Shop-floor management is featured with dynamic and mixed-product assembly lines, where the real-time positions of manufacturing objects are critical for effective information interaction and management decision making. This paper proposes to adopt RFID technology to constantly acquire wireless signal sent from tags mounted on manufacturing objects. To build the mapping mechanism between RFID signals and object positions, online sequential extreme learning machine (OS-ELM) is applied for training and testing. Besides extremely fast learning speed and high generalization performance, OS-ELM could avoid retraining for new arrived objects and disturbance existed in dynamic shop-floor environment. The verification through experiments demonstrate that the proposed OS-ELM based RFID indoor positioning system is superior than other prevailing indoor positioning approaches in terms of accuracy, efficiency and robustness.