In light of the lack of prior information and the difficulties in characterizing when detecting novelty faults of the driving systems of astronomical telescopes， the method based on the SVM observer is proposed. First， the principles of the proposed method are introduced and a diagnostic model of the general state observer is established in terms of the driving systems of astronomical telescopes. Then， the telescope's operation data are used as samples to test the proposed method. It takes 0.047 s to detect novelty faults and the accuracy is 94%. Meanwhile， the back propagation （BP） observer needs 7.628 s and finishes with a accuracy of 85.5%， and the radial basis function （RBF） observer only reaches a accuracy of 58% within 1.985 s. Compared with them， the method based the support vector machine （SVM） observer has obvious advantages. Finally， the proposed method is implemented on a fault diagnosis and self-healing semi-physical simulation platform， which proves its feasibility in engineering applications.