Options
2017
Conference Paper
Titel
Fault detection for heating systems using tensor decompositions of multi-linear models
Abstract
A model-based fault detection method for heating systems is proposed. Two examples of heating system units are under investigation. These systems can be represented as multi-linear systems. Subspace identification methods are used to identify linear time-invariant models for each operating regime, resulting in a parameter tensor. In case of missing data and models for some operating regimes, an approximation method is proposed, where the canonical polyadic tensor decomposition method is used. Low rank approximations are found using an algorithm specialized for incomplete tensors. The tensor of these approximations defines the models in operating regimes, where no measurements were available. Fault detection is done using parity equations and application examples using real measurement data of a heat generation unit are given.