Extracting a Whisper from the DIN: A Bayesian-Inductive Approach to Learning an Anticipatory Model of Cavitation
For several reasons, Bayesian parameter estimation is superior to other methods for inductively learning a model for an anticipatory system. Since it exploits prior knowledge, the analysis begins from a more advantageous starting point than other methods. Also, since "nuisance parameters" can be removed from the Bayesian analysis, the description of the model need not be as complete as is necessary for such methods as matched filtering. In the limit of perfectly random noise and a perfect description of the model, the signal-to-noise ratio improves as the square root of the number of samples in the data. Even with the imperfections of real-world data, Bayesian methods approach this ideal limit of performance more closely than other methods. These capabilities provide a strategy for addressing a major unsolved problem in pump operation: the identification of precursors of cavitation. Cavitation causes immediate degradation of pump performance and ultimate destruction of the pump. However, the most efficient point to operate a pump is just below the threshold of cavitation. It might be hoped that a straightforward method to minimize pump cavitation damage would be to simply adjust the operating point until the inception of cavitation is detected and then to slightly readjust the operating point to let the cavitation vanish. However, due to the continuously evolving state of the fluid moving through the pump, the threshold of cavitation tends to wander. What is needed is to anticipate cavitation, and this requires the detection and identification of precursor features that occur just before cavitation starts.
- Research Organization:
- Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States)
- Sponsoring Organization:
- USDOE Office of Science (SC)
- DOE Contract Number:
- AC05-96OR22464
- OSTI ID:
- 7463
- Report Number(s):
- ORNL/CP-102987; 60 03 01 01 0; ON: DE00007463
- Resource Relation:
- Conference: Artificial Neural Networks in Engineering, ANNIE '99, St. Louis, MO, November 7-10, 1999
- Country of Publication:
- United States
- Language:
- English
Similar Records
Developing a CD-CBM Anticipatory Approach for Cavitation - Defining a Model Descriptor Consistent Between Processes
An Anticipatory Model of Cavitation