Study On The Full Weighted Incremental Svm For Ultrsonic Detection Quantitative Recognition On Bonding Flaw Of Composite Material

With the advance of science and technology, adhesive structure is widely applied in aviation, petrochemical and other defense and civil fields, but in the process of their manfacture and use often appear bonding defects, such as bad adhesion, porosity, partial debonding phenomenon. The presence of bonding defects increase the instability factors of device, there are significant risks in the safe use of materials and products, if you don’t find accurately determine the harmfulness of the defects, may lead to a huge and even irreparable economic loss. Therefore, effective detection and identification of the bonding defects has become an extremely important research area and cutting-edge topics.Bonding defect recognition is an important research content of the non-destructive testing, has gradually become a hot research topic in the field of pattern recognition, the research and development direction can be roughly grouped into the following two aspects:First make improvements in selecting the classifier; The second is to make improvements in the method of solving multi-classification problem. In this paper, use the full weighted incremental Support Vector Machine (SVM) as classifier, to sheet composite bonding defects10standard debonding specimens, do classification based on extract effective features, and experimental work are summarized as follows:First, it introduces the mathematical principles of SVM, analyzing the nature of the support vector and incremental learning process, giving an incremental learning algorithm based on a generalized KKT conditions. The algorithm uses a generallized KKT conditions to select the training samples, in the work to preserve the useful information to reduce the training sample size, the algorithm can effectively improve the accuracy and speed of classification.Second, research the application of multi-classification, analyze and compare the pros and cons of different multi-classification method, propose combining the priori knowledge that is the center distance of the bonding defects adjacent grade classes is roughly linear with tree classification methods as the structure basis of this system, build a ten degree sheet bonding defect classification and recognition model.Third, in training the model of ten bonding defects, each incremental SVM has uneven number of positive and negative class samples lead to classification error rates tend to the small samples, at the same time the presence of noise and other uncertainties lead to some of the bonding defect sample seriously deviation from the category, and then three extracted characteristics has different classified contribution to each categories, taking into account the impact of these three aspects, therefore, this article uses the fully weighted incremental SVM of the classes, samples and characteristics weighted method effectively solve the above deficiencies.Fourth, under the same conditions, experimental results show that the fully weighted SVM has more advantages than the regular SVM, incremental SVM and weighted SVM, this method provides a more precise criterion for quantitative recognition of de-bonding in composite plate, and strengthens the foundation of automated detection for de-bonding in composite plate.