Bearing Fault Sample Generation Based on Genetic Algorithm

Data-driven PHM (Prognostics and Health Management) has been gaining increasing application in a wide range of industries, which relies on massive samples to ensure diagnostics performance. However, small and imbalanced datasets usually challenge data-driven diagnostics techniques in practical applications. Sample generation has emerged as an efficient solution to this issue. This paper introduces the GA (Genetic Algorithm) to realize sample generation, providing various configurations in steps like selection, crossover, and replacement. To improve the generated samples' similarity in both time and frequency domains, a new fitness weighting of both time and frequency domains is built. Experimental data from the CWRU dataset is used for validation. Results confirm the feasibility of utilizing GA for sample generation.