Image Based Processing for Weld Defect Detection

There is a growing need for automation in the welding industry due to a growing shortage in skilled welders. TIG [Tungsten Inert Gas] welding, a method of welding that uses an electrode shielded by gas and is fed externally by a wire, is incredibly advantageous for its precise heat control. TIG welding is considered the standard for nuclear application which requires highly precise welds to be performed. Robotic welding can address this issue, and one major problem that occurs during welding is welding defects. Typical weld defect detection requires a highly knowledgeable welder or destructive and nondestructive evaluation. Destructive evaluation such as sanding or cutting destroys part of the weld to investigate if a defect has occurred. Nondestructive evaluation such as x-rays ad ultrasonic imaging work well and do not require any damage to be done on the weld. However, these traditional measures must be done after a weld finishes due to the amount of heat that is dissipated from the system. The challenges associated with defect detection are that they are time consuming and often expensive to repair. If a defect is detected, then a repair must be done to ensure the structural integrity of the weld. The only way to repair a weld is by removing material until the defect is reached and the weld is redone. The sooner a defect is detected then the easier the rework will be. Therefore, robotic real time detection would be the gold standard for detect weld defects. Image processing is a solution to real time weld defect detection. This is especially useful so that an operator does not have to constantly monitor the weld camera to check for a defect and how much a defect is occurring. There are three types of defects that are commonly associated with TIG welding and will be investigated. Surface porosity which is described by its hole patterns in the surface of the solidification of the weld. Tie-in described by its nonlinear weld characteristics and jagged edges. Void formation which physically looks like a cavern underneath the weld. Currently an automated adaptive welding robot is in development by the University of Tennessee Knoxville in partnership with the Electric Power Research Institute. The goal of this research will be to incorporate the weld defect detection algorithm into the robot once it becomes operational. Therefore, image processing algorithms will be developed in LabVIEW to detect surface porosity, tie-in, and if possible void formation.