AI in manufacturing: lessons from Siemens’ Erlangen plant

Preparing a production line before the physical product is available and comparing different production schedules are the two uses of artificial intelligence described by Siemens in its 19 October announcement about the Erlangen plant in Germany. The project stems from its collaboration with NVIDIA.

According to the company, virtual preparation for producing new printed circuit board assemblies can reduce the engineering work for that stage by up to 85%. The system uses representations generated from digital twins to train visual inspection. For planning, an agent proposes scenarios that are checked in the plant model: Siemens reports potential savings of up to 20%. These figures come from the manufacturer, not an independent Economind measurement. Official source.

Assessing the value of earlier preparation

In Economind's economic analysis, the issue to investigate is the cost of waiting. For a company considering a similar project, the first question should be which activity can start earlier and which decision still depends on a physical test. A comparison is useful only when these boundaries remain explicit.

We suggest separating three items in the assessment: preparation hours, start-up duration and the system's recurring costs. An improvement in the first does not, by itself, demonstrate an equivalent improvement in the others. Likewise, time released gains economic value when the company can use it for a useful activity or avoid a cost.

Questions for a comparable investment

To compare this experience with another plant, we would ask for a precise description of the previous process, the activities included in the calculation and the variety of products observed. Integration costs, model maintenance, review of results and operator training should also be considered. These are evaluation questions proposed by Economind, not information about costs incurred at Erlangen.

Before attributing an economic return to a similar solution, the company therefore needs an assessment built around its own plant. The case documents a practical application worth examining; the announced percentages describe Siemens' stated scope and are not a forecast transferable to every factory.