From Sensor to Cloud and Back – End-to-End Analysis for Guided Manufacturing Using Machine Learning

Challenge

Our client faced the challenge of developing a scalable and efficient solution for machine monitoring. The existing machines did not support automatic data extraction, making it difficult to implement modern monitoring solutions. Additionally, detecting anomalies in the production process was crucial to minimize downtime and maximize productivity. The solution needed to be independent of existing hardware while seamlessly integrating into their current infrastructure. Risks included potential production downtimes, high costs due to machine failures, and the complexity of integrating the solution into the client's IT landscape.‍

Approach

During a proof of concept (PoC), an end-to-end solution was developed. Industrial Raspberry Pi setups with audio and acceleration sensors were installed to capture data directly from the machines. An anomaly detection model was implemented and executed at the edge of the manufacturing process to analyze the data in real-time. The captured and analyzed data was streamed to the cloud via Azure IoT Hub for further analysis and reporting. A custom ML container runtime was deployed to ensure seamless integration with the client’s existing infrastructure.‍

Outcome

The project resulted in significant improvements in production monitoring and control. Specifically, the following outcomes were achieved:

• Development of a standardized blueprint for future IoT use cases, serving as a template for similar projects.
• Industrial-grade Raspberry Pi setup, capable of flexible integration with various sensor types.
• Custom ML solution, enhancing anomaly detection capabilities and seamlessly integrating with the client’s infrastructure.

The solution allowed the client to scale manufacturing monitoring, boost productivity, and reduce the risks of machine downtimes.