Elia relies on SAP for many of its core business processes and is constantly looking for new ways to enhance them. Illustrating its commitment to innovative SAP technologies, the firm rolled out Fiori Launchpad and several SAP Fiori apps used at Elia have already been fully mobile-enabled.

A Fiori app called ‘Materials Display’, was already in use to help Elia’s warehouse technicians search for, categorize and find materials. But so many material descriptions and images are tough to keep up to date, and some components might have substitutes or updates that aren’t in the system. Even more, it’s not always easy to figure out the right keywords to use to search for unfamiliar materials in the SAP database.

During a design thinking session, Elia invited Flexso to come up with new ways of improving the materials identification and documentation process. Flexso’s solution was an add-on to the Materials Display app that can accurately recognise components and materials from images taken by employees in the warehouse or the field.

Together, Elia and Flexso took an agile approach to development, creating a machine-learning proof of concept powered by SAP Intelligent Technologies. They trained it using images of Elia materials, enabling it to accurately recognise 187 materials in 3 groups. The proof of concept didn’t need to be capable of identifying a material perfectly on the first try, but it should be within the top five suggestions. The Flexso team delivered a solution that surpasses Elia’s requested inclusion rate.

In the future, the proof of concept will be trained and tested to recognise assembled parts as well, since it’s not often possible for technicians to disassemble a component and take pictures of its individual pieces in the field.

• Faster, more accurate warehouse processes with fewer errors.
• Technicians can easily identify materials by snapping an image of it using their smartphones – in the warehouse or in the field.
• Users can help train the solution by pressing ‘like’ or ‘dislike’, letting it know if the material or component in question was shown in the list of probable items displayed.
• Fewer reference materials are needed in the field.
• More efficient infrastructure repair and service planning.

The possibilities of this image recognition solution extend beyond stock control processes. They can be applied in various ways to solve a wide range of business problems in production, logistics, operations and more.

• Find a material simply by taking a picture with your mobile phone
• Detect damage, quality issues and anomalies
• Take action immediately, e.g. order spare parts at the push of a button

While the applications of this technology are diverse, during the course of this project, we learned some concrete lessons regarding how to take pictures that machine-learning image recognition platforms can work with. By keeping the following photography tips and tricks in mind, it’s possible to maximize the accuracy of the solution.

• Take pictures with the object lying down naturally
• Use the same background for the objects as you would in real-life scenario (if possible)
• No other objects or items should be visible
• Objects shouldn’t be shadowed or over-illuminated
• Use different angles by turning the object, not the background
• Try to take the images from the different sides

We offer these functionalities bundled in a packaged solution, the ‘Flexso Material Wizard’ – find out more . 

Want to add intelligence to your assets or business processes? Get in touch!