Combination of Information from Several X- Ray Images for Improving Defect Detection Performances – Application to Castings Inspection

The results presented were obtained in the frame of the European QUME project, during which three Non Destructive Techniques were developed to improve inspection of castings : multi-angle radioscopy, X-ray spectrometry and vibration analysis. We present the concept of confidence level (the so-called mass function of the Evidence theory) assigned to the information delivered by each method. A simple model is introduced, based on the deviation of the measurements from the mean of a normal measurement, with respect to the standard deviation. Information is combined using the Dempster rule of combination. The final decision on the sample acceptability can be done in an objective way by a threshold in the confidence level. Results obtained on a set of 171 castings samples, containing 62 defects, show that the combination of images acquired at different angles allow both to detect more defects than a single projection approach, but also that the defects detected present a higher confidence level, thanks to the data fusion step. X-ray spectrometry was not so sensitive as expected, and thus, fusion was done in a cautious way, considering spectrometric results only when fused with radioscopy. Vibration analysis results were not reliable enough to allow a successful data fusion step. However, the potential of the different modalities will be discussed.