Nonlinear Modeling of Mechanical Gas Face Seal Systems Using Proper Orthogonal Decomposition

Abstract

A nonlinear reduced-order modeling approach based on Proper Orthogonal Decomposition (POD) is utilized to develop an efficient low order model, based on ordinary differential equations, for mechanical gas face seal systems. An example of a coned mechanical gas face seal in a flexibly mounted stator configuration is presented. The axial mode is modeled, and simulation studies are conducted using different initial conditions and forcing inputs. The results agree well with a fully meshed finite difference model, while the resulting model order is significantly decreased.

Real-Time Force and Moment Estimation for Mechanical Gas Face Seal Systems Using Reduced-Order Kalman Filters

“…Zhang et al [6] developed a computationally efficient reduced-order gas film model applicable to a wide range of operating conditions using POD and Galerkin projection techniques. In this model, a pressure profile approximation, ˆ( , ) p r θ , is generated as the product of time varying weight coefficients, α j (t), and a set of N optimal spatial basis functions (POD modes), φ j (r, θ), as …”

“…Examples include the small perturbation method [2][3][4], the step jump method [4], and the direct numerical frequency response method [5]. However, these analytical and semi-analytical methods are only applicable for small motions about a specific operating point, and the obtained force coefficients are valid only within that range [6]. When the system parameters change, the gas film properties vary and profoundly alter the overall system dynamics.…”

“…In one scheme, the force and moments are considered as independent system states directly. In the other scheme, the gas film force and moments are calculated indirectly from intermediate state estimates of Proper Orthogonal Decomposition (POD) time varying weighting functions, which are components of a reducedorder gas face seal model developed in [6]. The gas film force and moments will then be reconstructed from the estimated weighting functions and the POD modes.…”