Cyber Physical System for Data-Driven Modeling of Fused Filament Fabrication (FFF) Extrusion Process

Fused Filament Fabrication (FFF) is one of the most common additive manufacturing tools. With the aid of the open-source community, constant improvements in the mechanical build, electronic components, thermal design, and software algorithms allowed for increased reliability and part quality. However, issues persist in extrusion control, manifesting as deposition errors that degrade the visual and mechanical properties of the printed part. Current feedforward control techniques use a simplified model of the extrusion process as a first order system. However, such assumption fails to capture nonlinear extrusion dynamics. Therefore, to build a controller that can improve quality under varied extrusion velocities, a more descriptive model is required. In this work, we introduce a cyber-physical system with industrial grade motion stages and controllers that can record feedback information from the drive motors of the 3D printer such as position, velocity, acceleration, and motor current. Additionally, a temperature sensor near the nozzle exit records the surface temperature of the nozzle. The printed geometry is scanned using a laser line profilometer. Using this system, single bead lines were printed with sinusoidally varying bead width. To create a model for extrusion dynamics, a nonlinear auto regressive exogenous (NARX) neural net is employed to predict the bead area given the extrusion velocity, tangent velocity, temperature, and motor current. The resulting trained NARX net provided predictions with low mean absolute error values between 0.0923 and 0.0994 using testing sets that are independent from training sets.