Image recognition as an additional sense in data collection and analysis
Inspecting a production line with an iPhone, a warehouse with a robot or a bridge with a drone? Developers can now provide applications with an off-the-shelf module for image recognition: IBM Maximo Visual Inspection. This technology contains AI that analyzes camera images to detect maintenance issues or control self-driving vehicles, for example. How does this module work in practice?
An additional sense
This is the first time that developers can have a framed module of image recognition software with AI as an integral part. This makes it possible to provide solutions with an additional sense that not only identifies, but also analyzes. In this article, Ronald Teijken, Business Partner Manager Benelux for IBM MAS and Damiaan Zwietering, IBM Developer Advocate and Datascientist, shed their light on the possibilities of this module using some practical examples.
Grown into mature technology
"The image recognition technology behind Visual Inspection is not new. We have years of practical experience with this at IBM" says Ronald Teijken. "For example, we have previously applied it to allow self-driving vehicles to recognize obstacles and traffic signs. Moreover, the technology has been available to developers as open source software for some time." Meanwhile, the image recognition technology is so mature that IBM has included it under the hood of IBM Maximo Application Suite (MAS) and added AI as an integral part to it. On the platform, developers can include this combination of image recognition with AI as a separate module in their applications.
End users who get started with a Visual Inspection application start by tagging videos and photos. "For each image of a good situation they put a checkmark, for each image of an undesirable situation they don't," explains Damiaan Zwietering. "In the future, the module itself can distinguish between 'good' and 'wrong'. Users can set themselves when they receive a notification to go and look and which parameters AI also takes into account in its judgment, such as temperature. In addition, they can indicate at what limit value they want to receive a message." In this way, Visual Inspection helps signal better and faster when a current camera image is reason to take action and takes repetitive inspection tasks off the hands of users.
Image recognition in combination with AI can save a lot of time and improve product quality for various industries. Visual Inspection is therefore already used in various ways, from predictive maintenance in infrastructural objects to the inspection of factories and warehouses. To give an example: Toyota uses Maximo Visual Inspection to monitor production lines. How does that work? "An iPhone takes pictures of the production during the day. Via an internal 5G network at the production site, these are transmitted in real time to a data platform for analysis. In this way, Toyota is able to accurately and quickly check thousands of factory parts per day," says Ronald Teijken.
From robot dog to self-propelled ship
Another compelling example is how Boston Dynamics uses Visual Inspection. This tech company has equipped its robot dog Spot with a camera linked to image recognition technology. Like its flesh-and-blood counterparts, Spot is a loyal helper: the robot dog is used to inspect production facilities and warehouses. Especially in locations with security risks, Spot has a lot of added value. "Also at sea, Visual Inspection is already proving its strength," adds Damiaan Zwietering. "The ship Mayflower can sail completely autonomously thanks to this module. The image recognition technology recognizes buoys, other ships, bridges, driftwood, quays and more. Then AI combines the actual images with all kinds of other data to determine the correct speed and direction. Think about the number of knots of other ships nearby, wind speed and water depth."
What is the current water quality in a basin? Damiaan Zwietering: "Water companies use Visual Inspection to constantly keep abreast of the water quality and to be able to identify trends in it. In doing so, AI bases its judgment not only on current camera images, but also on current and historical values of other parameters such as temperature." Denmark's Sund & Baelt is also using the module to perform inspection, but in a slightly different location: the world's second-longest bridge connecting Denmark to Sweden. "At the top of a high bridge pillar, maintenance workers don't get there easily. In addition, working at great heights is time-consuming and not without dangers." Therefore, the bridge is inspected by camera drones flying around the bridge. Visual Inspection analyzes the images and detects defects such as concrete cracks or rust. Given its added value for inspecting infrastructure objects, IBM included Visual Inspection in its new solution IBM Maximo for Civil Infrastructure.
IBM's AI image recognition module is "pretrained. As a result, the output is high-level right from the start. That means developers and data scientists spend less time feeding and training the underlying model until it's mature enough to put into practice. Moreover, the module is self-learning, so the results become more and more accurate the more image data is analyzed. However, 100 percent accuracy is never possible without human adjustment - human expertise is indispensable when fine-tuning the model to remove the last flaws from the output or to perfectly tune the algorithm to the industrial application. That is where a clean task lies for developers and data scientist.