Edge Analytics

As soon as relevant data is available on site via data management, it can be analyzed with Edge Analytics algorithms, e.g. to identify the quality of a product during manufacture or the predictive maintenance condition of a plant.

The term »Edge« stands for an analysis directly in the production area, not in the cloud. The advantage of this approach is that no data has to be transferred to external cloud servers via potentially insecure communication media. All data relevant for the analysis can be accessed with minimal delay (low latency) and under real-time conditions. This increases the reliability of systems enormously and hardens systems and analysis results against cyber attacks, since relevant data remains on site and thus securely in the possession of the system operator. If cross-location analyses in the cloud require specific data, these can be specifically selected and sent to the cloud.

Our know-how lies in the fact that we can draw the right conclusions on the basis of embedded systems from data on site. Furthermore, we are able to reflect these results into the systems of our customers.

For the implementation of Edge Analytics tasks, we use different methods depending on the requirements. The learning process distinguishes between supervised and unsupervised learning.

Edge Analytics uses e.g. following procedures:

• Deep learning based on different architectures of neural networks, if sufficient training data is available
• Clustering or classification methods
  
• Regression methods

Our algorithms can run on resource-limited systems such as embedded systems based on ARM processors. It is possible to extend the functionality of components already used in production with sufficient free computing resources. If the computing power of the processor used is not sufficient for the necessary analysis tasks or for a learning process taking place on site, we are able to extend the system with appropriate hardware acceleration methods and thus assist in accomplishing its task. Depending on requirements and needs, we use complex programming logic in the form of FPGAs, graphics processors or commercially available neuromorphic processors.