| Performance prediction for hydro-mechanical systems is a difficult problem. Extreme system operating conditions cause the sensors measuring the performance of critical components to fail or lose accuracy over time, making failure diagnosis difficult. In the case of a weapon system, the prediction model must account for many variables such as: mechanical wear, elevation, firing rate, temperature and environmental conditions that influence weapon performance. Information from the model must be useable for reliability, availability and maintainability calculations. Finally, the model should be adaptable to monitor weapon production testing at manufacturing and rebuild locations.;The diagnostic and prognostic model was developed using standard weapon performance data from a 10,000 large caliber durability test. Since no major failures occurred during testing, an artificial neural network prediction model was built to diagnose "high stress" conditions in the weapon's output performance attribute.;To develop a model capable of diagnosing small variations in weapon performance, careful attention was given to network architecture, input and output attributes selection, and performance measurement. A unique output encoding system was developed to utilize the model's sensitive performance monitoring ability for prediction of the weapon's health up to 10 rounds in the future.;This research determined the probability of the weapon operating properly, and also generated a prediction model for future rounds, to create a survival curve, to predict if a weapon failure would occur in the near future. The failure diagnosis results from the model were good and will improve when actual failures are present in the test data. The model also generated accurate prognostic data to predict future weapon failures.;This research is very applicable to industry, especially for monitoring hydro-mechanical systems during testing, or use in the field. The research can also be applied to almost any other type of equipment monitoring. The same prediction model developed to predict weapon failure can be used to evaluate system reliability during production testing, or monitor system performance. The model can be used to monitor equipment where critical performance parameters are difficult to monitor because of temperature, shock, or vibration. |
