Gas Turbine Health Prognostics Using Artificial Neural Network
The field of prognostics has gained the attention of companies in effort to reduce costs or losses by predicting the equipment’s future life. This project aims to develop a prognostics model that is capable of accurately predicting the remaining useful life (RUL) of gas turbines using an artificial...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Year Project |
| Language: | English |
| Published: |
IRC
2017
|
| Subjects: | |
| Online Access: | http://utpedia.utp.edu.my/17977/1/Gas%20Turbine%20Health%20Prognostics%20Using%20Artificial%20Neural%20Network.pdf http://utpedia.utp.edu.my/17977/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The field of prognostics has gained the attention of companies in effort to reduce costs or losses by predicting the equipment’s future life. This project aims to develop a prognostics model that is capable of accurately predicting the remaining useful life (RUL) of gas turbines using an artificial neural network model. The model developed is to incorporate data fusion and can handle multiple-sensory signal input. Lastly, suitable application guidelines for the practical usage of the prognostics model is proposed. The Turbofan Engine Degradation Simulation Data Set published by Saxena and Goebel is used as the benchmark dataset in this research.
In this research, the modified Wu’s method ANN model structure is tested in combination with wavelet-based denoising and exponential function fitting. The ANN model takes in input in the form of past two time-step parameters, and outputs the gas turbine’s life percentage, which can be further used to calculate the gas turbine RUL. This model has the advantage of being capable of handling multiple-sensory input at once The prediction results are evaluated using five parameters: total score, MAPE, MAE, RMSE, and correlation coefficient. The proposed ANN model has shown best results using original data only without denoising or exponential function fitting. Without any data filters, the ANN model scored 1,093 for the FD001 with a MAPE of 27%, MAE of 17, and RMSE of 23. Study from this research have also shown an increase in prediction accuracy when data fusion is incorporated in the form of increasing number of input variables.
Analysis of results show that the ANN model has similar prediction effectiveness as other publications in terms of MAPE and MAE. However, the ANN model shows higher score value due to its bias of late predictions. Like other health-based prognostics models, this ANN model suffers from data insufficiency in practical applications due to the unknown current life percentage of the gas turbine. |
|---|