Siemens Electronics Factory Erlangen Reduces Machine Learning Deployment Time by 80% with AWS and Siemens Industrial AI on Industrial Edge

The factory has been at the forefront of innovation in electronics, producing PCBs that control motors for industrial machinery. This includes products like SINAMICS converters and the high-performing SINUMERIK CNC controllers. The factory serves a variety of industries, from intralogistics, to aerospace, to automotive, and many more.

The factory has been using computer vision (CV) for more than 20 of its 50 years in operation. It relies on those images to ensure its PCBs have all their components in the right places before they’re soldered. The factory’s engineering team uses the images as training data and creates ML models to inspect and spot anomalies in PCBs during production. In the early days, engineers trained models locally on computers running on premises. But after a while, computing limitations with Graphics Processing Units (GPUs) and a lack of elasticity became challenging. The team needed a more flexible and faster way to train and retrain their ML models.

Siemens Electronics Factory Erlangen leveraged AWS services and Siemens Industrial Edge solutions to cover the complete AI lifecycle: training and retraining ML models in the cloud, and deployment and operations on the shopfloor.

“We were running into GPU bottlenecks before deploying to cloud-based ML training on AWS,” said Marvin Herchenbach, process engineer and application owner for computer vision at the factory. “Also, for model results analysis we have the advantage of scalability and flexibility to handle large volumes of image data collected on the shopfloor and send them directly to the cloud. This allows a continuous improvement of the computer vision model, ensuring optimal, fast performance any time.”

The factory onboarded AWS services for model training and deployment through Siemens Industrial Edge and Siemens Industrial AI. Images from the shopfloor are sent via Edge applications to Amazon Simple Storage Service (Amazon S3) for storage, before going through Amazon SageMaker or AWS Lambda for model training. The training results from Amazon SageMaker, as well as the prediction results from AI Inference Server from each Industrial Edge Device, are sent to AI Model Monitor for observability measures.

Additionally, all the models are visible within AI Model Manager, which is a component that allows central management of AI models on the shopfloor. With the integration of AWS and AI Model Manager, the team can easily move models to different edge devices, where AI Inference Server processes images at scale, and AI Model Monitor provides full stack observability.

“The factory needed close connections between AWS and Siemens applications, considering everything we do within model architecture,” said Herchenbach. “Now it’s a smooth process with seamless integration from cloud to edge, built for security, and the maintenance is easy compared to traditional on-premises use cases. Moreover, for each retraining and deployment of a new model no further IT resources and colleagues are needed since the cloud-to-edge infrastructure ensures here an automated and easy to use process for us as domain experts.”