Effect of Sampling Rate on Parametric and Non-parametric Data Preprocessing for Gearbox Fault Diagnosis

PurposeData preprocessing is one of the key steps in any fault diagnosis process. The real data obtained from machines carry a lot of noise and inferred signals from other parts of the machine or environment. The intensity of this contamination varies with the sampling rate of data acquisition. To filter out these components and enhance the quality of the features generated from these data, several data preprocessing techniques are described in the literature. But the major concerns are the limitations of these techniques and the proper selection of sampling rates for data acquisition.MethodsThis paper presents a comprehensive overview of parametric and non-parametric data preprocessing techniques for gearbox fault diagnosis and how these techniques preserve their properties under different sampling rates. Both analytically simulated signals and experimental signals are used in this work to check the effectiveness of these techniques at different sampling rates.Results and ConclusionsThe obtained results clearly show that data preprocessed by a non-parametric filter contains significantly more information than data preprocessed by a parametric filter or without a filter. Even for a low (affordable) sampling rate, the non-parametric filter works well as compared to the parametric filter and with no filter. The proposed work has potential relevance in the industrial IoT for online condition monitoring of gearboxes.