A novel defect detection algorithm based on intrinsic rotational symmetry was proposed to address the issue of low accuracy in defect detection for deformable rotational industrial devices. The initial model, reconstructed using 3D scanning technology, was used as input. Through an optimization iteration process constrained by intrinsic rotational symmetry and fidelity, the initial model was gradually optimized to approximate an ideal, defect-free standard model. Finally, the defect detection results were obtained by comparing the standard and input models. During the optimization process, intrinsic rotational symmetry was employed to constrain the model, enabling compatibility with deformable rotational industrial devices. Fidelity constraints were introduced to prevent excessive deformation during optimization, allowing for reasonable correction of defective areas while avoiding masking of real features caused by over-smoothing or other non-physical deformations. Constraint terms were constructed by utilizing the coplanarity property of triangle normal vectors with the rotation axis and centroid projection direction vectors, and the solution was obtained using the L-BFGS optimizer. Compared with traditional methods based on extrinsic rotational properties, intrinsic rotational symmetry can handle locally deformable features and avoid false detections. Experiments demonstrated that the proposed algorithm achieved high accuracy in high-precision defect detection for rotational devices, especially for deformable rotational industrial devices.