Accelerated design of curved-beam-based torsional isolator with quasi-zero stiffness

Low-frequency torsional vibrations pose significant challenges in precision engineering systems, compromising stability, control accuracy, and structural integrity. To address this, a novel torsional quasi-zero stiffness (QZS) isolator based on curved beams is proposed. The design employs Non-Uniform Rational B-Splines (NURBS) to precisely define the beam’s centroidal axis, enabling the realization of QZS behavior through geometric shaping alone. Unlike traditional QZS designs relying on complex mechanical elements, the proposed structure is compact, lightweight, and easily miniaturized. To efficiently achieve target torque–stiffness responses, a deep neural network (DNN) surrogate model is integrated with an improved real-coded genetic algorithm (IRGA), forming an inverse design framework that eliminates the need for repeated finite element analyses. The resulting beam designs exhibit strong robustness and allow programmable behavior through modular assembly. Quasi-static tests validate the design methodology. Dynamic analysis using the method of averaging reveals typical nonlinear features, including frequency jumps, sub/superharmonic resonance, and chaos. Most configurations yield stable period-1 responses. Finally, theoretical and finite element evaluations show that the proposed isolator achieves a 68.6 % reduction in isolation onset frequency compared to a linear isolator of equal stiffness, confirming its superior performance in low-frequency torsional vibration isolation. • A novel NURBS beam-based torsional isolator with QZS characteristics is proposed. • A deep learning-accelerated inverse design framework enables rapid optimization. • The nonlinear torsional dynamic behaviors are captured by theoretical and numerical methods. • The onset frequency is reduced by 68.6 % compared to linear isolators.