rule-based algorithm detecting faults based on images. However, recent advances in deep learning technology have presented a unique opportunity to elevate this automated visual inspection process with Deep Learning algorithms.
Recognizing the potential of leveraging deep learning in visual inspection processes, a domain already known for upholding superior product quality, Novo Nordisk is developing a technological solution that could significantly reduce unnecessary waste of good product while maintaining the high detection accuracy for faulty products.
/f/239067/2933x1650/ab946139d4/visual-inspection-deep-learning.jpg)
The shift
The shift from traditional rule-based detection algorithms to more advanced deep learning detection algorithms was not about following trends of AI or machine learning. It was about using technology to drastically improve the product process while adhering to strict regulation for pharma manufacturing i.e. Good Manufacturing Practices (GMP).
This move required a structured approach, and a multi-expertise team comprised of process experts, operators, quality assurance, data scientists and machine learning engineers —individuals committed to pushing boundaries and achieving even better vision inspection.
The technical aspect involved creating deep learning algorithms that were specifically designed to identify defects in glass cartridges filled with insulin. The algorithms outperformed the prior rule-based computer vision approach, significantly reducing the false rejection rate of good products, leading to a notable increase in manufacturing capacity and a decrease in wasted products.
A real leap of faith
The reason it was so challenging was because the pharmaceutical industry, which includes Novo Nordisk, have the most stringent quality control standards, and integrating deep learning detection algorithms required adhering to those principles. Nevertheless, the team embraced the idea, going above and beyond to contribute to the project.
“This was about embarking on a journey, not only adopting new technology but also developing our mindsets together.” Gitte Bjørg Windfeldt, Head of Data Science in DD&IT PS Technology Foundation & Transformation.
The key to successful implementation has been the development of a thorough roadmap for effectively using this technology within Good Manufacturing Practice (GMP). The approach strongly emphasised “Integrating Deep Learning” into established quality management principles to foster a quality-driven mindset and promote organisational comprehension, which should in turn benefit future deep learning projects.
Time to get technical
A huge realisation also became apparent during the project – technical roles such as data scientists and machine learning engineering were no longer just “nice to have” in pharmaceutical manufacturing, they were essential.
Using Deep Learning detection algorithms for visual inspection places Novo Nordisk at the forefront of applying emerging technology to the benefit of both patients and the environment.
/f/239067/1920x1080/a26513bd2e/digital-innovation-hub-hero2.jpg)
/f/239067/1920x1080/35e085cad4/hero_fallback.jpg)
/f/239067/1920x1080/94892ea7ee/hero_fallback.jpg)
Automation Engineer I/II
NNPILP API M1-US
Systems Product & Innovation Analyst
NN Brazil
IT System Manager
NN Pharmatech Koege
Supply Chain Project Manager (Maternity Cover)
NN Bagsvaerd
Novotunity - MI Engineer
NN Hilleroed
Senior Analytical Project Manager
NN Taastrup
IT Tester
NN Service Centre Mexico SA CV
Supply Chain Project Manager
NN Bagsvaerd
CMC Project Director, Early Portfolio
NN Taastrup
Full-Stack Engineer
NN Hilleroed
Business Development and Partnerships Manager
NN Switzerland - 1
Project Execution Model Lead
NN Kalundborg
Submission Specialist
NN Soeborg
Integration Project Manager
NN Anagni Production
Senior IT & Automation Project Manager
NN Anagni Expansion
Advanced Digitalisation Professional
NN Lyngby
Senior Data Engineer
NN GBS Bangalore
Advanced Data Analysis & Insights Professional
NN Gentofte
Finance Solution Architect (maternity cover)
NN Bagsvaerd
Lead solution architect for IT and Automation in API expansion projects
NN Kalundborg
