Quality inspection
AI with an eye for sweetness
In its first deep learning project, Bi-Ber has implemented visual product inspection for a manufacturer of chocolate and wafer products. The final inspection ensures that there are no foreign objects on the products and that they meet aesthetic standards.
Until now, this task could only be fulfilled by visual inspection because conventional rule-based automatic image processing (machine vision) cannot reliably distinguish defects on irregularly structured surfaces from normal variances. AI-based automatic quality inspection is now set to increase detection reliability.
The inspection system reports an OK or NOK signal for each chocolate product in a mold - this requires an evaluation speed of up to 54 segments per second. Bi-Ber used Cognex VisionPro ViDi software to develop the AI, which includes an efficient segmentation and defect detection tool and requires only a few images to train an artificial neural network.
Bi-Ber only taught in 49 images of good products and 58 of rejects for the test run. Optimizing the recognition accuracy and speed for the system cycle was particularly labour-intensive. Once Bi-Ber had mastered this task, new molds and products can now be taught in within just four minutes. Bi-Ber has trained the fully integrated software and hardware system with sample molds. It is currently being tested in a six-month trial run, expanded to include new molds and fine-tuned. The hardware is based on Bi-Ber's previous successful machine vision projects in the confectionery industry: GigE cameras, flat design, shielding from the product area, stainless steel housing, touchscreen outside the product area. as








