In allusion to the characteristics of complex structure of rescue lifting vehicle, the poor independence of working condition and fault data, and the difficulty in fault diagnosis, we proposed a fault diagnosis algorithm of rescue lifting vehicle with the semi-supervised support vector machine in this paper. In the algorithm, the information on hidden features of the original fault data was mined with the idea of spectral clustering, to effectively distinguish the independent structural features of the fault information in the component system with different degrees of coupling. Firstly, a fault map was established based on the original input data. Then, a kernel function more consistent with the clustering hypothesis was obtained by establishing the Laplace matrix. Finally, the semi-supervised support vector machine model was built and solved by the gradient descent algorithm to obtain the final classification result. Besides, we applied the above algorithm to the working condition fault diagnosis system of XCA30_JY rescue lifting vehicle, to classify the collected working condition data by building a real simulation environment. In order to evaluate its performance, comparison was made with the traditional support vector machine and the gradient descent semi-supervised support vector machine. The experimental results show that the proposed method can reduce the error rate of fault classification for resure cifiting vehicles to 10.2%, which can effectively solve the problems of fault diagnosis in complex working conditions. Conclusively, this algorithm could be widely used in vehicle fault diagnosis system and has a universal application and guiding significance in data classification and pattern recognition, due to its extraordinary capability of arbitrary sample spatial clustering and optimization solution of non-convex function.