Speed-Induced Dilation and Prototype Alignment in Federated Bearing Fault Diagnosis
Journal of Applied Science, Engineering and Technology for Development · 16.11.2026 · 6 (1)
DOI: 10.33803/8judye.2 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Applied Science, Engineering and Technology for Development, online erschienen am 16.11.2026. Verfasst von Joseph Munene Disamalu und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Federated bearing fault diagnosis degrades when clients run at different shaft speeds. The literature treats this as generic non-identically-distributed drift and corrects it with general-purpose alignment, leaving unexamined whether the structure of a speed change admits a simpler remedy, and whether a model can be told in advance where it will fail. Because fault frequencies scale with shaft speed, two clients differ by a one-parameter dilation, which on a logarithmic frequency axis becomes a translation of magnitude $\log\rho$; on bearing envelope spectra this holds with no fitted parameters to within 5.6% for ratios up to 1.67. On leave-one-speed-out splits of the public JNU dataset the log axis changes random-forest and logistic-regression accuracy by $-2.6$ and $-0.6$ points, while for a convolutional encoder with global average pooling it yields $+18.8$ points; on CWRU, whose conditions differ by 4.4% in speed, it yields $0.0$ points. For that encoder the log axis is statistically indistinguishable from order normalisation with either the nominal shaft speed or one estimated from the vibration, as in tacholess order tracking ($-1.6$ points, $p = 0.4885$), while both order axes also raise the bin-wise classifiers to at least 84.6%. The log axis therefore removes the speed estimation step rather than outperforming tacholess tracking. Under federation across six clients the log axis raises accuracy at the unseen speed from 59.6% to 81.0%, at a small cost at the training speeds. Prototype alignment adds +1.8 points, which is not significant over 15 paired runs ($p = 0.25$). A derived margin condition tracks accuracy in simulation (Spearman 0.94). This is an association in simulation, where the displacement is measured on paired samples; it does not establish that the condition predicts accuracy on measured data, where it is never met.
Zitieren
APA
Disamalu, J. M., Mwangi, M. K., & Kang'iri, S. M. (2026). Speed-Induced Dilation and Prototype Alignment in Federated Bearing Fault Diagnosis. Journal of Applied Science, Engineering and Technology for Development, 6(1). https://doi.org/10.33803/8judye.2
Vancouver
Disamalu JM, Mwangi MK, Kang'iri SM. Speed-Induced Dilation and Prototype Alignment in Federated Bearing Fault Diagnosis. Journal of Applied Science, Engineering and Technology for Development. 2026;6(1). doi:10.33803/8judye.2
BibTeX
@article{Disamalu2026,
author = {Disamalu, Joseph Munene and Mwangi, Morris Kimathi and Kang'iri, Samuel Macharia},
title = {Speed-Induced Dilation and Prototype Alignment in Federated Bearing Fault Diagnosis},
journal = {Journal of Applied Science, Engineering and Technology for Development},
year = {2026},
volume = {6},
number = {1},
doi = {10.33803/8judye.2}
}
EUVIMED