- AI
- NVIDIA
- Case story
Discover how food manufacturing can optimize production efficiency by using AI & Machine learning. Imagine having a real-time monitoring (counting) system that accurately counts products on the production line, recognizing new product varieties without requiring retraining of the model. That's exactly what we achieved together with Kohberg Bakery Group.
By combining computer vision and powerful edge computing devices we've created an affordable, AI powered solution that addresses key challenges in bakery production and seamlessly integrates into existing industrial applications, like Modbus.
Do you want to know more about our product counting technology from our team?
Business Challenges
Kohberg is Denmark’s largest Danish - and family-owned bakery with approximately 500 employees and a wide variety of bread products. They share a strong passion for good bread and pastries and also for technology. They are always considering solutions that match current market trends, and their customers' and employees' needs. They keep this approach also for manufacturing operations.
Inaccurate counting: Accurately counting the number of finished bread products produced each day.
Difficulty in counting new products without additional learning process: Implementing a AI based system to count newly introduced bread varieties without requiring manual adjustments.
Difficult configurations and use of the previous counting system: The new AI system must be easy to use and scalable.
While computer vision and AI excel at recognizing and counting objects they've been trained on, identifying novel objects or variations they haven't encountered remains a challenge. The existing vision system, based on light sensors, while capable of counting items, suffered from significant inaccuracies. The company was unable to confidently determine the exact production volume on a daily or monthly basis. The previous system overestimated the count because it assumed the presence of a product at every stage, even if a piece of dough was missing from the line during the early cutting phase. Also due to a wide variety of bread products it is challenging for traditional counting systems to accurately identify and count each unique type of product.
AI Powered Counting Solution
A computer vision solution was developed to count bread products using a single camera, making it a cost-effective solution requiring only minimal hardware. By leveraging a deep learning model based on YOLO, we achieved accurate product recognition and counting, even for diverse bread types.
The system was deployed in a containerized environment, enabling efficient scaling and management.
Key components of the solution:
- Nvidia Jetson™ Orin NX module
- Two waterproof IP cameras
- YOLO model
The solution was seamlessly integrated into the company's existing Modbus industrial app, eliminating the need for additional user interfaces. This simplified the deployment and operation of the system.
By training the model on a variety of bread types, we achieved high accuracy and reduced the need for human intervention in the counting process.
How AI Can Recognize New Products on the Production Line
A significant challenge was training the system to automatically identify new bread types not included in the initial dataset. For example, if the bakery wanted to introduce seasonal Christmas buns, we needed a method to enable the system to recognize these new products without extensive retraining.
Our solution involved training the system to focus solely on moving objects on the conveyor belt. By doing so, the machine can effectively identify any new bread shapes or sizes on the conveyor. This approach provides a flexible and adaptable solution for future product introductions.
AI Powered Manufacturing: Easier, Cheaper, Better
The bakery and food manufacturing industry is undergoing a significant transformation, driven by the power of artificial intelligence. Traditional methods and outdated technologies, once considered state-of-the-art, are now being replaced by AI-powered solutions that deliver superior results with greater efficiency and cost-effectiveness.
'For us, the process has become significantly more simplified compared to the past, especially with the use of Radars. Previously, light sensors would only inspect the top of the products, and if a product moved even slightly, the system required reprogramming and also for every different kind of product. This has all been resolved. Now, the camera simply views the buns, and know how to count buns. Similarly, it can identify and count other bread products. Moreover, the current AI-based solution is more cost-effective than its predecessors due to the reduced complexity’
says Michael Kjær, Head of IT at Kohber
AI and Quality Control
The next step for production is implementing AI for quality control. This presents a unique challenge as the bakery aims to achieve a handmade look, which inherently involves natural variations and intended inconsistencies. To overcome this, we will train the AI model to recognize a range of acceptable variations within the desired "handmade" aesthetic.
Specifically, we will define a period of "perfect" production (e.g., the last 5 minutes where all products were deemed acceptable) and use this as a reference for future product quality.
The system should be able to distinguish between acceptable and unacceptable variations in product color, size, and shape. It will utilize self-learning capabilities, constantly refining its understanding of acceptable quality based on ongoing production data.
Let AI Improve your Manufacturing
AI for product counting enables process optimization, increased efficiency, and cost reduction. Accurate production data obtained through AI helps better plan production, manage inventory, and identify deviations in a timely manner.
This case study highlights just one example of the transformative power of AI. Discover the full spectrum of our AI & Machine Learning offerings designed by Danoffice IT to revolutionize your business operations.
Do you want to know more about our product counting technology from our team?
-
CASE: Advanced harbor trackingDiscover how a harbor can become more efficient and secure with AI. Explore our advanced harbor tracking system that shows all the traffic in real time. -
CASE: Machine visionDiscover the benefits of implementing machine vision technology and IoT (Edge AI) for optimizing label inspection in a candy factory.
