<p>The surface roughness of CNC-machined frozen sand molds strongly influences the resultant surface quality of the final casting. In this paper, the surface roughness detection and prediction method of CNC machining of frozen sand mold is studied, and a surface roughness detection method based on surface-structured light is proposed to achieve the low-temperature rapid detection of CNC machining of frozen sand mold with complex structure. Planar array structured light is adopted to detect the roughness of frozen sand molds made from different materials under multiple process parameters, and the significant impact of these parameters is determined through extreme difference analysis. This analysis serves as the basis for process judgment in the online regulation of surface roughness during CNC machining of frozen sand molds. In the study, the angular coefficient is introduced as a parameter to characterize the different material sand molds, combined with the CNC machining process parameters of spindle speed, cutting speed, cutting depth, and cutting width as inputs, and the surface roughness of the frozen sand pattern as outputs, to build a BP neural network prediction model with five inputs and one output, and to realize the prediction of the surface roughness of the different material sand molds under the multi-processing process with prediction error of 2.75% or less, to provide a judgment standard for the prediction of the performance of the frozen sand pattern before machining, and to further promote the engineering application of the green casting technology. The prediction error is within 2.75%, which provides a judgment standard for predicting the performance of frozen sand molds before machining and further promotes the engineering application of green casting forming technology of digital frozen sand mold.</p>