The FBG sensor technologies used for real-time monitoring of the mechanical vibration have attracted more and more concerns from both academic and industrial domains. In practical applications, the vibration frequency of the equipments generally reach up to a few hundred to several thousand Hz, however, the existing FBG demodulator is incompetent to meet the actual requirements of vibration monitoring in terms of demodulation speed and algorithm structures. To solve this problem while achieving a higher accuracy at a relatively lower computation cost, a novel adaptive semi-peak-seek algorithm based on state machine is proposed, which can be divided into three parts including data collection, data analysis and data verification. The experiment result shows that for the demodulation speed of 2 KHz, we can reach a demodulation accuracy of 1pm and a static noise within ±2pm. On the basis of a long term test, the stability error that less than 2pm and the system dynamic range of 0~ -30db are concluded, and the measurement error with the optical power attenuation is limited to 4pm. Thus, the proposed algorithm is capable to meet the requirements of high-speed demodulation devices in terms of speed, precision, stability and anti-interference properties.