A Latent Variable Mining strategy for unanticipated state diagnosis in a telescope drive system has been proposed in this study. Compared with the existing analysis methods of explicit variables commonly used in fault diagnosis, the latent variable mining method proposed in this paper has the advantage of mining the deep hidden information of the system. First, a strategy for extracting latent variables with Stable Kernels Representation as the detection statistics is established in this paper. Then, principal component regression is used to construct an unanticipated fault diagnose strategy. Finally, an experimental platform for unanticipated load variation of the telescope drive system is established in the article, and the method proposed in this article is used to perform diagnostic tests on it. The test results show that for the unanticipated faults that cannot be correctly identified by the explicit variable analysis, the latent variable analysis method can clearly distinguish them to achieve the purpose of unanticipated fault diagnosis. All of the above show a good performance of the proposed unanticipated diagnosis strategy based on Latent Variable Mining with telescope drive system.