Data fusion in automated robotic inspection systems

Teams of small modular inspection vehicles for automated inspection tasks offer the possibility of employing a variety of different NDE inspection methods simultaneously. By synergistically utilising information derived from multiple sources, individual deficiencies and limitations can be partially compensated, leading to a more accurate and precise evaluation of the condition of the engineering structure under test. This paper presents approaches based on fusion of NDE data that have been obtained by a heterogeneous team of small inspection robots which are equipped with payloads for magnetic flux leakage, eddy current and ultrasonic inspection. Any potential uncertainties in individual measurements regarding the location of defects constitute the basis for fusion methods based on statistical and probabilistic algorithms. Images of a two-dimensional test structure have been constructed from data derived from different scans, indicating the positions of detected artificial defects. Applying the Dempster-Shqfer theory of evidence and Bayesian analysis, the confidence level in the accuracy of these images is increased and the uncertainty reduced