Experimental investigation and development of Artificial neural network modeling of 3D printed PEEK bio implants and its optimization

Abstract

Abstract Additive manufacturing (AM) initially used as validation tool and now growing rapidly with promising results and challenges. Three-dimensional (3 D) printing is sub set of AM and it develops 3D parts from digital model data by adding materials ‘layer by layer’; it found applications in automotive, aerospace and medical sectors with a competitive advantage of reducing in product development cycle. Fused deposition modeling (FDM) is of the popular 3D printing approach used to fabricate polyether ether ketone (PEEK) parts being used as biomedical implants. In this work, an experimental investigation on PEEK 3D printing, artificial neural network (ANN) modeling and parametric optimization for obtaining improved 3D prints are investigated. Four process parameters viz., infill density (ID), layer height (LH), printing speed (PS) and infill pattern (IP) that affect the surface roughness (SR) and mechanical strength (UTS) of the prints produced. An ANN model having 4-12-2 network architecture found optimum with an average prediction error of 2.98% for SR and 1.92% for UTS. The developed ANN model is compared with response surface methodology (RSM) modeling and it exhibits excellent agreement with ANN. Higher ID and lower PS is required for producing better surface quality with improved strength so as to have regular grains with less irregularities. The microstructural study is carried out by obtaining SEM images. The Multi objective optimization results using desirability analysis (DA) obtain SR value of 4.80 µm and UTS of 61.90 MPa for ID = 79.4%, LH = 0.14 mm, PS = 25 mm/s with octet pattern is the best combined quality characteristics having composite desirability of 0.8221.