Ingroth Automates Barrel Defect Inspection with Datature

About Ingroth

Ing. Büro Roth GmbH is an engineering company that has been providing turn-key solutions in industrial automation for 30 years. The Company represents a blend of traditional engineering values and a forward-thinking approach, positioning itself as a key partner in navigating the complexities of today's and tomorrow's automation challenges. Through its expertise in Programmable Logic Controller (PLC) programming and automation, the company has successfully streamlined operations across various sectors, including manufacturing, packaging, and process industries. Notable projects include the implementation of advanced automation systems for production lines and the development of custom software solutions that enhance machine performance and efficiency, demonstrating their commitment to delivering comprehensive and adaptable solutions.

Industry

Food & Beverage, Manufacturing

Headquarters

Neunkirchen (NRW), Germany

Ingroth’s Operational Challenges: Efficient Keg Inspection

At the heart of many German breweries is the task of filling beer into various types of kegs, such as PU, Steel, and Eco-KEGs. A frequent issue encountered with PU-KEGs, particularly given their extensive reuse over several years, involves damaged handles. This kind of defect complicates their handling and necessitates their removal from the production line. The conventional approach to detecting these defects was heavily reliant on mechanical techniques, involving several pneumatic steps before inspection — a process ripe for innovation.

AI-Driven Solutions with Datature

To tackle their challenges, Ingroth has harnessed Datature's comprehensive MLOps platform, Nexus, to revolutionize their keg handlebar defect inspection processes. Through our partnership, Ingroth seamlessly progressed from data collection to a first-iteration, proof-of-concept trained model in a mere two weeks, thanks to the intuitive nature of the Nexus platform and its no-code environment.

The journey began with the collection of data using an Industrial Edge device. This step was critical for training a bespoke AI model capable of autonomously identifying the necessary features for precise object recognition, all under human oversight. The core of this task was to ensure a diverse and varied dataset, capturing the wide range of conditions and scenarios in which PU-KEGs are used. The type and variation of the collected data are crucial, as they directly influence the model's ability to accurately recognize and differentiate between normal and defective kegs. This diversity in data ensures the AI system's robustness and adaptability to real-world complexities.

Once the image data had been captured and imported onto Nexus, Ingroth swiftly transitioned to the annotation phase using Nexus' Annotator. Leveraging Datature's suite of annotation tools, including the Rectangle tool and smart labelling tools such as Intellibrush, Ingroth precisely labelled handlebar defects with bounding boxes and polygons, ensuring accuracy and efficiency throughout the process. 

Following annotation, Ingroth expedited model training, achieving a robust, first-iteration object detection model within a short time.The model managed to learn how to predict the handlebars with decent accuracy within just a mere couple of hours. Datature’s team of seasoned domain experts, engineers, and developers collaborated closely with Ingroth at every stage of the MLOps lifecycle, providing invaluable guidance and support to develop a functional model seamlessly and effortlessly. Through targeted modifications to the training pipeline, such as leveraging Image Augmentation techniques based on the nature of the dataset for increased variability, hyperparameter tuning based on model validation metrics, and coupled with fast, iterative model retraining, Ingroth was able to fine-tune their first-iteration model and reach an accuracy of 98.8%, which was extremely promising at such an early stage.

Deploying to the Real World

The solution was deployed on a Siemens Industrial Edge device, chosen for its integration into a comprehensive ecosystem. This system's modular nature allowed for the flexible use of applications, facilitating various adjustments and ensuring the solution's adaptability to different scenarios through simple app exchanges.

Following deployment, the solution was tested under real production scenarios. The AI's performance was analyzed, with its efficacy and any limitations thoroughly evaluated. An essential part of this process involved parameterizing the interaction between the camera, AI on the edge, and the PLC to ensure that the data flow and decision-making process were finely tuned for optimal operation within the production line's environment. Adjustments made during this phase allowed for effective processing and decision-making, optimizing the system for the dynamic and complex conditions of the brewery. Insights gained from these tests informed further model refinement, enhancing the solution's overall effectiveness and efficiency.

Continuing and Expanding AI Applications

To enable ongoing iterative development to stay robust in the face of model drift in changing data and environmental conditions, Ingroth can leverage Datature's Active Learning capabilities and Management API. This empowers them to identify images with low-confidence predictions, facilitate Model-Assisted Labelling of new batches of images, and automate training and deployment of new models, ensuring continuous improvement and adaptability to evolving conditions.

Leveraging automated computer vision solutions with Datature not only addressed Ingroth’s immediate challenge of keg handle detection, but also set the stage for future innovations. The successful application of AI and Edge Computing heralds a new era of potential advancements, including broader AI-based inspections like fitting and logo recognition, all while conveniently utilizing the same end-to-end MLOps pipeline with Datature Nexus. Such innovations promise to revolutionize not just brewery operations but potentially other sectors seeking efficiency and precision. This exemplifies the potential of combining traditional operational challenges with advanced AI technologies, marking a significant step forward in industrial automation and operational efficiency.

Datature's innovative tools have empowered Ingroth to not only develop a cutting-edge computer vision solution for automated keg handlebar defect detection in real-time, but also establish a streamlined pipeline for future use cases.

‍

"For us, as an automation company, maintaining a high level of quality in our components is crucial so we can offer our clients the solutions they need. Datature's platform not only enables us to implement AI projects but also allows us to develop industrial-grade AI capable of operating within production environments. Since the software supports the entire process from data management to model export, we always have a clear overview of which data currently influence AI models. This flexibility ensures we can swiftly adapt to various use cases with the platform."
‍- David Wörster | Project Manager, Ing. Büro Roth GmbH

‍

Datature is proud to support Ingroth’s mission to automate and transform the manufacturing process through AI-powered inspection, and we are excited to see what’s next for Ingroth.

If you are an innovator looking to unlock the potential of deep learning to transform the Manufacturing industry, get in touch with us!

About Ingroth

Ing. Büro Roth GmbH is an engineering company that has been providing turn-key solutions in industrial automation for 30 years. The Company represents a blend of traditional engineering values and a forward-thinking approach, positioning itself as a key partner in navigating the complexities of today's and tomorrow's automation challenges. Through its expertise in Programmable Logic Controller (PLC) programming and automation, the company has successfully streamlined operations across various sectors, including manufacturing, packaging, and process industries. Notable projects include the implementation of advanced automation systems for production lines and the development of custom software solutions that enhance machine performance and efficiency, demonstrating their commitment to delivering comprehensive and adaptable solutions.

About Datature

Datature is a leading end-to-end MLOps platform that allows teams and enterprises to build computer vision models without a single line of code. Teams can manage datasets, annotate, generate synthetic data, train and deploy - all in a single, secure cloud-based platform. With the rise of citizen data scientists, deep tech companies, and enterprises looking to adopt deep-learning - Datature equips these startups / experts with the tools required to build their own capabilities easily within weeks.