Methods to Estimate Machine Remaining Useful Life Using Artificial Neural Networks
GEORGIA SOUTHERN UNIV STATESBORO ENGINEERING STUDIES PROGRAM
Pagination or Media Count:
In this paper, a general methodology for remaining useful life estimation based an indirect methodology is presented. Gearbox failure data, recorded using a mechanical test bed at the Applied Research Laboratory, Penn State University, is used. The machine remaining useful life estimation method used in this paper is indirect method, in the sense that it predicts first the behavior of some system parameters known to be sensitive to the machine operating status, use those predicted values in order to find the predicted machine status through the fuzzy system definitions, and then estimate the remaining useful life by measuring the time from the present time to the time where the death status was detected. Some machine parameters such as temperature, vibration spectrum and level, and acoustic emission, are used in such analysis. Machine operating regions are divided into normal operation, abnormal operation, and no operation or death. Every parameter limits is defined in each region. Prediction models are used to predict the time trajectory of the machine parameters starting from some history measurements. Those predicted trajectories could be used to determine the machine death status point in time. The remaining time to death can be estimated form such models within some appropriate certainty and error tolerance. Neural networks and fuzzy logic system modeling techniques are used for machine parameter prediction due to their known ability for non- linear system modeling, robustness, generalization, and modeling decision uncertainty.
- Theoretical Mathematics