Support vector machine (SVM) has excellent learning, classification ability and generalization ability, which uses structural risk minimization instead of traditional empirical risk minimization based on large sample. The perfect performance of SVM will be realized only if the parameters are rightly selected. The accuracy and efficiency of classification largely depend on the quality of the parameters selection. Focusing on the problem of the parameters selection in least squares support vector machine (LSSVM), a new method is proposed to optimize the parameters in LSSVM using adaptive genetic algorithm. The research is provided using this method on the fault diagnosis of a certain type of helicopter’s helicopter-electrical-box. Simulated results show that the proposed method achieves perfect accuracy and efficiency in fault diagnosis.