Build Scalable AI Solutions with MATLAB Production Server in Kubernetes on Azure

With more than 150 years of experience in the industry, the AERZEN Group is one of the top 3 application specialists for high-performance blowers. The compressors, blowers, and turbos are mainly used in wastewater treatment plants, in the process industry, and in oil-free pneumatic conveying of bulk materials. Sustainability, smart energy, resource usage, and reliability of machinery are important concerns for AERZEN’s customers when operating their plants all over the world. The Aerzen Digital Systems business unit is working on smart services so that users can operate the machines even more efficiently and reliably with the help of AI and machine learning.

Aerzen Digital Systems designs AI models for forecasting, condition monitoring, and predictive maintenance for this purpose. Functions and models may be used interactively for data exploration as well as automatic processing of streaming IoT data. While planning to operationalize these models, several challenges surfaced and certain requirements for the platform were defined:

• Flexibility and scalability for many unique plants and sets of machinery
• MLOps to monitor and adapt AI models over decades of equipment lifetimes
• Agile and quick deployment of new functionality on availability of improved AI techniques
• Integration with applications from different frameworks developed both in-house and by customers

In this talk, we detail a sample solution to these challenges centered around MATLAB Production Server™ running in Kubernetes on Microsoft Azure. The Aerzen Digital Systems libraries of MATLAB functions and AI models are deployed to MATLAB Production Server through an MLOps pipeline. In the present case study, data from a large wastewater treatment plant is analyzed and an anomaly detection algorithm for a single blower is developed. This model is then uploaded to the cloud and trained individually for all blowers. During runtime execution, every model is monitored and automatically retrained if necessary.