<p>This research explored the feasibility of utilizing servo signals from a tube bender machine to estimate the wrinkle states of bent tubes. Five experimental cases were designed to simulate varying tube wrinkle qualities, including normal conditions, slight and severe mandrel position mistuning, and slight and severe mandrel wear. Features were extracted from the servo signals of the feeding (Y), bending (C), mandrel (V2), and assist boost (V1) axes during the rotary draw bending process. The wrinkle states were estimated with high accuracy using the random forests algorithm, achieving classification accuracy between 77 and 89%. Analysis revealed that torque and velocity data from the C-axis and V2-axis are strongly correlated with tube wrinkle states. Incorporating Fisher scores for feature selection further enhanced the model’s accuracy by prioritizing the most relevant signal features. The novelty of these findings provides the tube forming industry with an efficient, indirect method for quality estimation and strategies for improving production efficiency.</p>