Implement applications more easily and cost-effectively
AI could revolutionize automation. How can Time-Sensitive Networks (TSN) make AI applications easier and more cost-effective? What network requirements does AI need for the future? What new applications and solutions can TSN enable, and what is the path forward?
Since detailed TSN information is widely available, this article will explore AI-focused applications benefiting from TSN, excluding motion control. These examples aim to inspire exploring options beyond current limits.
The rapid development of AI tools, accelerated hardware capabilities, and integrated Ethernet interfaces is opening new applications and segments. These generally require:
- Transporting large amounts of data from the field to the AI system.
- Providing forwarded data with highly accurate time stamps for comprehensive time series data analysis.
- Feeding AI processing results back into the field.
Conventional 100 Mbit industrial Ethernet can’t fully meet these requirements. Time-Sensitive Networking (TSN) offers solutions. The following examples show how AI-focused applications can be implemented or improved by a convergent TSN network.
Camera-based quality control in the active process
In production, special industrial cameras (GigE Vision) capture images of processes or products. Machine learning detects quality defects and controls production. Current hardware evaluates images in milliseconds, enabling real-time quality control. Multiple synchronized cameras allow for 3D calculations or time series analyses. The evaluation algorithm, running on external hardware, adapts to different products and conditions. Industrial vision applications need high bandwidth, often several 100 Mbps, and use jumbo frames, benefiting from TSN preemption. Convergent networks are rare in vision applications but could become common with TSN.
Preventive maintenance of large machines
In predictive maintenance for large drives, sensors measure temperature and vibration, while a power device tracks energy use. This data trains an AI model for normal motor operation, with precise temporal correlation aiding the process. A system-wide time understanding is needed for multiple drives. Deviations from normal parameters allow operators to act before failures, optimizing maintenance and control. The minimal training effort is outweighed by the benefits.
Synchronized feed-in of renewable energy
The energy transition introduces a new problem: replacing conventional power stations with alternative generators like wind turbines and solar parks loses the grid-supporting effect of large rotating masses. These masses help manage load fluctuations and provide reference frequency. Accurate synchronization of all generators and their feed-in converters via TSN can solve this, using the existing network for operating data. Installing separate networks in a large wind farm would be costly.
Root cause analysis of events
Another application is the temporal and localized analysis of operational events, required by insurance companies for root cause evaluation. Current controllers may not suffice as they lack all relevant data. TSN enables recording events with highly accurate time stamps (<1 ms) in devices. The convergent network ensures simultaneous event transmission to an analyzing unit, which may not be the controller. AI helps identify anomalies, supported by precise time stamps, uncovering and rectifying hidden problems during operation.
Implementation is already possible
Beyond the examples mentioned, many AI applications in automation rely on convergent TSN networks and time synchronization, which simplify and reduce costs. TSN excels in these applications compared to specialized real-time bus systems, offering significant benefits. The good news is that these applications are already feasible. Phoenix Contact switches support functions like Quality of Service (QoS), Precision Time Protocol (PTP), and preemption. The GigE Vision standard also specifies PTP. AI and TSN will drive a new generation of automation applications, with development just beginning.
Read more about Time-Sensitive Networking
