Hybrid Algorithm and Abrasive Machining Cut 3D-Printed Composite Surface Roughness

A hybrid computational optimization framework combining Cuckoo Search and Differential Evolution algorithms has successfully identified parameter settings for abrasive flow machining that significantly improve the surface quality of components produced by fused deposition modelling. In an experimental investigation published in Discover Mechanical Engineering, researchers addressed one of the persistent limitations of material extrusion additive manufacturing, where layer-by-layer deposition inherently creates staircase defects, chordal deviations, and surface irregularities that degrade functional performance and aesthetic quality. By applying a specialized vertical abrasive flow machining process driven by an environmentally benign slurry, the research team demonstrated that targeted parameter tuning can reduce the initial surface roughness of copper-reinforced polylactic acid composite specimens from 8.6 micrometers down to an experimentally confirmed value of 6.05 micrometers, reflecting a 26.7 percent improvement in surface finish while preserving the underlying geometry and structural integrity of the fabricated components.

Fused deposition modelling remains one of the most widely adopted additive manufacturing methods across industrial and biomedical engineering domains because of its low operational cost, equipment simplicity, and geometric flexibility. However, this discrete deposition mechanism introduces structural and morphological anomalies, including layer-interface ridges, orientation marks, inter-layer voids, and stepped topographies along inclined or curved contours. While this combination yields a biocompatible, non-toxic, and biodegradable composite structure suitable for medical and specialized engineering applications, it presents complex post-processing challenges because traditional abrasive and mechanical finishing techniques struggle to uniformly smooth intricate internal passages and external features without damaging delicate geometries.

To overcome these finishing limitations, the investigators employed vertical abrasive flow machining as a non-traditional post-processing intervention. In abrasive flow machining, a viscoelastic carrier loaded with solid abrasive particles is repeatedly forced under controlled extrusion pressure across or through targeted workpiece surfaces. To match the sustainable and non-toxic attributes of the polylactic acid and copper composite matrix, the researchers formulated an eco-friendly abrasive formulation using paper pulp, peanut oil, vegetable oil, and stainless steel 316L grade powder passed through a 325-mesh sieve. This natural viscoelastic blend provided sufficient rheological elasticity and viscosity to transport the stainless steel particles uniformly along the component contours, facilitating consistent microchip formation and plastic shear deformation across irregular surface profiles.

The test specimens were fabricated as tubular structures using a Creality Ender-3 additive manufacturing system featuring a maximum build envelope of 0.220 by 0.220 by 0.250 cubic meters. The composite filament, characterized by a nominal diameter of 1.75 millimeters, was deposited within a layer thickness span ranging between 100 and 400 microns in accordance with ASTM standard D638 testing criteria. Following additive manufacturing, the cylindrical specimens were transferred to the specialized vertical abrasive flow machining apparatus for surface smoothing. Surface roughness assessments were systematically recorded using a precision Talysurf stylus contact profilometer.

Because the physical interaction between the abrasive suspension and the component walls is governed by non-linear physical phenomena, classical optimization techniques relying on isolated trial-and-error runs or standard response surface methodologies often fail to identify global parameter optima across multi-level manufacturing domains. In this study, the primary operational factors evaluated were the number of processing strokes, the applied extrusion pressure, and the concentration of abrasive particles within the viscoelastic slurry. Increasing stroke counts generally promotes surface refinement through cumulative material removal, yet excessive cycling risks over-polishing, unnecessary cycle times, and geometric distortion. Similarly, elevated extrusion pressure enhances normal cutting forces and abrasive penetration into surface peaks, but excessive pressure compacts the media, reduces particle mobility, and causes deep scratch grooves. The abrasive concentration directly dictates cutting point density, where excessive particle crowding produces inter-particle collision and high media viscosity that impede fluid motion.

To model and optimize these interacting parameters, the researchers applied an orthogonal Taguchi L9 experimental array to sample the parameter domain without requiring hundreds of experimental fabrications. The experimental data points gathered from the Taguchi design were initially utilized to train an artificial neural network predictive model structured around a Multi-Layer Perceptron regressor. This preliminary neural architecture evaluated fitness responses across the design domain and generated synthetic data points within the experimental space, thereby expanding the training datasets. A subsequent neural network model trained on the expanded dataset provided a continuous computational objective function that mapped the relationship between stroke count, extrusion pressure, and abrasive concentration to predict arithmetic mean surface roughness values for any candidate parameter set evaluated during metaheuristic optimization loops.

