Enhancement of cyclic spectral coherence map by statistical testing approach—application to bearing faults diagnosis in electric motors

Abstract Efficiency of fault detection in rolling element bearings is heavily influenced by the quality of data. In controlled environments, such as test rigs designed for bearing diagnostics, data quality is relatively good. Similarly, diagnosing bearings that support shafts in industrial machinery is relatively straightforward. However, diagnosing bearings in electric motors presents greater complexity due to the influence of additional cyclic components on vibration signals. These extra components, originating from mechanical or electrical sources, complicate frequency-based analysis. This paper proposes a novel approach for diagnosing bearings in electric motors, utilizing statistical analysis within the bi-frequency domain through a cyclostationary framework. The method involves applying a statistical testing procedure to individual pixels on the cyclic spectral coherence (CSC) map. The statistical significance of these pixels is assessed based on quantiles of CSC maps obtained from a dataset representing a healthy bearing. This process results in an enhanced or cleaned CSC map, facilitating the identification of fault-related components. Consequently, this approach enables the detection of defects in electric motor bearings, even when additional signal components unrelated to the defect, but characteristic of a healthy bearing, are present.