In the 'Emilie' (Embedding Machine Intelligence Logic and IT Security into Edge Devices) research project, Real-Time Systems, Fraunhofer ITWM, Mobotix, Gebr. Pfeiffer, Magnetic Sense and Bremen University of Applied Sciences are jointly developing solutions based on intelligent edge electronics. The aim is to securely and robustly record data on edge devices and process it based on artificial intelligence (AI) in order to be able to use the resulting information to optimally control and monitor production processes in the control cabinet. In 'Emilie', the bridge between OT and IT is secured via an IoT gateway. The information from production and status data collected by various sensors and condensed using AI processes is bundled at this gateway and can be transmitted to the PLC, for example, as an additional input variable.

Smart sensor technology and AI for more efficient production

The potential of intelligent edge computing is illustrated using two different processes from practice: cement mills from Gebr. Pfeiffer and turbo sets managed by HIMA. The objectives include measurable improvements in throughput, higher energy efficiency and more effective estimation of the remaining wear stock. The integration of AI processes in signal-processing analyses within the OT enables reliable monitoring of operating states with resource-saving predictive maintenance to minimize planned and unplanned outages on the one hand, and automated, condition-based process control of the operating mode of energy-intensive systems in energy-optimized or more efficient areas on the other. In order to ensure the necessary safe, robust and near-real-time monitoring (of vibrations) in the production process at sensor level, the electronics of the following sensors and edge devices are being further developed to carry out intelligent data processing:

• To improve distance and temperature compensation, a magnetostrictive sensor for force measurement from Magnetic Sense is being enhanced. More intelligent algorithms are being developed for the sensor board, which calculate the speed and angle of rotation of the shaft to which the sensor is attached in addition to temperature and torque.
• A high-resolution industrial camera for vibration monitoring from Mobotix is being further developed in terms of hardware and software in order to extract housing vibrations directly from freely configured pixels on the camera and pass on the amplitudes as path signals to each point on a channel basis.
• The IoT gateway 'Arendar' from Real-Time Systems is extended via suitable co-processor cards, which allow multivariate analyses to be bundled via several sensor channels to match the connected sensors, so that comprehensive features on machine dynamics can be offered. On this basis, the gateway forms the foundation for monitoring, diagnostics, forecasting and optimization applications.

The data is analyzed on the edge devices. For more computationally intensive analyses, more powerful industrial PCs can be mounted on the DIN rail in the control cabinet. Due to the increasing requirements for miniaturization and heat transport, computing capacity decreases with proximity to production processes.

Short distances in the switch cabinet

Interaction and cross-cutting issues of intelligent edge computing in production processes: The IOT Gateway enables a kind of galvanic separation between OT and IT.

© Fraunhofer ITWM

In contrast to the heavily advertised IoT cloud solutions, edge computing offers four significant advantages that reduce hardware, effort and risk:

• Short distances: data storage scales even with high sampling rates (in the kHz range), as raw data remains local and is enriched locally by the intelligent data pre-processing units within the sensors and edge gateways or IPC within the control cabinet.
• Computing at the point of action: the events of recorded system situations are determined and evaluated on the edge device. AI processes for condition monitoring, predictive maintenance and control recognize condition patterns in real time and control the systems with the addition of the PLC in a resource-efficient manner within the control cabinet without a direct IT or Internet connection.
• Less communication effort: Edge computing enables the resource-efficient, digital transfer of secure information and relevant histories of detected events to IT without transmitting raw data. Local data processing reduces communication requirements in terms of bandwidth, throughput and response times; real-time system monitoring with available power electronics is simplified.
• Trust creates security: The trustworthiness (trusted computing) of the computing units involved (sensors, gateways, IPCs, IoT cloud) within the processing chain is secured through the use of 'cryptochips'. Based on this trust anchor, platform integrity can be ensured through root-of-trust. In addition, hardware trust anchors make it possible to implement remote attestation principles and thus securely query the integrity of a device remotely. Analyses for condition monitoring are carried out largely autonomously locally on the system so that company secrets are protected.

A view from outside helps

Imaging methods based on high-resolution, high-speed cameras enable the global monitoring of excited vibrations at any point in the system. In contrast to acceleration or displacement sensors, this imaging approach is not used directly on the machine, but at a defined distance. As a result, this approach is less susceptible to difficult environmental conditions and therefore offers an ideal supplement. A Mobotix camera is being further developed as part of the project. It is to be used as an edge device for evaluating the image data. To this end, the camera's hardware modules are being expanded with image processing software and AI functionalities for intelligent data processing. The image processing tool 'ToolIP', which is already available on ARM architectures, will be ported to the camera hardware.

Energy-efficient control

The processes used to produce cement are highly complex and, according to the German Cement Works Association (VDZ), account for 6 to 8% of global CO2 emissions. A medium-sized cement mill requires around 20 GWh of energy per year to grind the cement clinker. There is huge potential for optimization here.

Limestone grinding plants are operated close to limestone quarries, which are often in remote areas. Accordingly, local data processing is essential for effective operational automation. In the 'Emilie' project, Pfeiffer Bros. grinding mills are being expanded to include smart edge devices so that energy-efficient operating points can be achieved automatically. Initial tests are being carried out on a test mill at the technical center in Kaiserslautern.