Acoustic Emission Diagnosis of Lubricant Water Contamination Fault by Using Adaptive Multi-fidelity Bayesian Optimization

In rotating machinery, water contamination in lubricants can lead to oil film failure and exacerbate equipment wear. The acoustic emission (AE) technology demonstrates superior performance over traditional detection methods for early fault detection due to its sensitivity to weak signals. Regarding the specific features of AE signals during oil film shearing, the traditional Bayesian method lacks comprehensive diagnosis and is constrained by single-fidelity optimization, dedicating most evaluations to high-cost configurations. Accordingly, this paper proposes an adaptive Multi-fidelity Bayesian Optimization (MFBO) method for detecting water contamination in lubricants. The approach denoises acoustic emission (AE) signals by combining the detection index (DI) and the interquartile range (IQR) with wavelet packet transform (WPT), which markedly improves the signal-to-noise ratio. Within MFBO, an adaptive penalty parameter α dynamically adjusts to reduce costly evaluations. This framework is used to optimize CNN hyperparameters, including the learning rate, mini-batch size, and number of training epochs. Experiments on twelve groups at different shear rates yield a recognition accuracy of 99.15% for water contamination, representing an 18.48% improvement over the non-optimized baseline; introducing α further reduces computation time by 7.35%. Overall, the method achieves an effective balance between exploration and exploitation, maintaining reliable identification across varying moisture levels.