We gratefully acknowledge support from
the Simons Foundation and member institutions.

Data Analysis, Statistics and Probability

New submissions

[ total of 4 entries: 1-4 ]
[ showing up to 2000 entries per page: fewer | more ]

New submissions for Tue, 19 Oct 21

[1]  arXiv:2110.09099 [pdf]
Title: Vibrational quality classification of metallic turbine blades under measurement uncertainty
Comments: END&CM2021
Subjects: Data Analysis, Statistics and Probability (physics.data-an)

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.

Cross-lists for Tue, 19 Oct 21

[2]  arXiv:2110.08508 (cross-list from hep-ph) [pdf, other]
Title: Improving Variational Autoencoders for New Physics Detection at the LHC with Normalizing Flows
Comments: 10 + 3 pages, 7 figures
Subjects: High Energy Physics - Phenomenology (hep-ph); High Energy Physics - Experiment (hep-ex); Data Analysis, Statistics and Probability (physics.data-an)

We investigate how to improve new physics detection strategies exploiting variational autoencoders and normalizing flows for anomaly detection at the Large Hadron Collider. As a working example, we consider the DarkMachines challenge dataset. We show how different design choices (e.g., event representations, anomaly score definitions, network architectures) affect the result on specific benchmark new physics models. Once a baseline is established, we discuss how to improve the anomaly detection accuracy by exploiting normalizing flow layers in the latent space of the variational autoencoder.

Replacements for Tue, 19 Oct 21

[3]  arXiv:2012.11026 (replaced) [pdf]
Title: Independent Approximates enable closed-form parameter estimation of heavy-tailed distributions
Authors: Kenric P. Nelson
Comments: 30 pages, 8 figures, 7 tables
Subjects: Methodology (stat.ME); Information Theory (cs.IT); Data Analysis, Statistics and Probability (physics.data-an)
[4]  arXiv:2103.09743 (replaced) [pdf, ps, other]
Title: Deep Learning-based Extreme Heatwave Forecast
Comments: 20 pages, 5 figures
Subjects: Machine Learning (cs.LG); Atmospheric and Oceanic Physics (physics.ao-ph); Data Analysis, Statistics and Probability (physics.data-an)
[ total of 4 entries: 1-4 ]
[ showing up to 2000 entries per page: fewer | more ]

Disable MathJax (What is MathJax?)

Links to: arXiv, form interface, find, physics, recent, 2110, contact, help  (Access key information)