logo SBA

ETD

Digital archive of theses discussed at the University of Pisa

 

Thesis etd-01292011-180645


Thesis type
Tesi di dottorato di ricerca
Author
OPREA, CONSTANTIN-VIOREL
URN
etd-01292011-180645
Thesis title
Methods and Algorithms for 3D Characterization of Defects in Piping Inspection performed with a Guided Wave Multichannel Magnetostrictive System
Academic discipline
ING-IND/31
Course of study
APPLIED ELECTROMAGNETISM IN ELECTRICAL AND BIOMEDICAL ENGINEERING, ELECTRONICS, SMART SENSORS, NANO-TECHNOLOGIES
Supervisors
tutor Raugi, Marco
Keywords
  • defect characterization
  • guided wave
  • NDT
Graduation session start date
11/04/2011
Availability
Full
Summary
The Guided Wave NDT technique (GW-NDT) is a relatively new non-destructive testing method used for long-range inspection and monitoring. Guided waves are mechanical or elastic waves propagating in a confined medium (plates, pipes), at ultrasonic and sonic frequencies traveling along the structure, guided by its geometrical borders. They can propagate in various frequencies and modes: torsional, flexural, and longitudinal.
GW-NDT can serve as a useful tool for evaluating the integrity of metallic structures as integrated parts of in petrochemical industrial plants, used also in nuclear and electrical power generation and distribution of water and gas. GW-NDT are able to inspect long sections of pipeline from a single application point to detect defects present even under insulation and to reach pipe segments placed in inaccessible areas like river crossings, bridges etc. They are reliable and become even more cost effective if the sensors are installed permanently to perform continuous monitoring of the pipe integrity, while in-service. Instruments that generate and receive GW are based on piezoelectric or magnetostrictive transducers which are disposed as a collar completely surrounding the pipe circumference.
Piezoelectric systems (PZT) are configured as a multi-channel collar able to generate and receive GW signals in multiple points around the circumference. This configuration allows also techniques like focusing that help in the process of defect characterization. The focusing technique is however, limited in the cases of small OD pipes or at long distances sensor-defect. Another method to estimate the defect size is based on information provided by the amplitude of the reflected waves or the amplitude of the wave modes other than the transmitted one. This has its limits related to the number of elements of the collar surrounding the circumference and to the pipe geometry that in some cases can be complicated.
Magnetostrictive guided-wave systems can be described as a single continuous transducing element attached to the pipe and surrounding the circumference independently of the pipe size. With this configuration the generated wave mode is always symmetrical (torsional) and no asymmetrical modes are generated. In addition, less wave control can be achieved with a single channel transmitter. In the acquisition process, asymmetrical modes generated by defects can be observed with certain difficulty because of the use of a single receiver.
The detection of a possible flaw is an important achievement, but the inability to tell how serious the situation of the analysed structure is, leaves space to further research and investigation. Some defects that occur in a pipe structure can be negligible while others, on contrary, must enforce the immediate replacement of the flawed pipe for the safety of the pipeline system. Moreover, even if one succeeds to find out the percentage of the defect’s cross-section from the total cross-section of the pipe, he cannot completely asses the gravity of the situation.
There is extensive research work documented in the NDT literature that describes algorithms for defect characterization. They are mainly based on classification techniques like neural networks, k-nearest neighbour, space mapping or support vector machine (SVM) and signal processing methods for feature extraction like wavelet and Fourier transform or rely on the extraction of features like correlation coefficients or wavelet or Fourier coefficients. The implementation of these algorithms is, however, related in most of the cases to local NDT techniques, while results of defect characterization with guided-wave NDT use amplitude or phase information or focusing techniques and are limited to large approximations of the cross-sectional loss or axial length of the defect. Guided-wave NDT remains therefore a method of screening which is reliable only for the good localization of defects along the pipe axial dimension.
Estimation of the cross-sectional loss gives limited information on the gravity of the defect, while knowing the radial depth of a defect or the remaining wall thickness would allow the plant operator to understand how imminent a fluid leak is and to decide immediately on further actions to ensure the safety of operation.
As previously described, existing guided wave technology, provide low-level information on the geometrical characteristics of defects that were detected during inspection or monitoring of pipes. The existence of a method or software able to provide this information would bring an important improvement to this long-range NDT technology.
There is therefore the need for a new method for defect characterisation in guided-wave NDT that would be able to give a fair approximation for all the three dimensions of a defect.
This thesis describes the development of new methods and algorithms for 3D characterization of defects identified from signals that were acquired with a guided-wave NDT instrument.
The methods are based on the evaluation of flexural components extracted from data sets acquired with a multitude of magnetostrictive sensors uniformly distributed around the pipe circumference. Moreover, the algorithms presented take advantage on classification techniques like Support Vector Machines (SVM) that are trained using training parameters extracted from signal information like Sum of Euclidean Distances (SED) or Flexural-Torsional Ratio (FTR).
To prove its efficiency, the method was applied to both simulated and experimental data and the results were compared. Finally a statistical study of the errors is presented and possible further developments are discussed.
File