Impact of Epistemic Uncertainty on Performance Parameters of Compressor Blades

Abstract An well-established tool for probabilistic analysis is the Monte Carlo simulation (MCS), which can be used to gain insight into an unknown system behavior. The variability of the input variables is usually defined by parametric probability density functions (PDFs) or parametric cumulative density functions (CDFs). An example is the Gaussian normal distribution, which is characterized by its mean and standard deviation. When analyzing compressor blades, e.g. in the context of robust design optimization, the variation of profile variables such as chord length or stagger angle may be of interest. The distribution parameters are oftentimes obtained from a limited set of optical measurements. When performing the MCS, the distribution parameters must be fixed. However, the use of fixed parameters neglects the uncertainty and the effects resulting from the uncertain distribution parameters. In this paper, the influence of epistemic uncertainty is illustrated using the example of compressor blades, which are subject to both manufacturing variability as well as wear and tear. Only a limited set of optical measurements is available. The resulting effect of the epistemic uncertainty is quantified using parametric probability boxes. The results of this approach are compared with the result from classical confidence interval analysis. This allows an estimation of the required number of measurements to achieve the desired statistical confidence.