In high-mix low-volume textile production, visual inspection remains essential, and inspection machines continuously acquire process data that can be used as industrial sensing information for production management. We present a sensor-data-driven method to predict inspection speed and completion time in a fabric inspection process. The variability in inspection time complicates the coordination between inspectors and setup workers, leading to process inefficiencies. The aim of this study is to predict the inspection time per meter (IPM) to facilitate efficient process management. We analyzed 2098 inspection logs collected from a dyeing and inspection factory. Using random forest regression, we modeled the relationship between IPM and factors such as fabric length, defect count, and inspector characteristics. While the global model showed limited accuracy (R2 = 0.191), a stratified analysis based on prediction error revealed that the top 75% of “normal” cases were predicted with high accuracy (R2 = 0.807, mean squared error= 1.062). Conversely, the remaining 25% represented “exception” states (R2 = −0.013) governed by unrecorded delay factors. These results suggest that while IPM prediction is highly effective for normal operations, identifying exception states requires additional data on irregular events.