The optimization strategy was implemented by combining two distinct metaheuristic frameworks into a unified Cuckoo Search and Differential Evolution algorithm, denoted as CS-DE. The Cuckoo Search component simulates the obligate brood parasitism of cuckoo bird species and incorporates stochastic Levy flight random walks generated through Mantegna’s algorithm, establishing broad global exploration capabilities that prevent candidate solutions from becoming trapped within local minima. Conversely, the Differential Evolution algorithm contributes fine local exploitation through randomized vector mutation, binomial parameter crossover, and deterministic selection operators. In the integrated architecture, the initial candidate population is partitioned each generation according to a predetermined ratio, assigning one subset to Levy flight exploration and the alternate subset to DE/rand/1 mutation and binomial crossover operations, maintaining algorithmic diversity across a designated ceiling of iterations.

In parallel with the neural network analysis, the investigators developed a third-order polynomial regression equation in MATLAB to formulate an explicit mathematical representation linking the process variables to the resulting surface finish. Statistical evaluation demonstrated that this third-order polynomial model achieved a coefficient of determination of 0.9831, indicating that 98.31 percent of the total variation in surface roughness could be explained by the mathematical formulation. The predictive model also exhibited a mean squared error of 0.002322 and a mean absolute error of 0.03948, demonstrating high predictive reliability. The algorithmic optimization run executed through the hybrid CS-DE framework tracked progressive reductions in predicted roughness across successive iterations, dropping from an initial population best of 6.3178 micrometers down to an optimized theoretical minimum of 6.192333 micrometers under an identified parameter set of 138.1284 strokes, 0.0098 MPa extrusion pressure, and a 32.8423 weight percent abrasive concentration.

To validate the computational model under physical conditions, the researchers performed confirmation machining experiments within the predicted optimum operational window. Using machine settings of 138 cycles, an extrusion pressure of 0.009807 MPa, and an abrasive concentration of 33 weight percent, the confirmation run produced an experimental surface roughness of 6.05 micrometers. Comparing this measured roughness against the regression model prediction of 6.193624 micrometers yielded an absolute percentage error of only 2.37 percent, confirming close alignment between empirical observations and algorithmic predictions. In alternate parametric evaluations reported in the study, a combination of 100 strokes, 0.00981 MPa extrusion pressure, and 35 weight percent abrasive concentration predicted a surface roughness of 6.193624 micrometers, with physical confirmation producing 6.30 micrometers at a prediction discrepancy of approximately 1.71 percent, confirming consistent predictive accuracy across tested operating regimes.

Surface topography profiles and scanning electron microscopy examinations verified the physical mechanisms driving the measured roughness reductions. Micrographs captured prior to abrasive treatment revealed pronounced surface defects, including deep inter-bead valleys, sharp stair-step ridges, and irregular voids resulting from the additive filament deposition paths. Following abrasive flow post-processing under optimized parameters, scanning electron images showed substantial eradication of these sharp ridge crests and layer-induced asperities, yielding a continuous, smoothed, and leveled surface devoid of surface tears or micro-cracks. The smoothed moving-average topography profiles obtained from the stylus instrumentation confirmed that repeated micro-cutting and micro-ploughing by the suspended stainless steel particles successfully leveled macro-scale irregularities without compromising the underlying structural walls of the tubular components.

The study highlights specific procedural boundaries and avenues for subsequent manufacturing research. While the combination of a third-order regression model and hybrid CS-DE metaheuristics reliably navigated the parameter envelope, the investigators emphasized that comprehensive rheological and physicochemical characterization of the green abrasive slurry remains essential. Subsequent investigations must measure media viscosity, long-term slurry stability, particle dispersion degradation over extended operational life, and shear-thinning flow properties to better understand carrier transport mechanics.

Subject of Research: Surface finish optimization of additive manufacturing components using abrasive flow machining and hybrid metaheuristic algorithms

Article Title: Optimization of surface finish in fused deposition modeled components using hybrid metaheuristic techniques and abrasive flow machining

Article References: Shah, V., Sehgal, S. S., Chohan, J. S., Kalsi, S., & Hunde, B. R. (2026). Optimization of surface finish in fused deposition modeled components using hybrid metaheuristic techniques and abrasive flow machining. Discover Mechanical Engineering, 5(1), Article 174. https://doi.org/10.1007/s44245-026-00356-y

Image Credits: AI Generated

DOI: 10.1007/s44245-026-00356-y

Keywords: fused deposition modelling, abrasive flow machining, surface roughness, cuckoo search, differential evolution, hybrid metaheuristics, additive manufacturing, copper reinforced polymer, biodegradable abrasive medium, Taguchi experimental design, artificial neural network, polynomial regression, extrusion pressure, scanning electron microscopy, micro cutting mechanics, surface topography