Model-data fusion based thermal error measurement and prognostics for motorized spindles
Nowadays,thermal error prediction of machine tool spindles has not effectively linked physical degradation patterns with machine tool state data,resulting in redundant sensor layouts and insufficient interpretability and accuracy in predictive models.To tackle these challenges,this study accurately predicted the geometric error at the grinding wheel end of a forming grinding machine's motorized spindle by employing the approach of numerical modeling and data-driven decision-making fusion.First,a finite element numerical model of the grinding machine spindle was created to identify feasible regions for steady-state temperature field measurement points.Following this,a temperature measurement point layout method that combines unsupervised and supervised learning was developed using a multi-objective optimization algorithm.In addition,an autoregressive modeling theory in time-series prediction was utilized to design the multi-channel inverted Transformer algorithm.This algorithm,leveraging an encoder-decoder architecture,maps temperature signals to thermal error deformation in variable-length prediction mode,overcoming weak generalization in thermal error prediction caused by long lag steps.Ultimately,the effectiveness of this intelligent thermal error prediction method within the digital-physical linkage framework was validated through grinding experiments on the grinding machine.
NETL
NSTL
万方数据
