Loading...

A Data Mining Approach for Prognostics and Health Monitoring Using Age Based Clustering: A Case Study on a Gas Turbine Compressor

Mahmoudian, Ali | 2020

463 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 53229 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Durali, Mohammad; Saadat, Mahmoud
  7. Abstract:
  8. In recent years health monitoring and prognostics of complex systems have been considered more than ever.. In the present researh, data - based approach has been selected among various prognostics and health monitoring approaches. One of the most challenging issues in data-based methods is how to map system sensor information to its health status. In this research, different methods of mapping are discussed. The results show that sensor fusion by principal component analysis (PCA) offers acceptable performance. This pattern produces a single-dimensional signal for health monitoring with high reliability.The second challenge is to predict the status and design of the prognostics module. For this purpose, two common methods of regression and neural network have been used. Performance analysis of designed prognostics modules showed that using a unique and common prognostic module to predict the life of units with different age conditions causes large errors. Therefore, the method of prognostic using age based clustering (ABC) is presented in this project. The main idea is that the base age of the unit is involved in prediction. This intervention is done by age clustering of test data, then reproduction of train data for each cluster, subsequently creating and training of prediction modelus for each cluster and finally life estimation of the test unit. In this thesis, a step-to-step algorithm for implementation of ABC is introduced, which consists of grouping, reproduction, mapping and finally estimation of the remaining useful life. The results show that the concept of clustering is efficient in prognostics; even without any optimization processes and for its simplest form of performance. The effectiveness of ABC method was demonstrated in comparison with other methods. The proposed structure can be combined with many forecasting methods and can significantly improve their accuracy.The main parameters of ABC model are discussed and genetic algorithm is used to find the optimal model parameters. Finally, fuzzy technique is used to assign the unit under test to the appropriate cluster. By optimizing the parameters of the structure and the fuzzy clustering, the accuracy of the prognostics is increased.For the case study, prognostics datasetes from the NASA Prognostic Data Repository are used. The data set is known to be one of the best and most widely used test data to develop and validate algorithms
  9. Keywords:
  10. Reliability ; Sensor Data Fusion ; Health Monitoring ; Fracture Prediction ; Data Driven Method ; Prognostics Failure ; State Prediction ; Principal Component Analysis (PCA) ; Artificial Neural Network

 Digital Object List

 Bookmark

No TOC