Vibrational quality classification of metallic turbine blades under measurement uncertainty

Liangliang Cheng, Vahid Yaghoubi Nasrabadi, Wim Van Paepegem, Mathias Kersemans

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

Abstract

Non-destructive testing on metallic turbine blades is a challenging task due to their complex geometry. Vibrational test-ing such as Process Compensated Resonance Testing (PCRT) has shown an efficient approach, which first measures the vibrational response of the turbine blades, and then employs a classifier to determine if the quality of the turbine blades are good or bad. Our previous work mainly concentrated on the development of Mahalanobis distance-based classifiers which are fed by the measured vibrational features (such as resonant frequencies). In practice, however, measurement errors could lead to a biased trained classifier, potentially resulting in the wrong quality classifications of the turbine blade. In this study, we investigate the classification problem of turbine blades under measurement uncertainty. For this, the concept of Interval Mahalanobis Space is employed, leading to the Integrated Interval Mahalanobis Classification system (IIMCS) classifier which has high robustness against measurement uncertainty. The IIMCS employs Binary Particle Swarm Optimization (BPSO) to filter out those resonant frequencies which contrib-ute most to the information in the system. A Monte Carlo Simulation scheme is employed to analyze the sensitivity of the resonant frequencies to the measurement uncertainty. This yields an indicator of reliability, indicating the confidence level of the final classification results. The developed IIMCS methodology is applied to an experimental case study of equiaxed nickel alloy first-stage turbine blades, showing good and robust classification performance.
Original languageEnglish
Title of host publicationEuropean NDT&CM2021
Number of pages6
Publication statusPublished - 2021
Externally publishedYes
EventEuropean NDT & CM2021 - Prague, Czech Republic
Duration: 4-Oct-20217-Oct-2021

Conference

ConferenceEuropean NDT & CM2021
Country/TerritoryCzech Republic
CityPrague
Period04/10/202107/10/2021

Fingerprint

Dive into the research topics of 'Vibrational quality classification of metallic turbine blades under measurement uncertainty'. Together they form a unique fingerprint.

Cite this