Edge Cloud Continuum for Production

When used intelligently, production data offers huge potential to optimize industrial manufacturing processes: from automated tool changes in the event of wear and tear, to automated filtering of reject parts, to dynamic parameter setting for greater efficiency and precision. The essential data is provided by tool machines. They continuously generate valuable information about the status of tools or components. The challenge is to collect and process this data efficiently and to gain insights from it.

The research and development project “Edge Cloud Continuum for Production” (ECC4P), developed by the Fraunhofer Cluster of Excellence Cognitive Internet Technologies CCIT, is the first holistic solution to directly address the challenges of automated production processes such as machining (milling, drilling), generating grinding and forming. ECC4P helps manufacturing companies to monitor, manage, document and automate their production processes, from collecting sensitive process data at the local edge device, to AI training in the cloud, to making smart adjustments to on-site processes.

From machine to cloud and back

ECC4P combines the benefits of two technological approaches in a single loop. Processing data in edge devices close to the sensor enables fast response times and local data monitoring, while cloud computing in central data centers provides the scalability needed to carry out complex analyses. Data is automatically processed in the most efficient location.

Smart, wireless sensor systems on the machines collect parameters such as temperature, vibration or pressure. An edge industrial PC merges the collected sensor data and the general machine data and can play a regulating role if necessary. The merged data is then prepared on site for processing by usage-specific machine learning models so it can be utilized further through standardized interfaces. These are automated pipelines developed specially for training and developing usage-specific ML models.

These models are trained in the cloud, where the available processing capacity is greater than on the edge. The trained algorithms are able to identify patterns and deviations, such as production faults, and to make predictions on factors such as tool wear and tear.

Once the training has been completed in the cloud, the algorithms with their decision-making rules and predictive models will be taken from the cloud and once again applied to the edge IPCs. In regular production, the AI evaluates measurements locally, interprets them and automatically adjusts the feed rate or rotation speed on the basis of the information it has learned.

The collected and processed data and models remain under the control of the data owner at all times. Dataspace technologies such as the EDC (Eclipse Dataspace Connector) are used to ensure that the data, once it has been shared with authorized business partners, can only be used in accordance with the agreed data usage rules.

The edge and cloud infrastructures are subject to continuous security audits and vulnerability analyses to ensure that they are in line with security and compliance standards.

A modular, practical approach to production

ECC4P provides the infrastructure necessary for a networked overall architecture while also supplying the components for each individual step, from monitoring to control loop. The modular design means they can be connected to existing structures. These include:

• The edge software infrastructure to enable the collection and synchronization of high-frequency measurement signals in real time.
• The highly sensitive local sensor systems “smartGRIND” (for generating grinding), “smartTOOL” (for milling, drilling and grinding) and “smartNOTCH” (for metalworking in presses). These can communicate with the IPC, the machine controller or separate monitoring systems via a gateway.
• The “LinkedFactory” solution, a flexible data architecture for managing production data relating to products, processes and machines.
• Usage-specific AI algorithms for evaluating data in parallel with processes.
• Eclipse dataspace connectors (EDCs) to transfer selected measurements to the cloud simply and securely while maintaining data sovereignty.
• Automated auditing of security compliance rules on the edge and cloud infrastructures.
• An automated MLOps pipeline that (re)trains, tests and validates usage-specific AI models in the cloud.
• A user-friendly live dashboard to visualize the data.

Michael Fritz, head of central office, Fraunhofer Cluster of Excellence Cognitive Internet Technologies CCIT: “Through ECC4P, the Fraunhofer CCIT offers a scientific and technological solution resulting from intensive research into the edge-cloud continuum by various Fraunhofer institutes — research that will continue to be expanded and developed in the future. It includes reliable IoT sensor technologies, innovative machine learning methods and secure dataspaces. Fraunhofer CCIT combines the scientific findings, expertise and problem-solving skills from these fields, so it can support the entire digitalized value chain as a neutral technology provider for industrial businesses.”