is an open access repository that collects the work of Arts et Métiers ParisTech researchers and makes it freely available over the web where possible. Abstract. In this paper, a probabilistic multi-class pattern recognition algorithm is developed for damage detection, localization, and quantification in smart mechanical structures. As these structures can face damages of different severities located at various positions, multi-class classifiers are naturally needed in that context. Furthermore, because of the lack of available data in the damaged state and of environmental effects, the experimentally obtained damage sensitive features may differ from those learned offline by the classifier. A multiclass classifier that provides probabilities associated with each damage severity and location instead of a binary decision is thus greatly desirable in that context. To tackle this issue, we propose an original support vector machine (SVM) multi-class clustering algorithm that is based on a probabilistic decision tree (PDT) and that produces a posteriori probabilities associated with damage existence, location, and severity. Furthermore, the PDT is here built by iteratively subdividing the surface of the structure and thus takes into account the actual structure geometry. The proposed algorithm is very appealing as it combines both the computational efficiency of tree architectures and the classification accuracy of SVMs. The effectiveness of this algorithm is illustrated experimentally on a composite plate instrumented with piezoelectric elements on which damages are simulated using added masses. Damage sensitive features are computed using an active approach based on the permanent emission of non-resonant Lamb waves into the structure and on the recognition of amplitude disturbed diffraction patterns. On the basis of these damage-sensitive features, the proposed multi-class probabilistic classifier generates decisions that are in excellent agreement with the actual severities and locations of the simulated damages.

is an open access repository that collects the work of Arts et Métiers ParisTech researchers and makes it freely available over the web where possible. . At each node of the tree, a bi-class SVM along with a sigmoid function are trained to give a probabilistic classification output. For each branch, the outputs of all the nodes composing the branch are combined to lead to a complete evaluation of the probability when reaching the final leaf (representing the class associated to the branch). To illustrate the effectiveness of PDTs, they are tested on benchmark datasets and results are compared with other existing approaches.