الفهرس 
Prognostics and Health ManagementOnly 14 pages are availabe for public view
 |  | from | 206 |  |  |
| | | from | 206 | | |
Abstract
Complex engineering systems suffer from internal wears and tears that cannot be measured by sensors. Sudden failure of such systems is hazardous and may endanger human lives. On-board real-time remaining useful life (RUL) estimation of complex engineering systems prevents sudden failures and safes lives. The proposed research is intended to solve the problem of RUL estimation of aircraft turbofan engines on-board in real-time. The decision maker can use prognostics information to take appropriate decisions and corrective actions to maintain safety and performance in the meantime. The on-board real-time RUL estimation problem is solved in this study by development of on-board prognostics system for real-time RUL estimation of aircraft turbofan engines based on data-driven approach. This system takes the multivariate sensors and operational data and adds indicators about historical engine run for RUL estimation. Kalman filter and artificial neural networks are used together for accurate engines’ health state prediction. The predicted health state is then projected into future for RUL estimation. The proposed prognostics system continuously predicts engine health state (health index) until the prediction point is reached. The system uses the last ten values of engines’ health indices (HIs) and project it to the future to forecast the damage progression for RUL estimation. To achieve the research objective, the proposed prognostics system is developed through three main stages: training, testing, and implementation. All heavy computational tasks are performed offline in the training stage such as: data preprocessing and artificial neural networks training. The neural networks are trained for HI inference and projection. In the testing stage the onboard part is developed and the system output correctness is evaluated. The trained neural networks and Kalman filter are connected together to work in a homogeneous way in the form of algorithm for HI prediction and RUL estimation. Mean squared error, average scale independent error, and average bias are used for system output correctness evaluation. The implementation stage is considered as the harvest of the previous stages. In this stage the online part is implemented on a single board computer as a universal windows platform application. This implementation brings the internet of things technology to prognostics which increases its technology readiness level for onboard deployment as needed by the prognostics and health management community. Data from PHM08 data challenge competition is used for system development and testing. Results showed that the proposed methodology can be used for RUL estimation onboard in real-time with high prediction and computational performances.