LEDs have become an integral part of our everyday lives, whether in the lighting of living spaces, in vehicles or in safety-critical applications such as street and building lighting. They are energy-efficient, multifunctional and offer great scope for special lighting design. Technically, LEDs are power electronic components that convert electrical current into light but also heat. In order to make the advantages of LEDs available to end users, developers of LED modules must ensure that the heat generated can be reliably dissipated from the LED to the environment, i.e. the LED does not heat up excessively. The specified service life of LEDs can only be achieved if this is always ensured. Small changes in the thermal conductivity between the LED and the heat sink, whether due to ageing or mechanical effects, lead to unexpected failures. These increase operating costs and can compromise safety. Researchers are therefore looking for ways to evaluate the installation situation of individual LED modules based on existing measurement data and to predict the service life of individual LED modules. The automatic determination of the thermal characteristics of LED modules after their installation is of central importance for this.
An interdisciplinary team at MCL is working on the thermal modeling of power electronic components and modules, including LEDs. As part of the COMET research project Hybrid20, a physically inspired, probabilistic thermal model of an LED module was created. The determination of the model's parameters was investigated and successfully implemented in collaboration with Prof. Peharz from the Institute for Machine Learning at Graz University of Technology as part of a master's thesis. With the model created, the thermal behavior of LED modules can be better understood and manufacturing fluctuations and different installation situations can be specifically taken into account.
The researchers rely on probabilistic models (i.e. the individual model parameters are modelled as random numbers) and Bayesian optimization to estimate the parameter values (i.e. original assumptions about the distribution are updated based on observations). In this way, existing knowledge - for example about the materials and construction of the LEDs - can be combined with new measurements. By including uncertainties in the model parameters, the reliability of the lifetime estimate based on them can also be evaluated. This is crucial for practical applications, where the decision for or against a service intervention must weigh up the costs against a potential failure.
The inclusion of prior knowledge about the structure of the LED enables an enormous reduction in the model parameters from 200 to 4-8 thermal components. These few model parameters can then be physically interpreted, which significantly simplifies and accelerates the evaluation, making it possible to integrate the monitoring algorithms directly into the control units of lighting units.
Impact and effects
The results show that LEDs can be better understood by combining prior physical knowledge with probabilistic modelling and Bayesian optimization. The developed model provides reliable estimates of the thermal parameters including uncertainties. This creates a solid basis for future lifetime models, which can build on the uncertainty estimates. In turn, these models will make it possible to monitor the service life of LED luminaires in operation in the future and to repair or replace them shortly before they fail and darkness falls.
Project coordination (Story)
Dr. Manfred Mücke
Group Leader Embedded Computing
manfred.muecke(at)mcl.at
IC-MPPE / COMET-Zentrum
Materials Center Leoben Forschung GmbH
Vordernberger Straße 12
8700 Leoben
T +43 (0) 3842 45922-0
mclburo(at)mcl.at
www.mcl.at
Project partners
• Montanuniversität Leoben, Austria, Cair of Automation and Measurement, Austria
• Technische Universität Wien, Informatics, Austria
• Technische Universität Graz, Signal Processing and Speech Communication Laboratory, Austria
• FH JOANNEUM Kapfenberg, Institute Industrial Management, Austria
• Universität Heidelberg, Institute of Technical Informatics, Deutschland
• Linz Center of Mechatronics GmbH, Austria