Loading...

Quality Estimation of Resistance Spot Welding Using Ultrasonic Testing and Artificial Neural Network Approach

Ghafarallahi, Ehsan | 2020

412 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53184 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Farrahi, Gholamhossein
  7. Abstract:
  8. One of the most common nondestructive tests is ultrasonic testing which has been paid great attention from specialists of this field. Apart from being economical and efficient, ultrasonic waves are sensitive to small changes in the structure and thus have a high degree of reliability. The most common method of ultrasonic testing is manual single-element A-scan inspection, carried out offline using longitudinal waves with pulse echo technique which is used in this thesis. The purpose of this thesis is to monitor structural health of thin metal joints and estimate quality of resistance spot welds by simulating ultrasonic testing using a finite element software. Initially, acoustic properties of the steel are calculated using elasticity formulas. After examining concepts of transmission and reflection of ultrasonic waves from interface of two environments, theoretical equation of ultrasonic signal and the formulas to calculate diameter from signal in three-sheet joints were extracted. Then, spot welds, sizing from zero to 4 millimeter in diameter (by 0.2 mm steps) were modelled using COMSOL software and 441 signals were obtained. Thereafter, using theoretical formulas, spot weld diameters of numerical models were calculated and showed an average of 14% error. Also, comparing peak amplitude of theoretical and numerical models resulted in an average error of 21% which was mainly caused due to dispersion effect. In the following, sources of error in the experimental model were studied and attenuation was introduced as the most important factor. Then, attenuation was applied to the numerical data using a trial and error method and results were given to an MLP neural network in which inputs were amplitude peaks of the signals and outputs were the diameters of spot welds. Results of the trained network had an average error of 5% which showed the capability of artificial neural network to predict spot weld diameter. At last, an instruction was introduced for rejection or passing spot welds; then, performance of the instruction applied on experimental samples was compared with the results obtained from destructive tests which indicted over 82% accuracy
  9. Keywords:
  10. Artificial Neural Network ; Ultrasonic Test ; Nondestructive Test ; Spot Weld ; Electric Resistance Welding ; Numerical Finite Element Simulation ; Resistance Spot Welding

 Digital Object List

 Bookmark

No TOC