id: 05953700
dt: a
an: 05953700
au: Dai, Xuewu; Breikin, Timofei; Wang, Hong; Kulikov, Gennady; Arkov, Valentin
ti: Modeling a complex aero-engine using reduced order models.
so: Mohammadpour, Javad (ed.) et al., Efficient modeling and control of
    large-scale systems. New York, NY: Springer (ISBN
    978-1-4419-5756-6/hbk; 978-1-4419-5757-3/ebook). 89-111 (2010).
py: 2010
pu: New York, NY: Springer
la: EN
cc: 
ut: model identification algorithm; data driven reduced order modeling; dynamic
    nonlinear least squares (DNLS) identification; Hessian approximation
ci: 
li: doi:10.1007/978-1-4419-5757-3_4
ab: Summary: This chapter describes the modeling phase of the engine
    model-based condition monitoring. The aim is to find a fast OE model
    identification algorithm suitable for the limited on-board computation
    facilities of the on-board condition monitoring system. The chapter is
    organized as follows. Section 2 provides an introduction to gas turbine
    systems. Section 3 discusses the performance criterion and structure
    selection during data driven reduced order modeling. Section 4 defines
    the objective function for OE modeling and presents the development of
    DNLS (Dynamic Nonlinear Least Squares) identification algorithm. In
    order to accelerate the identification speed, two techniques are
    employed. The first one is iterative calculation of the gradient (the
    Jacobian). The dependency within output errors is taken into account by
    adding an weighted sum of past gradients to the current Jacobian
    matrix. Secondly, Hessian approximation is used. In order to further
    accelerate the convergence speed at a relatively low computation cost,
    the second order Hessian matrix is used and approximated by the first
    order information. Finally, the potential of the proposed DNLS for
    reduced order modeling of a gas turbine engine is illustrated in Sect.
    5. Data gathered at an aero engine test-bed serves as test vehicle to
    demonstrate the improvements in convergence speed and the reduction of
    computation cost.
rv: