Time series as the future of production

Artificial intelligence and machine learning are the next logical step in industrial automation. While classic automation solutions are becoming increasingly interchangeable, companies are on the lookout for new differentiating features. The following holds true: Many AI‑applications are already working cost efficiently today. Progress does not require autonomous agents.

Currently, there are four fields that hold great potential: Engineering, simulation, vision and, above all, time series. In the engineering area, AI‑systems support design, parameterisation and documentation. In the field of simulation, AI-based models are created that map complex physical processes faster and more easily than conventional tools. Vision remains a cornerstone of industrial AI, while the next technological leap is on the horizon in the area of time series.

Vision systems are already demonstrating just how well AI can function in an industrial environment: Quality testing, object recognition and inspections are production ready. The analysis of time series is the next major field on the agenda. Machines communicate via sensor data, process values and energy consumption – and modern AI‑models are able to analyse this information to a high level of precision. Consequently, predictive maintenance, predictive quality, energy efficiency or demand forecasting can be realised cost efficiently, often without new hardware and even on existing CPUs.

While we hear a lot of talk about autonomous agent systems or self-optimising machines, these approaches are still around 10 to 15 years in the distance. A lack of standards, safety‑requirements and business models are putting the brakes on the direct coupling of AI‑models to control systems. Today, optimisations are usually carried out in the digital twin of the system before they are transferred to real production.

Humans remain in the control loop – for now.

AI-based assistance systems, such as large language models (LLMs), could facilitate the operation and configuration of machines in the future and thereby help to offset the shortage of skilled labour. But the same applies here: The costs for applications and operation must pay off. Not every innovation is immediately commercially viable – while there are many proofs of concept, actual, genuine scaling remains rare.

Especially in the time series area, large foundation models are currently being developed, which – similar to text data – are trained on a broad data basis.

Those who control this data can not only optimise their own products, but also develop new business models – including the optimisation of competitor products.

Change will come, but not overnight. Although automation cycles are shortening, fully autonomous production chains are still years away. One thing is certain: AI will not replace automation, but rather expand it – and gradually lead the industry from traditional control systems on to data-driven, learning systems.