Dynamic characteristics of a hybrid suspension system for flexible conveyor belts integrating deep learning and multi-body transfer matrix

The hybrid suspension system based on flexible conveyor belts is a critical architecture for hybrid magnetic suspension conveyor systems. Traditional multi-body dynamics (MBD) methods face challenges such as computational complexity and difficulties in rapid prototyping and parametric studies. This study employs an extended transfer matrix method (TMM) method to discretize and analyze the system, proposing a hybrid multi-rigid-flexible structure combining mass-beam-damping spring elements. To address modeling errors and random disturbances, a Gated Recurrent Unit (GRU) neural network is integrated to correct the dynamic responses at suspension support points. Comparative simulations with long short-term memory (LSTM) networks demonstrate that GRU achieves lower mean, minimum, and maximum MSE values. An experimental platform for flexible conveyor belt hybrid suspension was developed, conducting interference response tests. Results show that under four-point suspension conditions, the TMM model maintains steady-state vibration response errors around 0.3[Formula: see text]mm and multifrequency disturbance response errors around 0.5[Formula: see text]mm. GRU-corrected steady-state predictions achieved minimum MSE values of 0.0114 and an average MSE of 0.021, while under multifrequency disturbances, MSE values reached 0.0757 (minimum) and 0.1127 (average). These findings validate the applicability and reliability of the GRU-TMM modeling approach in complex rigid-flexible systems.