Thesis

Classification of defects for non-destructive inspection using contact sensors and data analysis

Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2021
Thesis identifier
  • T15859
Person Identifier (Local)
  • 201856954
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • The requirement for Non Destructive Testing of composite structures is paramount in Aerospace to maintain structural integrity. There are several established Non Destructive Testing and Inspection methods available dependant on the individual requirements of the structure and situation. Most established non contact methods require an uninterrupted line of sight to the structure to produce an accurate report of the underlying condition. Methods using contact sensors, such as ultrasound, require some form of surface preparation or a couplant to provide accurate readings. Within the Maintenance Repair and Overhaul (MRO) environment, such requirements results in the removal of a component from the maintenance and repair processes, as well as surface preparation to conduct an inspection. This additional time results in a detrimental impact in regards to both financial costs and the time taken to conduct the full maintenance and repair processes. This thesis will focus on the development and application of a tooling system utilising contact temperature sensors which is compatible with the MRO environment, requires minimal surface preparation and is unaffected by line of sight issues created due to the repair process (such as bagging films and breather materials). Once thermal data has been captured, data analysis should be conducted in order to indicate areas of interest (e.g. delaminations and disbonds) as well as to infer useful material properties (i.e. fibre orientation).;The indications will be presented in such a manner visually, so as to require minimal training for operators to adopt. Representative composite structures are thermally profiled within simulations and experimentally, to produce library data of expected ther- mal responses as well as for use in training Machine Learning algorithms for classification of defects. The methods developed are then tested using data previously unseen by trained algorithms or within the library data, to score their performance. The thesis presents a two part tool capable of heating a composite sample and record the resultant thermal response, without impact from the vacuum bagging process or a requirement for special surface preparation. It is shown to successfully present impact damage within composite sandwich structures in an areal form familiar to existing inspectors. The tooling in conjunction with an experimentally captured Mean Temperature profile library successfully indicates fibre orientation of biaxial 5 harness satin weave carbon fibre reinforced polymer laminates common within aerospace. This method is performs with an overall accuracy of 87-93% on samples with artificially introduced Gaussian noise. The analysis of transient thermal conductivity profile within a sample is demonstrated to successfully indicate delaminations of maximum acceptable tolerances within composite structures. This method utilises machine learning algorithms in the form of Support Vector Machine (SVM), and Random Forest (RF), achieving overall accuracy of 90% with SVM. Existing methods of tap testing produce accuracy between 73-81%. The main contributions of this Thesis can be summarised as: the design and creation of a versatile contact-based thermography tool that can be used for a variety of NDI tasks; the development of a contact-based thermography technique that utilises contact temperature sensors to assess impact damage; the creation and validation of a mean temperature response library capable of identifying the fibre orientation within a composite laminate panel based on its thermal response; the development of a contact temperature sensor based thermography method to indicate delaminations within a composite laminate utilising step heating and machine learning
Advisor / supervisor
  • Tachtatzis, Christos
Resource Type
DOI
Date Created
  • 2021
Former identifier
  • 9912982090702996

Relations

Contenu