Automating product defect detection is possible with AI-based visual inspection - Vinodh Venkatesan, Co-founder and COO, Jidoka Technologies

Gartner conducted an online survey from October through December 2021 among 699 respondents in the US, Germany, and the UK at organizations that deployed AI or intend to deploy AI within three years. It found that 80% of executives think automation can be applied to any business decision. AI is already becoming a part of their automation strategies and supports decision-making across business processes. One-third of organizations that are applying AI across several business units are creating a highly competitive differentiator to stay ahead of the curve.

AI is being applied in almost every industry vertical including manufacturing and production too, enabling the workforce to work smarter, not harder, and utilize the time that is saved for higher-level, strategic tasks. AI is leveraged especially in visual quality inspection of products on the assembly line which is done to establish quality control. Manufacturers have placed visual inspection on high priority as they can avoid additional costs due to errors, reduce scrapped items and arrest customer churn. In a production setting, the Quality Inspectors check for product quality such as surface inspection of defects of parts or contaminants, product orientation, product sorting or classification, and product completeness. Superior automation processes are already replacing the human workforce because the former is economical, offers speed and consistency, and is more accurate, thereby saving significant time and effort. 

Limitations of manual quality inspection

Manufacturers consider investing in AI-based automatic quality inspection after they start witnessing the challenges in manual inspection. This includes inconsistent output, where one manual inspector differs from another as each one’s knowledge, understanding, and experiences are subjective leading to the differences in their visual perception and detection of defects. The tribal knowledge usually stays with the concerned quality inspector and many a time not documented for reference or future use. Manual inspection is also not cost-effective as a significant number of workers are required to be deployed for visual inspection with organizations scaling up rapidly post covid.

According to Mckinsey’s report, by employing advanced image recognition techniques for visual inspection and fault detection, a productivity increase of up to 50% is possible. Image recognition may increase defect detection rates by up to 90% as compared to human inspection. With constantly evolving product changes, the traditional inspection methodology requires frequent reconfiguration. The delays due to time-consuming manual inspection cause bottlenecks on the assembly line.  Moreover, according to research, manual visual inspection error typically ranges from 20% to 30%. Through frequent training and practice, these errors can be reduced but cannot be eliminated.

False negatives which are the result of the quality control engineers missing an existing defect or false positives, which occur due to incorrectly identifying a defect that does not exist are the two kinds of visual inspection errors.  These errors reduce the quality of the product or lead to an increase in production costs and wastage.  Customer complaints, returns, and replacements also cannot be discounted here.

Automating manual visual-inspection tasks

Now, how can manufacturers smartly detect defects on products before they reach the next process or end customers?   It is here AI comes into the picture and is today more than just a buzzword.  The AI-based defect detection solution is a form of visual inspection technology based on deep learning and computer vision.

It is the advances in AI that are making it possible to automate several visual inspection tasks.  Furthermore, there is no fatigue factor setting in with this technology as it works 24X7.  It does not call in sick or get distracted.  Deep learning and Machine vision help in building smart systems to perform a complete quality check, down to the granular level.  Automating the visual inspection process with a limited number of equipment and making it smarter with deep learning is possible with this new-age method for visual quality inspection, which is AI-based.  For the processing of data and developing an AI-based model, computer vision and image processing are done.  Computer vision recognizes the object by receiving an image as an input and producing an output.  In case of blurred or noisy images, image processing is done to deblur it and give visible clarity to the image and later be used as an input for computer vision.  Image processing removes noise, filters, detects the edges, and processes the colors.  The output image is used by a data scientist to label and train the machine through computer vision algorithms in AI and machine learning.  It helps to recognize certain patterns and stores the processed data which can be later used for predicting the results for real use-cases. 

Technologies for automation of visual inspection

Deep learning leverages computer vision to analyze various kinds of data sets.  Patterns are recognized to understand visual data which powers many images for machine learning, which is a subset of AI.  Deep learning is a subset of machine learning and uses neural networks that are capable of mimicking the human brain or intelligence.  It recognizes anomalies, distinguishes parts and characters, and endures natural variations.  All complex patterns are interpreted by aligning human inspection with a computerized system.

Automating machine vision helps in delivering at high speed, and is precise, standardized, and reliable as it follows the provided instructions.  Furthermore, it can be deployed in any environment, even where it is risky for humans to operate.  However, by itself, it fails to assess any change between similar-looking images.  This challenge can be addressed by combining it with deep-leaning based systems for accurate visual inspection. 

Building an automated visual inspection system involves the installation of cameras, capturing, storing, and annotating images.  This will have to be followed by the training, deployment, and validating models.  Then a dashboard has to be established which can be used to evaluate the inspection results by the team that controls the process. In recent times AI + machine vision has been industrialized to work in tough manufacturing environments 24X7.

Automated visual inspection is being successfully leveraged across industry verticals – Automotive, FMCG, Pharma, Healthcare, Textiles, Computer Equipment Manufacturing, and Printing among others.