부산대학교 도서관

Data-driven predictive maintenance of modern machinery has the potential to increase equipment lifespan and decrease manufacturing costs. Vibration analysis can effectively diagnose potential problems in machines. A Doppler radar can be used as a sensor that provides noncontact, inexpensive real-time machine vibration data collection without necessitating line-down time. Implementation with software-defined radio (SDR) allow easy adjustment of excitation signals based on application needs. Conventional fast Fourier transformation-based vibration analysis requires large amounts of data to achieve high spectral resolution needed for fault detection, which can be inefficient and computationally too expensive. In this work, we propose to use a sweeping lock-in amplifier (LIA) to achieve high frequency resolution with small amounts of data by processing windowed sections of Doppler-shifted radio signals. This algorithm can reliably measure the Doppler shift frequency corresponding to the traveling speed of a moving object and identify the frequency of oscillating target with small amplitude, with the latter widely present in machine vibration. The distinguishing condition of the two cases is mathematically derived. The proposed algorithm is studied in simulation with triangular displacement waveform for simplicity of analysis and sinusoidal waveform for generic applications. For experimental verification, speaker vibration at a known frequency is analyzed to achieve an accuracy of 0.025 Hz within the known vibration frequency. This method is robust to the presence of noise frequencies and capable of detecting multiple frequencies.