Abstract: To address the problems of insufficient real-time acquisition accuracy of vibration signals and the difficulty of extracting fault features under complex operating conditions in railway locomotive braking system air compressors, a vibration signal acquisition and analysis system was designed. The system employs the STM32F407 microcontroller as the core controller and establishes an embedded acquisition platform composed of vibration sensors, a signal conditioning circuit, a data acquisition module, and a communication module, enabling multi-channel real-time acquisition and analysis of vibration signals. In the software design, high-speed continuous sampling was achieved through the coordinated operation of Direct Memory Access (DMA) and timers. An improved wavelet thresholding algorithm was introduced to denoise non-stationary vibration signals, while Fast Fourier Transform (FFT) and wavelet packet energy entropy were combined to extract time-frequency features for operating condition identification and fault feature analysis of the air compressor. Experimental results show that, compared with the conventional frequency-domain analysis method, the proposed algorithm effectively improves the signal-to-noise ratio of vibration signals and enhances the discriminability of fault features. Moreover, the wavelet packet energy entropy increases from 0.415 to 0.691, accurately reflecting variations in fault conditions. The results demonstrate that the proposed system provides high acquisition accuracy, excellent real-time performance, and stable operation, offering reliable technical support for condition monitoring and predictive maintenance of railway locomotive air compressors.