Advanced Pattern Recognition Techniques for Fast and Reliable E-nose Response Analysis in NDTs Scenarios

摘要

Non Destructive Tests (NDT) technologies have strong requirements in terms of fast response and reliability, these constraints are particularly stressed in aerospace industry, for safety and efficiency reasons. Composite panels bonding, essential to green aircraft concept, is still lacking a validated NDT technique that can guarantee the bond quality. E-noses appear to be a promising choice but they have to be adapted to the surface analysis task, furthermore they may require a significant amount of time to acquire a complete response pattern slowing the testing of large panels. In this paper, we present the results obtained by combining real time classifiers response on an adapted e-nose platform to obtain a rapid response while keeping the possibility to maximize accuracy awaiting for the end of the measurement cycle. A reject option is casted on the base of combined perceived reliability to nullify high safety costs false negatives while keeping low the false positive rate.