Industrial AI needs holistic IT

Many industrial companies have already successfully implemented their first AI use cases and are now embarking on a broader rollout. From Dell Technologies' point of view, they should avoid implementing separate IT infrastructures for individual use cases. Such isolated solutions are usually costly, inefficient and do not scale well.

Instead, it is better to build a holistic IT platform - also known as an AI factory - as a common foundation for all current and future use cases. Such a platform enables industrial companies to better utilize and reuse resources, reduce costs and use AI in a more agile way.

However, such a platform must be able to cover the different requirements of AI applications in terms of the location and speed of data processing. Three main classes of industrial AI can be identified:

1. AI for trend analyses
Examples of this application class include predictive maintenance, systems for improving energy consumption and solutions for optimizing production flows. Such applications need to combine data from different sources and are not dependent on AI decisions in milliseconds. Both the central data storage and the training of AI models and the making of AI decisions can take place in the cloud - in a public or private cloud, depending on compliance requirements.

2. AI for real-time decisions
This class includes applications such as image processing for detecting rejects, systems for optimizing process parameters or solutions for real-time monitoring of machines. The decisions made by these applications are directly integrated into the production process and must therefore not be dependent on external connections. This is why they require decentralized data storage and decisions to be made directly at the production site. The training of AI models, on the other hand, can take place in the cloud.

3. physical AI
Typical applications in this class are self-driving transport vehicles, robots that perceive their surroundings or industrial co-pilots. They require high computing power very close to or even in the machines themselves so that the AI can make its decisions at the necessary speed. Further training with large amounts of synthetic data allows the AI models of these applications to adapt to new requirements. The initial training and further training can take place in the cloud, while the AI decisions have to be made directly on the store floor.

"In order for an AI Factory to cover all of these different requirements, it must follow a multi-level approach with three levels," explains Dr. Stefan Muthmann, Field CTO at Dell Technologies in Germany. "It combines IT systems on the store floor with local data centers at the production sites and private or public clouds. Through close networking, it ensures continuous feedback between these three levels and enables the orchestration of workloads across them."