This paper addresses the problem of analyzing measurement data to estimate the variations in turbine blade life in the presence of manufacturing variability. A methodology that employs existing denoising techniques, namely, Principal Component Analysis and Fast Fourier Transform analysis, is proposed for filtering measurement error from the measured data set. An approach for dimensionality reduction is employed that uses prior knowledge on the measurement error obtained from analyzing repeated measurements. The proposed methodology also helps in capturing the effects of manufacturing drift with time and the blade to blade manufacturing error. The filtered data is then used for generating three-dimensional representations of probable manufactured blade shapes from the limited number of available measurements. This is accomplished by using a Free-Form Deformation based approach for deforming a nominal mesh to the desired shapes. Estimations of life on the probable turbine blade shapes manufactured over a span of 1 year indicate a reduction of around 1.7 in the mean life relative to the nominal life, with a maximum relative reduction of around 3.7, due to the effects of manufacturing variability.
展开▼
