Fusing neural networks, genetic algorithms and fuzzy logic for diagnosis of cracks in shafts

Abstract

During the last decades, the engineering community has extensively studied crack identification in rotating machine elements. Although the proposed analytical models may be capable of identifying cracks based on modal analysis, response measurements, or other techniques, the required time for performing the underlying computations is restrictive in real-time diagnosis applications. This chapter introduces a framework for implementing soft-computing techniques, namely artificial neural networks (ANN), fuzzy logic (FL), and genetic algorithms (GA), for identifying cracks in rotating shafts while diminishing the required computational time. Cracks are considered to lie on arbitrary angular positions around the longitudinal axis of the shaft at any distance from the clamped end and characterized by three measures: position, depth, and relative angle. The reduction in computational time is achieved by approximating the analytical model with a neural network and by replacing the exhaustive search of the solution space with a genetic algorithm whose objective function relies on a fuzzy logic representation. Results concerning the efficiency of the proposed framework in terms of accuracy and computational time are presented in the chapter.