Fault detection and diagnostics

In many buildings, systems and components are operated far from their optimum. Frequently occurring faults include incorrect operating times, faulty sensors, defective actuators such as valves and dampers and inadequate control strategies.

Fraunhofer ISE uses mathematical methods to detect faults in the operation of building systems and components in near real time. In order to optimally monitor a system and maintain energy-efficient operation, numerous signals must be observed and evaluated in their temporal course and their interdependence. Using machine learning methods, building systems such as heat pumps or heating circuits can learn their optimum operating mode themselves and generate a diagnostics for the operator in the event of deviations between measured and predicted values. With feedback, the operator can further increase the performance of the self-learning method. Thus, a higher transparency of the system operation can be achieved and energy and costs can be saved.

We develop data processing algorithms for monitoring the operation of building systems with the help of which faulty operating conditions can be automatically identified and reported to operators. The algorithms are based on expert knowledge, on methods of machine learning such as qualitative models, decision trees or on a combination of these methods. Together with software and device manufacturers, we test and integrate our methods into cloud-based platforms or directly into devices as edge computing solutions.

Our R&D Services include:

• Automatic fault detection
  We develop, test and integrate methods for automatic fault detection in the operation of building systems. We use machine learning methods and expert knowledge and achieve a high degree of flexibility, adaptability and accuracy by selecting a suitable method or combining different methods.
  

• Self-learning diagnostic systems 
  We develop diagnostic systems that recognize and diagnose errors learned from training data with high accuracy. In addition, we use methods that detect unknown anomalies in operation and assign them to individual sensors or subsystems. The direct involvement of the user in the diagnostic process enables a tailor-made adaptation of the algorithms and an increasing accuracy of the diagnoses in the course of the application.

• Integration on cloud-based platforms or in devices
  The fault detection and diagnostics methods developed by Fraunhofer ISE can be integrated into cloud-based platforms or directly into the manufacturer's devices and components. In addition, the methods can be directly integrated into the data analysis platform "mondas", which is now being further developed by the Fraunhofer ISE spin-off mondas GmbH.
  

• Coupling with BIM data
  If required, existing digital descriptions of certain system parts or components can be used by the customer to parameterize the algorithms for error detection and enrich the diagnostics. Linking the operational monitoring with additional digital information about the relevant system reduces setup effort and increases benefit.

• User-friendly alarm management
  When a fault is detected in the system, corresponding alarms can be generated in the form of emails, push messages to a mobile device or directly in the platform used. Additional information such as the time range of the fault, sensors involved, system and building parts affected and remedial actions can be linked to a BIM model or CAFM system to facilitate maintenance.

• Energy and cost savings
  In various national and international research projects and monitoring contracts we were able to achieve energy savings of 20% and in some cases even more. Such savings potentials can most of the time be realized through cost-effective measures such as fault correction and the adaptation of control strategies.