Based on Data-Augmentation long short-term memory​​ gear meshing accuracy and error compensation

Conventional compensation approaches for gear error (e.g., Monte Carlo simulations) have shown efficacy in uniform meshing scenarios, however, two drawbacks limit further applications: (1) high computationally cost and low efficiency in practical implementation; (2) inadequate adaptability to complex non-uniform meshing dynamics. Addressing these issues, this work proposes a Data-Augmentation Long Short-Term Memory (DA-LSTM) model to achieve high-precision manufacturing (GB5-level, GB/T 10095 - 2008 Grade 5) of gears machined through shaping processes. The developed DA-LSTM framework combines LSTM architecture with adaptive Data-Augmentation strategies, enabling efficient prediction and compensation of machining errors under non-uniform engagement conditions. By optimizing tools trajectories and process parameters, experimental results show that the proposed method achieves GB5-level machining accuracy while achieving a 40% reduction in computational cost relative to baseline methods. Notably, the framework maintains superior prediction accuracy across varying meshing conditions, effectively bridging the research gap in adaptive error compensation for complex gear shaping operations. The presented method provides a novel technical pathway for high-precision gear manufacturing in non-ideal meshing environments.