The unfluctuating running of on-orbit spacecraft equipment has a decisive impact on the smooth implementation of space exploration mission. However, due to the adverse work conditions and complex running states, it is really a challenge for the online monitoring of aerospace equipment. In this paper, an improved growing neural gas method based on incremental learning is proposed, which is dedicated to solving the problem of online anomaly detection. The learning rate of the proposed method is adaptively adjusted according to the process of model training, ensuring the weights update quickly at the beginning of model construction and converge steady at the end of model training. The optimized insertion mechanisms of neurons ensure that the necessary new neurons are inserted at the right time and location dynamically, while the innovative deletion mechanisms of neurons ensure that the worthless neurons be deleted timely and at the same time guarantee the representation ability of model. The comparison results with the conventional methods on public datasets show that the proposed method achieves the better performance obviously, both in the aspects of detection accuracy and computational efficiency, respectively. At last, as a case study, the proposed method is used for online anomaly detection of a real aerospace device, i.e., a gamma ray detector, and the final F1 score of anomaly detection is as high as 98.78%. The results show that the proposed method can be applied to online detection of aerospace equipment health conditions effectively.