Diagnostics and prognostics of rotating beam fatigue crack

This study developed a prognostic methodology to predict the residual life of a rotating beam that fails through fatigue cracking. The rotating beam was chosen because the results could be extended for axial compressor or turbine blade diagnosis and prognosis. An existing FEM crack propagation model, i.e., FRANC (Fracture Analysis Code) was employed to perform the residual life prediction. However, to make a prediction, FRANC requires two inputs that are not readily available, i.e., the current crack size and loading. To establish the utility of FRANC for predicting residual life, this study developed a diagnostic method to estimate crack size and inverse models to estimate loading from the output of a microphone positioned near the beam. First, a signal processing algorithm was used to estimate the modal frequencies of the beam from microphone output that primarily comprised pressure wave excited by the beam vibration. Then a diagnostic neural network was established to map modal frequencies to the crack size. Next, inverse models based on recurrent neural network were established to estimate the impact load from the microphone output. Finally, to predict the growth of a crack and hence the residual life, FRANC is used to extrapolate the current state into future based on the estimated crack size and the impact load.; To validate the proposed methodology, a simple, time-efficient and repeatable experiment was set up from which real measurements of crack size and microphone signal can be taken while a rotating beam goes through its fatigue life cycle. The real measurements taken from run-to-fail tests were used to validate the aforementioned diagnostic and prognostic methods. Experimental results showed that the diagnostic and prognostic methods gave good results.