The dual-redundant acceleration sensor of the maglev train is externally mounted on the electromagnet. It is affected by vibration, electromagnetic interference, temperature and humidity change during operation. Its measurement characteristics are unstable, which are reflected by dynamic measurement noise. At present, the comparison test method is often used to detect redundant sensors in engineering practice. The problem of false alarm and miss alarm is easy to occur under dynamic noise, and the detection accuracy is low. Thus, in this article, an adaptive multi point generalized likelihood ratio test algorithm is proposed for sensor fault detection. The multi point decision form is used to enhance the robustness to outliers. In addition, the sliding window variance estimation is introduced to recursively estimate the variance of parity vectors, and the decision function is adjusted to realize the adaptiveness to dynamic noise. The effectiveness of the proposed algorithm is evaluated by experiments on a small-scale suspension test-rig. Compared with similar traditional algorithms, the detection accuracy of the proposed algorithm increased by 15% in static noise experiments, and increased by 13% in dynamic noise experiments. The false alarm rate and missing alarm rate are significantly reduced with good robustness to dynamic noise.