TL;DR:
- Nestle’s plant in Germany utilizes SICK Deep Learning to improve product inspection.
- Challenge: Transparent plastic scoop detection on aluminum foil lids was problematic.
- Solution: SICK Deep Learning empowers operators to train neural networks intuitively.
- Real-world production images are categorized and processed in the Cloud for network training.
- Trained neural networks integrated into SICK 2D camera for precise decision-making.
- The inference process occurs directly on devices, enabling quick and predictable decisions.
- The system halts production on scoop detection failure and resumes upon scoop addition.
- Deep Learning reduces setup times and costs, enabling autonomous AI image classification.
- The convergence of technology and manufacturing prowess propels the industry forward.
Main AI News:
In the realm of nutritional product manufacturing, Nestle’s plant in Osthofen, Germany, stands as a symbol of innovation. Catering to the unique dietary needs of individuals, Nestle crafts specialized products like sip feeds and supplements. A pivotal aspect of their production involves a meticulous quality control process, where a measuring scoop is placed in each container before automated powder filling takes place. Historically, a vision camera employing a color pixel counting tool was employed to ensure the presence of this scoop within the container. This scrutiny was executed at a remarkable process speed of over 80 cans per minute.
However, progress brought forth a challenge. In pursuit of a higher recycling rate, Nestle introduced a nearly colorless plastic scoop. This, while environmentally commendable, posed a predicament. The transparent scoop, adorned with a subtle grey tint, became a conundrum for the conventional image processing system. Detecting it atop an aluminum foil lid that shared a similar color proved intricate. The lid was further complicated, being corrugated, embossed, and reflective.
Enter SICK Deep Learning, an intuitive resolution to this intricate issue. This cutting-edge technology empowers operators to navigate a user-friendly graphic interface, enabling the selection and training of neural networks through simple steps. Once the initial setup is complete, operators gather real-world production images and categorize them. These images undergo pre-processing and are transferred to the Cloud for the neural network’s training. The result is a finely-tuned network ready for application. This intelligence can then be downloaded to a Deep Learning-enabled device, facilitating automated decision-making without requiring a constant Cloud connection.
The training of SICK’s neural networks is notable. By exposing them to a diverse array of images showcasing the scoop in various orientations, the networks learn decision-making strategies. Subsequently, the trained algorithm is integrated into a SICK 2D camera.
Intriguingly, the inference process occurs directly on the device, ensuring quick and predictable decision times. This autonomy eliminates the need for an additional PC. The outcomes are transmitted as sensor values to the control unit. A remarkable feature of the system is its ability to halt the process if the Deep Learning system detects a missing scoop. Upon identifying the addition of a scoop, production can seamlessly resume without manual intervention.
The advent of Deep Learning revolutionizes manufacturing processes, significantly reducing setup times and associated costs. Through the deployment of Artificial Intelligence (AI) image classification onboard smart devices, decisions are taken autonomously. Neural networks, meticulously optimized for specific tasks, enable precise inspections that were once deemed unfeasible in high-speed automated procedures. This convergence of technology and manufacturing prowess propels the industry towards new horizons.
Conclusion:
The adoption of SICK Deep Learning technology by Nestle not only addresses a challenging quality control issue but also underscores the potential of AI-powered solutions in manufacturing. This advancement signifies a significant leap forward in the market, as it showcases how AI-driven image classification can enhance accuracy, reduce costs, and streamline production processes. Businesses across industries could benefit from such solutions, revolutionizing their operations and bolstering competitiveness in a fast-paced market landscape.
