Loading...

Data Analysis for Damage Detection Using Machine Learning Algorithms

Safaei, Mohammad Hossein | 2015

544 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 47310 (45)
  4. University: Sharif University of Technology
  5. Department: Aerospace Engineering
  6. Advisor(s): Abedian, Ali
  7. Abstract:
  8. The pressurized structure such as aircraft fuselage and pipeline (in the oil and water transportation) have a great importance in today’s world. There is a growing tendency in improving design factors, safety and maintenance of such valuable structures. Change of design’s philosophy during the time from infinite life to safe damage and safe life and after that damage tolerance, are a process leading to many changes in design and consequently the methods of inspection, evaluation and maintenance. In this way, using old methods like non-destructive test and evaluation (NDT/E) is encountered with different problems when damage is tiny and there is no coincidence in examination (inspection) process which result in delaying the working process and need a lot of time to detect damage in big inspection. These methods also depend on the user performance so because of such limitations, those lose their proficiency and using new methods that, at the first place, can give the information for large inspection and also can recognize the existence, location, type, the reasons for occurrence and the origins of flaws in-situ are significant. Recently, there is growing tendency to assess the performance of structure and detect any anomaly and damage before the critical situation with structural health monitoring that is considered as a practical method for evaluate the safety of structure and the main goal is to detect abnormal behavior which in term show the undesirable structural conditions. In this study, full assessment is considered on the pipeline as a pressurized structure. The crack and corrosion are considered as the main damages that occur commonly in pipeline. The main goal for modeling such damage on pipeline is to generate data to detect damage existence with some algorithms and evaluate (interpret) moving principal component analysis (MPCA) and robust regression analysis (RRA) as two methodologies for automatically detect damage (which result in abrupt change in measurement data obtained from long term static monitoring). Performance of both methods are evaluated in the presence of virtual noise and outlier. This thesis includes the study about evaluating the performance of the algorithms and the capability of model-free data analysis that detect anomalous behavior in structure (data generated from FEM model) when the number of sensors around damage are reduced (change). In addition, the result of changing damage severity has been assessed
  9. Keywords:
  10. Machine Learning ; Damage Detection ; Artificial Neural Network ; Structural Health Monitoring ; Data Analysis ; Pipelines Health Monitoring

 Digital Object List

 Bookmark

No TOC