Dr. Tilman Buchner, Director of Engineering, BCG Digital Ventures, gives a glimpse into the future of automation and explains how DV helped build a digital product shadow by leveraging Amazon Edge Computing Technology at re:Invent, Las Vegas.

Today’s industry robots are programmed once and expected to run without any interruptions, hour after hour. The problem? This programming accounts only for the conditions of a perfect world, but in reality, robots are faced with unplanned disturbances. Due to the laws of nature, machine tools can only manufacture within certain tolerances. If the deviations are too large, they default to failure mode, resulting in production interruptions and regular downtimes which have a negative impact on the overall equipment effectiveness (OEE).

In the future of automation, a resilient and self-optimizing production system could react to disturbances and adapt to handling operations automatically. At last week’s annual AWS re:Invent conference in Las Vegas, the IoT engineering team at BCG Digital Ventures joined together with Amazon Web Services, WZL RWTH Aachen and KUKA, to showcase the ways in which Edge Computing can enable self-optimizing robot handling processing in the automotive glazing industry.

IOT robots

“The joined engineering team leverages AWS Edge Computing technology to enable self-optimizing robot handling processing”

Digital Product Shadow

In windshield manufacturing, thermal effects can result in deviations in the glass geometry and disturb the subsequent robot handling processes at the cleaner, primer and quality stations. As a consequence, the production lines need to stop several times per day while technicians re-adjust the robot handling programs manually. This has a negative impact on productivity.

To solve for this, the joint team has developed a digital shadow of the physical windshield. The digital product shadow is a consistent digital representation of the physical workpiece and operates like a living document which reflects all changes on the physical object either in its geometric shape or material property. To create a digital shadow, you need input devices like sensors to measure any deviations in its properties. The digital shadow serves as a single point of truth and moves alongside the physical product through the production system.

At re:Invent, a force sensitive KUKA LBR IIWA robot was used as an input device to measure the 3D shape of the individual glass while cleaning the windshield. The 3D model of the windshield has been saved in the corresponding digital shadow. Handed over to the primer station, the robot applying the adhesion on the windshield can then leverage the product shadow to generate the individual glue path. Disturbances in previous operations (oven process), which had an impact on the geometrical dimensions of the physical product, could be compensated automatically on the fly. As a result, downtimes could be eliminated and the OEE be improved by up to 5 percent [BCG].

“The concept of a digital product shadow reverses the principles of today’s production systems. In the future, it’s not the machine taking the decision about the next step, it’s the digital product shadow asking for resources with dedicated capabilities. The business impact of such cyber-physical systems is tremendous because it offers a new dimension of flexibility which is highly appreciated in word of batch-size one,” said Dr. Tilman Buchner, Director of Engineering, BCGDV.

The benefit goes far beyond the prevention of downtimes – it also offers the ability to adapt the production automatically to new product variants. Time and labor-intensive retooling could now be a thing of the past.

Technology Enabler: Cloud/Edge

To turn this vision into reality, cloud/edge computing technology comes into play. Simple single board computers (e.g. RaspberryPi) attached to industry controllers (PLC, NC, RC) establish connectivity to the physical world. If needed, the Edge can be fitted with additional sensors to capture information which is not provided by existing controllers. In concert with the Cloud, edge devices synchronize the current status of the product shadow with the Cloud. 

The digital product shadow is the missing connector between Information Technology (IT) and Operational Technology (OT). The information exchange between IT and OT has been a well-known challenge in enterprise IT systems for decades. 

“Both worlds are separated by two orders of magnitude in speed. You can operate an ERP system with some seconds of latency, but you need 1ms cycle time or below on the Edge to control industrial processes in “real-time,” explains Dr. Markus Obdenbusch, Chief Engineer at the Laboratory of Machine Tools and Production Engineering (WZL), Aachen Technical University.

Leveraging the Cloud for processing huge amounts of data while having real-time capabilities at the Edge solves three major problems in today’s network architecture: latency, limited bandwidth and intermittent connectivity. The consistency in the flow of information is key to success in a world where data is the new gold. By leveraging cloud/edge technology, it has never been as easy as it is today to exchange information between the management level and the shop-floor and vice versa. You can increase the value of data you captured on the shop-floor if you are able to enrich the data with external information.

Machine Learning

IoT data is often noisy and contains gaps and false readings. AWS’s latest update to Greengrass offers IoT analytics services that filter, process and enrich captured data. Dedicated IoT device management services simplify the provisioning and registry of IoT devices. In combination with built in security features (AWS Device Defender), you can detect drifts in policy and identify anomalies to make sure your IoT devices stay secure.

AWS Greengrass also offers local machine learning inference. At re:Invent an optical quality inspection process was implemented by leveraging the ML framework Apache MXNet. Edge detection technology was used to validate the previous primer process. Hereto the team prepared a training set of pictures from previous primed windshields. Once the algorithm was trained in the Cloud it could be deployed to the Edge and run locally. 

Simply put, the automation architecture of the future will be a hybrid system leveraging cloud services for learning and optimization, but also to correlate field data with external information to generate new insights. In addition, on premise edge devices equipped with the requisite computational power and storage needed to process real-time sensor input will run complex and data intensive machine learning applications on field level.

There is no doubt that Cloud and Edge computing are the enabler technologies to put Industry 4.0 into reality today, rather than tomorrow. The first round of applications from the mining industry (RioTinto), to paper machine manufacturers (Valmet), to automotive suppliers (Denso) showcased in last week’s IoT keynote from Dirk Didascalou (VP IoT, AWS) share a glimpse of the huge business impact across industries.


Our next stop is the Hannover Messe – the world leading trade show for industrial technology from 23rd-27th of April. Stay tuned!