Experimental Investigations into the Accuracy and Precision of Material Extrusion Additive Manufacturing

Abstract

Material Extrusion Additive Manufacturing (ME AM) is one of the most popular AM technologies thanks to its ease of use and ability to produce complex shapes from a range of materials at a comparatively low cost. However, it is commonly characterised by poor quality, including unsatisfactory surface roughness, dimensional and geometrical accuracy, and dimensional precision. As a relatively recent technology, it is common for users have limited familiarity with design implications and process capability. Consequently, many of the purported benefits are negated as components do not conform to specification and trial-and-error approaches are often employed instead. Given these limitations, improvements to the process would help increase the proliferation of the technology and maximise potential benefits. The existing capability of the ME AM process was first established in this thesis through experimental work involving the production of identical components across multiple prints and machines of differing designs and price points. This highlighted that all axes on all machines exhibited average dimensional error of up to 0.25mm and error variability of approximately ±0.20mm. The Z axis, typically driven by a lead screw rather than belt drive, exhibited lower error and variability. Geometrical inaccuracy was also demonstrated to be prominent, largely due to swelling outwards at corners. Using a standard process capability value of 1.33, these errors suggest a tolerance of 1.2mm is needed such that any component feature aligned with any machine-axis can be assembled with another matching feature. However, with suitable adjustments for average error and by addressing the geometrical inaccuracy, this tolerance may be reduced substantially. A high-level characterisation was then presented to categorise error sources as arising either from machine positional error or from the extrusion and deposition of material itself. Positional error was determined in terms of fixed backlash and variable scale components. This showed that backlash errors were typically of the order of 0.05mm, and scale errors were undersized by up to 0.6% representing one of the largest error sources. In addition, filament diameter and extrusion lengths were measured and found to contribute a very limited amount to overall error. Extrusion errors were characterised in terms of steady-state filament morphology and the XY plane corner errors identified in the initial performance experimentation. A new cross-sectional morphology model was presented which accounted for volumetric flow rate and the deposited material aspect ratio. This represented the outer boundary of vertical surfaces as a series of arc lengths, which predicted a deposition approximately 0.035mm wider than the standard rectangular model included in current popular slicing software. XY plane corner errors were investigated for 11 separate angles, alternative perimeter deposition orders and the inclusion or exclusion of a perimeter weld. This showed typical geometric deviations of approximately 0.1mm relative to the straight depositions before and after the corner. Potential methods of process improvement were discussed, including existing studies which have sought either to avoid the errors, compensate for them or remove them via post-processing techniques. As a result, a new machine and nozzle design were proposed. The novel nozzle incorporated a side piece, which could guide over-extruded material at corners and improve the surface finish of vertical walls. In order to operate in the XY plane, this was combined with a 4-axis machine to allow the rotation of the nozzle. Steady-state experimentation showed that a doubled layer height to 0.4mm was required for controlled morphology with the new nozzle design. For an optimal combination of temperature, extrusion rate and print speed, a comparable surface roughness and improved dimensional error, build time and bonding width were observed compared to the lower layer height standard nozzle. However, application to corners in the XY plane revealed several limitations. First, pauses at the corners associated with the nozzle rotation introduced over-extrusion as residual pressure causes the continued extrusion of molten material. Whilst the side piece was demonstrated to interact with this additional material, some nonetheless flowed beyond it resulting in poor geometrical accuracy. Reducing the effects of the pause through faster rotation speeds, lower temperatures and filament retractions did improve this behaviour, though introduced additional geometrical error inwards from the corner. Ultimately, it was determined that immediate improvements can be made through adoption of the updated filament model and better positional control of machines, to include scale calibration and limitation of backlash through proper tensioning of belts and gantry arrangements. Whilst the novel nozzle design shows promise, particularly during steady state deposition, the root cause of over-extrusion at corners still requires addressing in popular slicing software. If these issues can be overcome, improved average dimensional error and tolerances of 0.05mm and ±0.10mm respectively are achievable.