A new approach for the identification of reduced thermal models

This paper describes a method aimed at producing reduced order models in state-space form for speeding up the transient thermal simulation of mechatronics products. Designed in a non-intrusive framework, this method is implemented as an offline tool that post-processes the inputs and outputs of a detailed reference model considered as a black box. The reduction algorithm is based on robust linear algebra tools, such as the householder decomposition and the least square approximation. Thanks to the underlying modal representation, the model order, typically equal to the number of observation points, can be also further reduced. A first linear test case illustrates the quality of the results. The investigation of various input signals ensuring the reliability of the identification process while minimizing the simulation time of the detailed model, as well as the derivation of non-linear reduced models, are the ongoing paths of improvement and extension.