Micro-milling technology is widely employed in manufacturing micro-precision parts due to its flexible processing and high accuracy. Compared to the macro-milling process, micro-milling has a smaller machining size, which leads to tool wear, tool runout, and other factors being more sensitive to the impact of machining quality. This work proposes a reliability evaluation framework for machining systems based on the micro-milling mechanical model and its surrogate method. Establish a cutting force model under shear and plow conditions with periodic variations in instantaneous uncut thickness. Subsequently, the probability distribution of the cutting parameters is obtained based on Bayesian updates. An improved buffer failure probability method is proposed to quickly get the micro-milling system failure probability. Developing new ensemble and adaptive sampling methods and introducing stochastic configuration network (SCN), an adaptive stochastic configuration network ensemble (ASCNE) model is established to alleviate the time-consuming problem of repeated calculations of mechanical models in reliability analysis. Experimental results indicate that the mechanical model can achieve good predictive performance, and the constructed ASCNE model can achieve high-precision surrogate effects. Additionally, the reliability analysis results of the system can guide tool replacement during the machining process, ensuring safe and efficient execution of the micro-milling.