ABB Uses MATLAB to Operationalize Causal AI Models
Approach Facilitated the Development of a Causal Fault Analytics System
“The microservices functionality … enables the deployment of models based in MATLAB that are scalable microservices. [This functionality offers] simplified and seamless integrations to other systems and frameworks.”
Key Outcomes
- MATLAB Compiler SDK enabled microservices with Docker containers to develop a customized pipeline
- RestAPI communication ensured data exchange with existing workflows on Amazon Web Services
- An automated deployment pipeline eliminated the need for recoding with a significant time reduction in manual translation, integration, code audit, and functional testing
ABB, a global leader in electrification and automation, is harnessing causal AI models based on cause-to-effect data relationships. These models provide better insights into customers’ decision-making processes. Any AI model faces deployment challenges such as manual translation, CI/CD pipeline issues, integration complexities with enterprise-level existing frameworks, and data exchange limitations. To address these challenges, the ABB team operationalized machine learning models in MATLAB® as containerized microservices and integrated them with existing workflows. As a result, a seamless pipeline enabled the development of a causal fault analytics expert system.
MATLAB Compiler SDK™ packaged the MATLAB functions into a format that can be easily shared and used in other applications. These functions, in turn, were used to create a Docker® image. AI algorithms were then deployed as a microservice to provide an endpoint and accept RESTful requests. The MATLAB microservice was deployed on Amazon® Elastic Container Service with Fargate, and a load balancer was used to distribute incoming requests evenly across multiple instances of the microservice to ensure smooth operation. Finally, Flask, a web framework, was connected to Amazon Simple Storage Service and DynamoDB to manage data storage and retrieval.
