Siemens demonstrates PepsiCo filling-line digital twin

Siemens has demonstrated how industrial digital twins and artificial intelligence can support more flexible beverage filling and packaging through its collaboration with PepsiCo and NVIDIA. The presentation at PACK EXPO International in Chicago on 21 October examined changing product specifications, mixed-product pallet configurations and the integration of manufacturing data within a virtual representation of production operations.

The demonstration builds on a partnership publicly detailed in January 2026, rather than marking the launch of a new agreement. Siemens Digital Twin Composer combines engineering models, simulations and operational information to represent the behaviour of equipment and production processes. NVIDIA technology contributes simulation and visualisation capabilities, allowing proposed changes to be evaluated before they are applied to physical machinery.

Andrea Sassetti, digital enterprise innovation manager at Siemens, explained the approach during the exhibition, describing how separate software systems contribute information about packaging, filling and robotic operations. Their outputs are brought together within a wider facility model, allowing interactions between equipment to be assessed rather than treating each production stage as an isolated system.

One focus of the Chicago presentation was flexibility in filling different beverage specifications. Products can differ in viscosity, ingredient composition and the presence of suspended particles, creating different demands on mixing, liquid transfer and filling machinery. The simulation environment is intended to help determine suitable operating arrangements for those conditions without relying exclusively on repeated physical trials.

Viscosity affects the rate at which a liquid moves through pumps, valves and filling nozzles. A relatively thin beverage may behave differently from a product containing more viscous ingredients or suspended solids, while particles can introduce requirements for maintaining dispersion and avoiding blockages. Changes to filling settings can also affect accuracy, cycle time and product handling after the container has been filled.

Siemens describes using AI models to propose equipment settings based on characteristics including viscosity and particle size. Such recommendations can be investigated through the digital representation before an operator decides whether they are suitable for use. The October presentation does not provide a validated operating envelope for every beverage formulation or establish that a simulated setting can be applied safely to any filling machine without further checks.

The wider production line must also accommodate the consequences of changing one operation. Increasing filler speed may create accumulation at a capper or downstream inspection station, while slower filling can leave packaging equipment underused. A virtual model can examine these relationships and help identify the stage constraining the performance of the whole system.

Simulation requires reliable information about the equipment and process under examination. Geometry, machine cycles, conveying arrangements and operating limits must be represented with enough accuracy to support the intended decision. Real production data can improve the model, but incomplete or outdated information may produce conclusions that do not match physical conditions.

Another part of the demonstration concerned palletising, particularly the preparation of loads containing more than one product format or flavour. Beverage businesses may need mixed orders for events or regional distribution rather than uniform pallets containing a single product. The arrangement of different packs influences robotic handling sequences, load stability and the time needed to assemble the completed pallet.

Digital simulation can evaluate proposed stacking patterns and equipment movements before a configuration is introduced. A palletising robot must still respect its payload and reach limits, while completed loads must withstand the handling and distribution conditions they will encounter. Siemens has not disclosed a new independently measured throughput improvement associated specifically with the mixed-pallet demonstration.

Changing product demand was another theme. The company described how production systems could use information about local markets and events when assessing the quantities and formats required. Such demand forecasting can inform manufacturing plans, although the presentation did not establish that a new temporary PepsiCo factory has been constructed or that all proposed production scenarios are operating commercially.

Connected production information also offers a potential basis for packaging data management. Producers may need to account for the materials and packaging formats placed on particular markets under extended producer responsibility systems. Equipment and product records can contribute to that work when material specifications, production quantities and product identities are accurately associated with one another.

The quality of those records remains dependent on data capture and governance. A digital twin cannot by itself verify that every packaging declaration is complete or compliant, while information from suppliers and other external systems may require separate checks. Siemens described the development of authentication capabilities with additional partners, but no universal certification covering every packaging reporting requirement was established.

Earlier PepsiCo work has produced quantified performance claims that must remain separate from the new exhibition presentation. In January, Siemens reported a 20% throughput increase in an initial deployment and projected capital expenditure reductions of 10% to 15% through the simulation and validation of proposed changes. Those figures concern the earlier programme and should not be presented as fresh results measured at PACK EXPO.

The January work also involved high-fidelity representations of machines, conveyors, pallet routes and operator movements. Bringing these elements into a connected model can help reveal bottlenecks and interactions that are difficult to identify when equipment is assessed separately. Its usefulness depends on continuing to compare predictions with physical operating results as processes, products and equipment change.

Siemens emphasised that AI-generated recommendations remain subject to operator oversight. Sassetti described the need for a human supervisor to assess proposed actions before they are accepted, particularly where incorrect settings could affect quality or machine operation. The distinction prevents simulation-generated suggestions from being treated as automatically validated instructions for industrial equipment.

The October demonstration extends the established PepsiCo programme into specific examples of filling flexibility, mixed-product palletising and connected packaging information. Further operating evidence will be needed to establish how the proposed adjustments perform across commercial beverage lines and different formulations. The latest presentation demonstrates applications of the existing digital twin platform rather than a newly completed manufacturing installation.