‘Amazon Lookout for Vision’ Detects Defects in Products and Production Processes

By employing a machine learning technique called ‘few-shot learning,’ Amazon Lookout for Vision is able to train a model for a customer using as few as 30 baseline images. Customers can get started quickly using Amazon Lookout for Vision to detect manufacturing and production defects (e.g. cracks, dents, incorrect color, irregular shape) in their products and prevent those costly errors from progressing down the operational line and from ever reaching customers.

Together with Amazon Lookout for Equipment, Amazon Monitron, and AWS Panorama, Amazon Lookout for Vision provides industrial and manufacturing customers with the most comprehensive suite of cloud-to-edge industrial machine learning services available. With Amazon Lookout for Vision, there is no up-front commitment or minimum fee, and customers pay by the hour for their actual usage to train the model and detect anomalies or defects using the service.

In today’s manufacturing industry, production line shutdowns due to missed defects or quality inconsistencies can result in millions of dollars of cost overruns and lost revenue every year. To avoid these expensive issues, industrial companies must maintain constant diligence to ensure quality control. Quality assurance in industrial processes typically requires human inspection, which can be tedious and inconsistent at best, or at worst, infeasible. Computer vision brings the speed and accuracy needed to identify defects consistently; however, implementing traditional computer vision solutions can be complex. Building computer vision models from scratch requires large amounts of carefully labeled images for each element of the manufacturing process. Then, teams of data scientists need to build, train, deploy, monitor, and fine tune computer vision models to analyze each individual phase of the product inspection process. Even small changes in the manufacturing process (e.g. replacing an out of stock component with an equivalent alternative, updates to the product specifications, or a change in lighting) means having to retrain and redeploy the individual model and perhaps other models downstream in the production process, which is tedious, complex, and time consuming. Because of these barriers, computer vision-powered visual anomaly systems remain out of reach for the vast majority of companies.

Amazon Lookout for Vision offers customers a highly accurate, low-cost anomaly detection solution that uses computer vision to process thousands of images an hour to spot defects and anomalies – with no machine learning experience required. Customers send camera images to Amazon Lookout for Vision in real-time to identify anomalies, such as damage to a product’s surface, missing components, and other irregularities in production lines. Utilizing a machine learning technique called few-shot learning (where the machine learning model is able to classify data based on a very small amount of training data), the service needs as few as 30 images of the acceptable and anomalous state as a baseline to begin assessing machine parts or manufactured products. In addition to enabling the service to detect anomalies without large amounts of training data, this capability also allows the service to be adaptable to a wide range of inspection tasks within industrial settings. After analyzing the data, Amazon Lookout for Vision then reports images that differ from the baseline via the service dashboard or the ‘DetectAnomalies real-time API so that appropriate action can be taken. Amazon Lookout for Vision is sophisticated enough to maintain high accuracy with variances in camera angle, pose, and lighting arising from changes in work environments. Customers also have the ability to provide feedback on the results (e.g. whether a prediction correctly identified an anomaly or not), and Lookout for Vision will automatically retrain the underlying model so that the service continuously improves. This feature allows the technology to adapt to changes in the manufacturing process and even understand when variations are permissible or not based on customer feedback. This means that customers can be more nimble and adapt their processes based on competitive advantages or external factors impacting their operations.

“Whether a customer is placing toppings on a frozen pizza or manufacturing finely-calibrated parts for an airplane, what we’ve heard unequivocally is that guaranteeing only high-quality products reach end-users is fundamental to their business. While this may seem obvious, ensuring such quality control in industrial pipelines can in fact be very challenging,” said Swami Sivasubramanian, Vice President of Amazon Machine Learning for AWS. “We’re excited to deliver Amazon Lookout for Vision to customers of all sizes and across all industries to help them quickly and cost effectively detect defects at scale to save time and money while maintaining the quality their consumers rely on – with no machine learning experience required.”

GE Healthcare is a leading global medical technology and digital solutions innovator that enables clinicians to make faster, more informed decisions through intelligent devices, data analytics, applications, and services. “We are excited about the encouraging early results from Amazon Lookout for Vision that will promise to help improve the speed, consistency, and accuracy of detecting product defects across our factories,” said Kozaburo Fujimoto, Operating Officer, General Manager, Manufacturing Division, Plant Manager, GE Healthcare Japan. “As one of the world’s most trusted healthcare companies with more than a century of technological progress and digital innovations, we look forward to capitalizing on the benefits that AWS’s industrial machine learning services will potentially bring to our manufacturing environments.”

Basler is a global manufacturer and solution provider in industrial vision, providing cameras and machine vision systems for applications such as semiconductor inspection, robotics, food inspection, postal sorting, and inspections of printed images. “Reducing faults is one of the most important KPIs to consider for manufacturing companies. Traditional manual inspection is labor intensive and difficult to scale. By using computer vision for quality inspection, this process can be automated and lead to a significant reduction of costs,” said Gerrit Fischer, Head of Marketing for Basler AG. “Basler and Amazon Lookout for Vision provide a very lean architecture to adopt vision based anomaly detection in any manufacturing site. We’re excited to jointly provide our customers with complete vision solutions by combining Basler’s expertise in industrial vision and edge platforms with AWS’s investments in industrial machine learning.”

For more information: aws.amazon.com/lookout-for-vision