Automatic flaw detection using recognition by synthesis: practical results

In quality control nondestructive techniques gain more and more importance. Optical methods as for instance holographic interferometry have the advantage of being sensitive and can be used contactless for inspection of technical components. The acquired interferogram contains fringe patterns, that hold information about the surface deformation subjected to the applied load. The detection of faulty parts is, usually done by an expert who is used to interpret the interferogram and to decide the criticality of the detected flaws. The automation of this procedure raises several problems, since the diversity of pattern produced by different objects and flaws makes an effective processing very complicated. This paper describes a method for the recognition of fault indicating patterns by synthesis of the interferograms, the comparison with the real pattern and modification of the simulation strategy with respect to the classification of the flaw. Taking into account the experimental conditions and a first hypothesis about the type of flaw within the object, a synthesized image of the fringes can be generated and compared to the experimental image. In case the synthesized and experimental patterns differ, this indicates that the assumed hypothesis wasn't correct and have to be modified. This process is repeated until both patterns correspond, and so the supposed hypothesis about the flaw was verified. This new approach was proposed earlier for holographic interferometry and a simple object. Here we demonstrate new results at a copy of satellite tank using holographic interferometry and show the advantage with respect to robustness of detection and the classification of the voids.