From the Pyramids into the Clouds

In traditional production, communication is usually hierarchical, from machines to management. In the so-called »automation pyramid,« data moves up and down between levels in the machine environment. The field level, with its sensors, actuators, and computer numerical control (CNC), is at the bottom. It generates high-frequency process data directly at the machine tool. This data is then passed on to the higher control levels, where it is used to monitor and optimize processes. Finally, at the top is the corporate level, where important strategic decisions are made.

Fieldbuses are used to exchange data between these levels, for example, the real-time system PROFINET at the field level. However, such real-time buses are not available at the higher levels. Therefore, the high-frequency data must be translated into other protocols, such as OPC UA. This causes delays and data loss, resulting in critical process information reaching the upper levels too late or not at all. Practical experience shows that it is not just the availability of data that is crucial, but also its consistency and availability exactly when it is needed.

Cloud and Edge 

This requires the implementation of agile systems capable of processing vast quantities of data with ease. Cloud computing allows for the dissolution of the traditional pyramid by providing services, data, and applications via distributed nodes. This is also reflected in the costs of IT infrastructure. The Federation of German Industries (Bundesverband der Deutschen Industrie, BDI) confirms that cloud computing reduces companies’ IT costs by an average of 50 percent and their organizational costs by around 20 percent.

Edge computing, on the other hand, processes data directly at the field level. Edge devices, installed in the control cabinet of the machine tool, collect and process high-frequency data where it is generated. They do so without compromising the performance of the machine control system. Large amounts of data no longer have to be routed through multiple levels, response times are shortened, and process reliability is increased.

The combination of both is particularly effective: Computing load and storage requirements are optimally distributed between the edge and the cloud, transmission paths are relieved, and machines operate more autonomously and flexibly. Older systems can also be retrofitted with edge devices, modern sensors, and additional interfaces.

Connecting to the network with the right equipment

For years, manufacturers have offered platform solutions that are compatible with commonly used machine controls. These solutions consist of the edge devices themselves, suitable apps, and the necessary connectivity and management infrastructure. At the field level, the edge devices make machine-internal protocols IP-capable, connecting the machine and the cloud. Apps for data storage, visualization, or digital twin applications are available via a »marketplace.« Corresponding software development kits (SDK) allow companies to develop their own apps tailored to their processes.

Researchers at the IWF at TU Berlin have taken advantage of one of these platform solutions. They expanded a CNC milling machine with an edge device, the necessary network technology, and a sensor adapter to capture various sensor data.

A network switch connects the respective components to the company and machine network so that process data can be processed locally or transferred to the cloud. The logic for automated operation was implemented in the machine control system and in an edge app. High-frequency machine data can be recorded 500 times per second, while sensor data is recorded in parallel. A specially developed edge app controls the process and automatically transmits countermeasures to the machine control system in the event of critical sensor data.

Control and edge precisely in sync

The prerequisite is that communication between the machine control system and the edge device runs reliably, as anyone who works with different systems knows the challenge of compatibility. The researchers solved this by synchronizing the edge device and the machine control system. This guarantees that all recorded data, whether high-frequency, low-frequency, or sensor data, can be clearly assigned. This allows  for precise commands to be transmitted to the control system in the event of critical sensor values.

To this end, so-called synchronous actions can be stored in the machine control system G-code. These are checked in each interpolation cycle of the tool control system using defined calculation parameters (R parameters) and executed in real time if the conditions are met. The machine and edge app exchange process states using the freely assignable R parameters from 0 to 999 to synchronize the process. One parameter, for example, can signal that the start position for a process step has been reached. The edge app automatically starts data acquisition and reports back that it is ready.

Only then, milling begins under the direct supervision of the edge device. If the app detects critical sensor data during the process, it reports this back to the control system via another R parameter. For example, wait commands can be integrated to continue milling and collect data even in unstable zones. The cut can then be stopped, the machining adjusted, or restarted. This close coupling allows machine and sensor data to be precisely linked and processes to be fully automated.

Without sensors, with AI?

One area of application for this technology is wear prediction. To this end, the researchers are conducting milling tests with new and already worn tools. During machining, the edge app records three types of data: high-frequency machine data, low-frequency control data, and external sensor data. Synchronization ensures precise coordination of all data sources, enabling accurate comparison. The tests also showed first signs that critical process states are evident in sensor data as well as machine data.

The objective is straightforward: identify signal changes that indicate tool wear and find these patterns in the machine data. An AI model uses sensor data to learn how to derive wear conditions from machine data alone. In later operation, the AI evaluates only this data, detects wear at an early stage, and automatically initiates a tool change – without manual intervention