Adaptive Subband Decomposition with Envelope Spectrum Consistency Voting for Cross-Domain Bearing Fault Feature Certification

Abstract Subband decomposition (SBD) provides a computationally efficient framework for bearing fault diagnosis, but existing implementations assume uniform spectral partitioning that does not account for dataset-specific resonance distributions. This work proposes an adaptive SBD variant in which filter bank edges are determined by a consistency voting procedure over envelope spectrum peaks, selecting the K frequency bands that appear most persistently across signal windows. The method couples adaptive decomposition with a feature certification protocol based on mutual information and Jensen-Shannon divergence bootstrap confidence intervals. Certification is evaluated across six transfer pairs spanning three heterogeneous bearing datasets — MFPT, Paderborn, and Fraunhofer LBF — with distinct fault mechanisms, sensor configurations, and operating conditions. Adaptive SBD with K=8 subbands and RMS inter-domain normalisation yields three certified-invariant features (kurtosis, shape factor, and clearance factor) against one for uniform SBD, with a 49% reduction in mean bootstrap confidence interval width. Analysis of per-pair certification scores reveals a structural asymmetry between distributional stability and discriminative transferability: the pair MFPT→LBF suffers a catastrophic accuracy drop (0.875→0.077) attributable to filter bank boundary mismatch when the source dataset uses narrow adaptive subbands (≈162 Hz) and the target requires a uniform fallback (1600 Hz), a condition detectable from the band-edge log prior to deployment. The per-window voting mechanism defines a narrow stability basin around its default hyperparameters; small deviations cause reversion to uniform fallback, motivating explicit diagnostic validation for new datasets. Distributional certification is a necessary condition for safe cross-domain deployment but is not sufficient to guarantee discriminative accuracy: a partition compatibility gate, formalised as Stage 0 of the certification pipeline, automatically detects the structural failure mode in which nominally identical subbands cover incompatible frequency ranges.

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-03
DOI
https://doi.org/10.1088/2631-8695/aebd3e
Primary Topic
Machine Fault Diagnosis Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Adaptive Subband Decomposition with Envelope Spectrum Consistency Voting for Cross-Domain Bearing Fault Feature Certification

Alisson V. Brito, P. S. de Almeida, Abel C Lima-Filho
Engineering Research Express
Machine Fault Diagnosis Techniques
article

Adaptive Subband Decomposition with Envelope Spectrum Consistency Voting for Cross-Domain Bearing Fault Feature Certification

Alisson V. Brito, P. S. de Almeida, Abel C Lima-Filho
article en

Abstract

Abstract Subband decomposition (SBD) provides a computationally efficient framework for bearing fault diagnosis, but existing implementations assume uniform spectral partitioning that does not account for dataset-specific resonance distributions. This work proposes an adaptive SBD variant in which filter bank edges are determined by a consistency voting procedure over envelope spectrum peaks, selecting the K frequency bands that appear most persistently across signal windows. The method couples adaptive decomposition with a feature certification protocol based on mutual information and Jensen-Shannon divergence bootstrap confidence intervals. Certification is evaluated across six transfer pairs spanning three heterogeneous bearing datasets — MFPT, Paderborn, and Fraunhofer LBF — with distinct fault mechanisms, sensor configurations, and operating conditions. Adaptive SBD with K=8 subbands and RMS inter-domain normalisation yields three certified-invariant features (kurtosis, shape factor, and clearance factor) against one for uniform SBD, with a 49% reduction in mean bootstrap confidence interval width. Analysis of per-pair certification scores reveals a structural asymmetry between distributional stability and discriminative transferability: the pair MFPT→LBF suffers a catastrophic accuracy drop (0.875→0.077) attributable to filter bank boundary mismatch when the source dataset uses narrow adaptive subbands (≈162 Hz) and the target requires a uniform fallback (1600 Hz), a condition detectable from the band-edge log prior to deployment. The per-window voting mechanism defines a narrow stability basin around its default hyperparameters; small deviations cause reversion to uniform fallback, motivating explicit diagnostic validation for new datasets. Distributional certification is a necessary condition for safe cross-domain deployment but is not sufficient to guarantee discriminative accuracy: a partition compatibility gate, formalised as Stage 0 of the certification pipeline, automatically detects the structural failure mode in which nominally identical subbands cover incompatible frequency ranges.

Engineering Research Express
Universidade Federal da Paraíba (BR), Instituto Federal de Educação Ciência e Tecnologia da Paraíba (BR)
Openalex Percentile: Top 19%
Machine Fault Diagnosis Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon