Data-driven identification of rotating machines using ARMA deterministic parameter evolution in the angle/time domain

Abstract

The functional-series angle-/time-varying autoregressive moving-average (AT-FS-ARMA) model was used to model and analyze vibration-based signals from internal combustion engines. This approach is derived from the formulation of the time-angle periodically correlated processes, a relatively new topic in the cyclostationary framework, which has gained attention for modeling of mechanical signals. The AT-FS-ARMA model consists of traditional time-varying FS-ARMA-like models, but with the projection coefficients expanded in terms of the angular variable, dependent on time. Therefore, the method has the advantage of considering the angle periodicities often present in vibration-based signals from rotating and reciprocating machines. The performance is illustrated by an experimental application of signals measured in a diesel internal combustion engine (ICE) with a constant operating speed. The accuracy of the model is evaluated through the residual sum of squares normalized by the series sum of squares. To illustrate the use of the AT-FS-ARMA for vibration analysis of ICEs, parametric angle-frequency spectrum was estimated and compared to angular-varying pseudo-Wigner-Ville distribution/ spectrum. The results showed that AT-FS-ARMA provides a useful complementary tool for analysis.

A Fast Identification Algorithm for Linear Parameter Varying Vector Ar Models of Short-Term Drivetrain Vibration

“…In the case of rotating machines, the main challenge is to devise a simple modelling methodology regardless of the complexity of the vibration response signals. Various time-series modelling methods have emerged, based on timevarying, parameter-varying or non-linear Auto-Regressive modelling [10,11,12]. In particular, the Linear-Parameter-Varying (LPV) approach can potentially provide efficient representations of the dynamics of rotating machines as a function of a reference shaft angle, and other operational variables.…”

Constrained mode decomposition method for modal parameter identification

“…Classical modal parameter identification methods are usually formulated and operate in time domain or in frequency domain. Time-domain methods include the widely used eigensystem realization algorithm (ERA) [1][2][3] and autoregressive moving average (ARMA) models, [4][5][6] the Ibrahim time domain (ITD) approach, 7,8 and stochastic subspace identification (SSI) method. [9][10][11] Frequency-domain methods include the simplest peak picking method, 12 maximum likelihood identification, 13 and the frequency-domain decomposition method.…”