Procter & Gamble is expanding an AI inspection system across its manufacturing operations worldwide, Siemens announced on September 16. The jointly developed Visual Inspection Cockpit checks products during production and connects defect findings to factory controls, allowing a line to respond while the affected item is still moving through it.
The system addresses a difficult part of consumer-goods inspection: acceptable products do not always look identical. Soft, textured materials can stretch or wrinkle, and packaging changes alter what a camera sees. Siemens says conventional vision systems often need extensive reconfiguration as these conditions change. P&G sought continuous inspection that could accommodate varying materials, formats and lighting without repeated reprogramming.
Visual Inspection Cockpit combines learned image recognition with conventional vision rules. Siemens' customer reference describes semantic segmentation, a method that assigns categories to regions of an image so the software can locate relevant features. The analysis runs on industrial computers beside the machine through Siemens Industrial Edge, its platform for operating software near production equipment. Keeping inspection there lets the resulting defect information feed into the manufacturing process; the announcement says it can trigger alerts or removal of a defective item.
Preparing those decisions for production requires examples from the actual line. Siemens' engineering guide tells teams to collect camera images in the deployment environment, select a representative set covering the defects of interest and mark the relevant regions with masks. Engineers then develop and validate the model and its processing steps offline. Next comes open-loop operation: the system analyzes images and records defects, but communication with the production controller remains disabled. This stage gives the team a way to check its judgments before enabling automatic responses.
Once validation is complete, closed-loop operation sends defect information to the programmable logic controller, or PLC, the computer that controls the machinery. The guide describes configurable reject reasons that can be mapped to controller outputs. It also supports recipes, inspection configurations selected for different product variants on one line. These connections need checking against the installed controller: the version 1.1.0 guide warns that some data transfers may be unavailable with vendors other than the fully supported Rockwell controllers.
Siemens reports lower scrap and faster commissioning, but its announcement does not disclose plant counts, evaluation periods or matched defect-detection results. The public evidence therefore establishes an expansion of a working inspection approach while leaving the size and consistency of its benefits difficult to assess. For factory teams, the documentation supplies a concrete path from labeled images to controlled rejection of products. The next useful deployment reports would show missed defects, unnecessary rejects and commissioning time for specified products and line speeds as the rollout expands.