Monitoring the generation and expansion of fatigue cracks in mechanical structures is critical to structural safety. To solve this problem, an optical sensing method for identifying crack propagation in mechanical structures is proposed. On-line monitoring of crack location, length, and expansion direction during crack propagation is achieved by combining micro-cavity array (MCA) fiber and optical frequency domain reflection (OFDR) system. Two adjacent ultra-short FBGs are used as a micro-cavity (MC) sensing element to obtain the strain distribution near the crack tip through a high spatial resolution distributed strain detection system. The crack state is obtained by combining the classical theoretical model, and a near real-time detection is achieved. Thereby, the system can perform an online monitoring and timely alarms on cracks. In this paper, we show the monitoring of the crack state during the process of preset crack length of 20 mm and crack propagation to 50 mm. An MCA fiber with 2542 MC elements with a spatial resolution of 1 mm is densely laid perpendicular to the crack tip direction. The crack propagation process is realized by using fatigue machine to apply cyclic load on aluminum alloy specimen, the distribution of non-uniform strain field of aluminum alloy specimen is obtained by detecting the wavelength drift of each MC element in the MCA fiber. In the test result, the distribution of the non-uniform strain field of the aluminum alloy specimen measured by the MC element is consistent with the simulation results. Consistently, the location of the crack tip and the detection of the crack length can be realized according to the distribution of the non-uniform strain field, and the feasibility of the aluminum alloy crack extension recognition system based on the MCA fiber is verified.