We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Physics > Data Analysis, Statistics and Probability

Title: Vibrational quality classification of metallic turbine blades under measurement uncertainty

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.
Comments: END&CM2021
Subjects: Data Analysis, Statistics and Probability (physics.data-an)
Cite as: arXiv:2110.09099 [physics.data-an]
  (or arXiv:2110.09099v1 [physics.data-an] for this version)

Submission history

From: Liangliang Cheng [view email]
[v1] Mon, 18 Oct 2021 08:44:52 GMT (722kb)

Link back to: arXiv, form interface, contact.