Component Reel Inventory & Labeling

The challenge

An automotive-electronics manufacturer takes in reels of surface-mount components from dozens of different suppliers. Every reel carries a label with the part number, lot code, and quantity, and that is the data the plant needs to bring the reel into its own inventory system.

The problem is that no two suppliers label a reel the same way. The label sits in a different place, at a different size, in a different layout. Some are in English, some in Japanese. The part number might be plain text on one reel and inside a barcode or a 2D data-matrix on the next. A conventional OCR setup depends on knowing where the text is and what it looks like, so a new supplier, or a supplier changing their label, breaks it. Reading these by hand is slow and error-prone, and a wrong part or lot number entering inventory is a traceability problem later.

What we built

We built a vision system that reads the reel the way a person would: it finds the label wherever it is and reads what is on it, without being told the format in advance. Four 10-megapixel cameras, each with its own linear light, image the reel so the label is captured cleanly regardless of how it is oriented.

The reading is done with SolVAI, our in-house deep-learning vision platform, trained on real reels from the actual supplier mix rather than on a fixed template. Because SolVAI learned from examples of how these labels really vary, it reads a supplier it has seen before even when the label moves or changes, and new suppliers can be added by training on more examples instead of rewriting rules. The part number, lot code, and quantity come off the label and go straight into the inventory record, and the reel is re-labeled to the plant's single internal standard so everything downstream reads the same way.

Reel handling is automated so reels are presented to the cameras and moved through the cell without manual placement.

System at a glance

• Imaging4 × 10MP cameras, linear lighting
• ReadingSolVAI deep-learning OCR
• OutputPart, lot, and quantity into inventory; re-label to internal standard

Why deep learning, not a template

Rules-based OCR works when the text is always in the same place and the same font. Here it never is. Training SolVAI on the real spread of supplier labels means the system recognizes the data by what it is, not by where it sits, so it holds up across suppliers and survives the label changes that would otherwise mean a re-program every time.

Capabilities used

Machine vision, SolVAI deep-learning OCR, multi-camera imaging and lighting design, robotic handling integration, and connection into an inventory system.

The label problem, in pictures

The same system reads all of these. Every reel below comes from a different supplier, and no two put the data in the same place, the same size, or the same symbology.