"ALFA"

The challenge​

Buildings account for a significant portion of energy consumption in the EU and Austria, estimated at 40% and 25%, respectively. Studies have shown that up to 50% of heat pumps operate at only 70-80% efficiency due to a lack of maintenance activities.

By detecting, diagnosing, and repairing faults in heat pumps early on, energy loss can be reduced by approximately 40%. Utilizing smart energy services, such as diagnosis, can enhance energy efficiency and reduce CO2 emissions. Further, diagnosis and repair are crucial components of maintenance that significantly extend the lifetime of products. There are generally two AI diagnosis avenues: knowledge-based, e.g., model-based diagnosis (MBD), and data-driven, e.g., machine learning (ML), approaches. While each research direction has been applied to industrial applications, it becomes crucial to adopt a holistic approach that combines these methodologies to leverage the strengths of both methods, offering a more robust and comprehensive solution.
​​

The project content​

In the project ALFA, we want to foster synergies between model-based and machine learning techniques in the context of fault diagnosis of building automation sector. Based on a set of real-world benchmarks, we will first develop a separate MBD and ML approach in the application domain. Second, we will focus on ways to integrate the methods unique strengths and capabilities. There is generally a lack of data for a newly commissioned building known as the cold start problem. To mitigate this issue, we will exploit simulations based on the MBD model to generate a synthetic diagnostic data set for training the ML model. Additionally, we emphasize the interpretability of ML diagnostics via Explainable AI (XAI), promoting transparency in decision-making processes. Finally, we will explore further synergy opportunities (e.g., incorporating probabilities into MBD based on data from the ML diagnostic process) to set new standards for the accuracy and efficiency of diagnostics in building automation.

​​

The innovation​

​

​Despite the vast amount of research on fault diagnosis using MBD and ML, each of the approaches still has many open research issues that need to be solved for use with HVAC systems of large commercial buildings. While ML methods have been applied to certain aspects of knowledge-based diagnosis, there is currently no framework that effectively combines these two diagnostic routes. Thus in ALFA, we focus on the combination of model-based and ML diagnosis to counteract the approaches’ weaknesses in the context of HVAC systems.