A new system for fully automatic inspection of digital flat-panel detector radiographs of aluminium castings
The aim of our work was the integration of various newly-developed methods into a system for fully automatic radioscopic inspection of arbitrary casting parts. Using a 16-bit flat-panel detector, projections in arbitrary directions through the part are acquired and analysed. The software package for testing can be separated into five stages: calibration, registration, filtering, selection, and quality assessment. Thereby, each step is realized with full 16-bit data processing to allow for quantitative measurement of the depth of details, i.e. their length in direction of the incoming radiation. Every algorithm used was optimized with respect to processing time such that a cycle time of less than 2°s per image was achieved. Up to now, the described system has been realized at several facilities and it's effectiveness has been proved in industrial routine. Within the first processing stage from the primary projections information about the physical length and density of the aluminium structures is extracted. Besides, various methods of correction, calibration and testing developed for high-end 3D-Computet Tomography system are utilized. These include algorithms to reduce scattered radiation effects, image afterglow and scatter inside the detector itself as well as calibration with dedicated test objects and correction for non-linear detection properties. Next, the primary image is registered with a reference image, which was acquired previously. The third stage combines both reference image-based and reference-less testing. Afterwards a filter operation is performed that adapts automatically to the local object structure by referring to the properties of the reference. Thereby, the self-adapting filter selects its size, direction and filter method optimally according to the local situation. Similar to the reference-less procedure, a subtraction is followed by a threshold operation, resulting in a map of regions that are suspected to be faulty. The fourth step aims at an elimination of false-positive detections (""pseudos""). Again, two methods are applied successively: evaluation of local image features at suspicious positions and a classification based on teachings independent of position and orientation of the faults. Within the last step the quality criteria given by the casting companies are applied which maybe fault size and depth, the density of faults in critical regions and a minimum distance between two or more faults.