Heuristic evaluation for progressive additive manufacturing of industrial bending tubes based on reconfigurable transfer learning

This paper introduced a heuristic evaluation method for progressive additive manufacturing (AM) of industrial bending tubes by integrating transcendental simulations based on reconfigurable transfer learning (RTL). Bending tubes occupy a significant proportion of industrial fields like automotive, aerospace, etc. The AM as a promising technology satisfies the increasingly diverse requirements in geometry and materials of bending tubes. The manufacturing evaluation of AM is an essential process to guarantee the quality of fabricated parts. The finite element method (FEM) as an effective simulation approach was applied to generate transient thermal–structure responses of AM process. To avoid the burdened and time-consuming simulation of FEM under various manufacturing parameters, the feed-forward multi-layer perceptron (MLP) was constructed to rapidly obtain temperature and deformation distributions of manufactured industrial parts. The internal parameters of MLP could be reconfigured and the effective predicting domain was expanded via RTL by retraining the neural networks with additional instance. The transcendental knowledge was configured and integrated as adaptive thresholds to heuristically evaluate the anomalous degrees of various regions. The effectiveness of MLP was verified through extensive experiments by comparing the predictions with FEM outcomes. The improvements of RTL in prediction accuracy were quantitatively evaluated. The predicting root mean square error of MLP was reduced by around 60% and the decision ecoefficiency improved by over 50% via RTL. The physical experiments were carried out to verify the RTL via selective laser sintering (SLS). The manufacturing quality of the outer and inner surfaces of the manufactured bending tube was observed with optical micrographs.