From April 23 through 27, Hannover Messe – the world's leading trade show for industrial technology – is set to showcase the technologies that set the benchmark for modern industry, both in the present day and beyond. With more than 5000 exhibitors and over 500 practical demonstrations, this leading trade show will transform the city of Hanover into a global hotspot for Industry 4.0. Among the major highlights of the event will be the exhibits on production-based applications for artificial intelligence, including our deep-learning-based component identification solution, which relies on image recognition technology. This system simplifies maintenance and damage detection in complex machine structures, significantly reduces service costs and enables anyone to create their own machine learning model. Trade show attendees can visit our exhibit at the HPE stand to learn about and experiment with artificial intelligence – even if they've never used the technology before!
At the 2018 Hannover Messe for Industry, we're enabling visitors to explore our component identification solution, which relies on deep-learning-based image recognition technology:
"Component identification" deep learning showcase
at the Hewlett Packard Enterprise (HPE) stand
stand A38, hall 6
Component identification based on image recognition – made possible by deep learning
Components installed in complex locations can often cause headaches during maintenance work. Individual components may be hidden, or tucked away in places that are difficult to access. In older systems, labels may no longer be legible or may have been removed entirely. To further compound the situation, some components are designed to look virtually identical – even though their functions differ.
At last year's Hannover Messe, we presented a solution that makes maintenance significantly easier: An application for optical component recognition. This solution is an extension of an image recognition process with a pretrained deep neural network core.
The application allows anyone involved in the production process to access information on maintenance and repair by simply photographing the component in question using their mobile device. The application provides a direct link to the required information, without the user needing to enter any type data or machine number. The solution also facilitates damage detection: Damage in the surface structure of components or errors in the analysis of the link between image characteristics in product images and sales data can be confirmed on the basis of a simple photograph.
Deep Learning case: component identification via image recognition 2.0
In 2018, we're going one step further. Now, a simple iOS smartphone app developed by us, in house, allows users to identify components or tools. For demonstration purposes at the trade show, we've trained a model that can recognize the tools on a pocket knife. Visitors will also get the opportunity to train a brand-new model from scratch. All they need to do is take a number of example images using the iOS app and label these images. The app works in a similar way to solutions that are already on the market.
However, what's new in this app is that we can generate a "heat map" (also known as a class activation map) for each identification. This heat map is based on a whitebox AI. From this heat map, the user can see which parts of the image the neural network has used to identify the object. The map – which closely resembles an infrared heat image – allows you to see why the identification process was successful, or why it failed. With the right tools at your disposal, neural networks need no longer be black boxes.
The process of developing a machine learning model is usually highly complex and can only be completed by specialists with the appropriate training and experience. Our application allows developers to create and train new models even if they have little or no prior knowledge of machine learning. This reduces the high development costs usually associated with such applications and allows the user to produce a launch-ready application quickly and cost-effectively.
Predictive analytics: a competitive advantage
These kinds of machine learning models form the basis for predictive analytics and predictive maintenance. In the process of identifying a component, the technology can also draw conclusions about its functionality and level of wear. In a mathematical model, previous and current data is combined with information from machine learning to produce qualitative predictions. In recent years, increasing numbers of companies have incorporated predictive analytics and predictive maintenance into their workflows, particularly for machines that form part of a production process.
This move benefits companies because predictive analytics eliminates unnecessary maintenance and repairs, saving the associated costs and resources. It also reduces machine and plant downtime. Detecting faults at an early stage lengthens the service life of the machines and enables the operator to perform maintenance when it has the least impact on the production process. A plastics and film manufacturer, for example, could save up to 50,000 Euros a month by using an application for predictive maintenance to reduce downtime and minimize material wastage in production.
We look forward to seeing you at stand A38 in hall 6. Click the button below to arrange a meeting with *um team member Philipp Schlüter:
This may interest you, too:
How to benefit from Artificial Intelligence as a Service – even for beginners