13 November 2024 in IEEE Access
Bearing fault diagnosis is a well-developed field and an active area of research in which the combination of model-free machine learning techniques with vibration data has become a popular approach. However, vibration data from rotating machines has the potential to contain domain shifts beyond the accepted causes in this research area (different part models, operating conditions and sensor locations) which can enable data leakage between training and test datasets. To demonstrate the impact of data leakage, six common bearing diagnosis methods are applied to two datasets using three data splitting methods to compare classification performance. Diagnosis is preformed using Principal Component Analysis (PCA), Supervised Principal Component Analysis (SPCA) and Linear Discriminant Analysis (LDA) in combination with frequency analysis and envelope analysis feature extraction methods. Datasets from McMaster University and Paderborn University are used as experimental data sources, and produce vastly differing results (over a 40% drop in accuracy) depending on the selected dataset splitting method, revealing a previously unknown domain shift. Despite great results for diagnosis methods using frequency response analysis on the data from McMaster, these results are not expected to generalize due to possible data leakage. Out of fifty-five previous works using the Paderborn dataset, ten are identified as likely to be affected and only six properly address the problem. Recommendations are given for future experiment design, model creation and model evaluation.
12 September 2025 in IEEE Access
The Bayes Error Rate (BER) is the fundamental limit on the achievable generalizable classification accuracy of any machine learning model due to inherent uncertainty within the data. BER estimators offer insight into the difficulty of any classification problem and set expectations for optimal classification performance. In order to be useful, the estimators must also be accurate with a limited number of samples on multivariate problems with unknown class distributions. To determine which estimators meet the minimum requirements for ‘‘usefulness’’, an in-depth examination of their accuracy is conducted using Monte Carlo simulations with synthetic data in order to obtain their confidence bounds for binary classification. To examine the usability of the estimators for real-world applications, new non-linear multi-modal test scenarios are introduced. In each scenario, 2500 Monte Carlo simulations per scenario are run over a wide range of BER values. In a comparison of k-Nearest Neighbor (kNN), Generalized Henze-Penrose (GHP) divergence and Kernel Density Estimation (KDE) techniques, results show that kNN is overwhelmingly the more accurate non-parametric estimator. In order to reach the target of an under 5% range for the 95% confidence bounds, the minimum number of required samples per class is 1000. As more features are added, more samples are needed, so that 2500 samples per class are required at only 4 features. Other estimators do become more accurate than kNN as more features are added, but continuously fail to meet the target range.