Automatic analysis and classification of the AIRIX single shot accelerator defaults

The AIRIX facility is a high current linear accelerator (2-3.5 kA) used for flash X-ray radiography at the CEA of Moronvilliers (France). The general background of this study is the diagnosis and the predictive maintenance of the AIRIX components. We are interested to the performances of the HV generators, which furnish the energy to accelerate the beam. The single shot functioning imposes to obtain the best performances at a given time. So we study and develop a prototype of monitoring tool using neural network and pattern recognition. Statistical models are used to define an error vector, representative of the state of the generators, which must be classified. To reduce the redundancy of this information and the computation time, we study two algorithms, the principal component analysis and the curvilinear components analysis. A classifier has been defined, with a three layer Radial Basis Function (RBF) neural network. We initialize the network by applying an unsupervised fuzzy technique to a training base. The configuration of the whole net is realized by a supervised training. Continuous learning must be enable to take in account new states, and to monitor the experiments to predict future failures. We will present the recent results obtained on the installation