Modelling the Labyrinth Seal Discharge Coefficient Using Data Mining Methods

This paper presents a data-based method to predict the discharge coefficients of labyrinth seals. At first, leakage flow rate data for straight-through and stepped labyrinth seals from various sources was collected and fused in one consistent data base. In total, over 15000 data points have been collected so far covering a 25-dimensional design space. Secondly, this leakage data set was analysed using open-source Data Mining software, which provides several algorithms such as Multiple Linear Regression (MLR) and Artificial Neural Networks (ANN). The suitability of MLR and ANN for modelling labyrinth discharge coefficients and analysing system sensitivity was tested and evaluated. The developed leakage models showed promising prediction qualities within the design space covered by data. Further improvements of model quality may be achieved by continuing data analysis using advanced methods of Data Mining and enlarging the existing data base. The major advantages of the presented method over numerical or analytical models are possible automation of the modelling process, low calculation efforts and high model qualities.