New Model Predicts Optimal Rubber Polishing for Aerospace Composites
In the high-stakes world of aerospace manufacturing, even microscopic surface imperfections can compromise the structural integrity of composite components. A team of researchers from Huazhong University of Science and Technology has developed a novel predictive model that significantly improves the surface finish of rubber soft dies used in the fabrication of composite stiffened panels. Their approach, which combines grey relational analysis (GRA) with response surface methodology (RSM), delivers unprecedented accuracy in forecasting surface roughness after robotic polishing—offering a practical solution to a persistent challenge in advanced composite manufacturing.
Composite stiffened panels are widely used in modern aircraft due to their high strength-to-weight ratios and fatigue resistance. However, these panels often feature variable ply thicknesses—so-called “dropped plies”—to meet localized load requirements. This design leads to uneven skin thickness across the panel, which in turn creates inconsistent fitting gaps between the stiffeners (ribs) and the skin during assembly. To address this, engineers traditionally apply a layer of rubber soft die onto a rigid mold. By precisely polishing this rubber layer, the gaps can be eliminated, ensuring a uniform interface during co-curing or bonding.
Yet, the polishing process itself introduces new complexities. Rubber, unlike metals, is highly susceptible to surface degradation under improper machining conditions. If polishing parameters such as abrasive grit size, spindle speed, contact pressure, or edge distance are not carefully controlled, the rubber surface can become “fuzzy” or excessively rough. This elevated surface roughness not only compromises the aerodynamic and aesthetic quality of the final part but also exacerbates a more insidious problem: static charge buildup.
Natural rubber has a volume resistivity exceeding 10⁸ Ω·cm, making it prone to electrostatic accumulation during dry abrasive processes. A rougher surface provides more micro-cavities and asperities for fine dust and grinding debris to adhere to, further intensifying contamination. This dust layer can interfere with downstream metrology—such as laser scanning or structured light inspection—leading to inaccurate surface measurements and potentially flawed quality assessments. Therefore, achieving a smooth, clean rubber surface is not merely a cosmetic concern but a critical process requirement.
Despite its importance, research on rubber surface finishing—particularly through robotic abrasive polishing—has been sparse. Prior studies have largely focused on cutting or milling operations. For instance, Jin et al. explored high-speed milling of soft rubber and found that higher cutting speeds and larger rake angles yielded better surface finishes. Yan and Strenkowski used finite element simulations to show that turning rubber with a tool rake angle above 50° produced smoother surfaces. Others investigated cryogenic turning of nitrile rubber or applied Taguchi methods to analyze turning parameters. However, none addressed the specific dynamics of end-face abrasive polishing—a technique increasingly adopted in automated aerospace production lines.
Recognizing this gap, Qiangsheng Shi, Xiaojian Zhang, Wei Chen, Zeyuan Yang, and Sijie Yan set out to systematically investigate the influence of key polishing parameters on the surface roughness (Ra) of natural rubber soft dies. Their work, published in China Mechanical Engineering (Vol. 32, No. 24, December 2021), introduces a hybrid GRA-RSM framework that first identifies dominant factors and then builds a high-fidelity predictive model.
The team constructed a custom robotic polishing system centered around an ABB IRB4400 industrial robot equipped with an Active Orbital Kit (AOK)—a compliant, force-controlled end-effector capable of maintaining consistent contact pressure during surface interaction. Integrated with an industrial vacuum system, the setup actively removes grinding debris in real time, minimizing re-deposition and static-related adhesion. The rubber specimens—3 mm-thick sheets of industrial-grade natural rubber with a Shore A hardness of 65—were bonded to Q235 steel plates and clamped securely to a workbench.
Four primary input variables were selected: abrasive grit size (P80 to P240), polishing pressure (20–50 N), spindle speed (3,000–6,000 rpm), and edging distance (10–40 mm)—the latter defined as the offset from the panel’s edge along the X-axis. To account for tool orientation effects, experiments were repeated at three tilt angles: 2°, 5°, and 8°. The response variable was the average surface roughness (Ra) measured post-polishing.
The researchers began with a qualitative assessment using grey relational analysis based on an L16(4⁴) orthogonal array. This method is particularly effective when dealing with limited experimental data and multiple performance characteristics. For each experimental run, they calculated the signal-to-noise (S/N) ratio using the “smaller-the-better” criterion, since lower Ra values are desirable. The S/N ratios were then normalized and used to compute grey relational coefficients and, ultimately, comprehensive grey relational grades.
The results revealed a clear hierarchy of influence: abrasive grit size emerged as the most significant factor, with a range (R) of 0.505 in the grey relational grade across its four levels. This was followed by spindle speed (R = 0.100), polishing pressure (R = 0.086), and edging distance (R = 0.083). Notably, finer grits (e.g., P240) consistently yielded lower Ra values, though they also showed increased clogging—especially near the center of the polishing disc, where linear velocity is lowest and centrifugal ejection of debris is weakest.
Additionally, the study confirmed that smaller tilt angles produce smoother finishes. At 2°, the average Ra was consistently lower than at 5° or 8°, likely because a near-parallel orientation maximizes contact area and promotes uniform material removal. However, excessively low angles (below 2°) risk inadequate chip clearance, leading to abrasive loading and premature tool wear—a critical trade-off for process designers.
While GRA provided valuable insights into parameter ranking and optimal discrete settings within the tested matrix, it could not interpolate or predict outcomes for untested combinations. To overcome this limitation, the team employed Box-Behnken Design (BBD), a class of response surface methodology ideal for three-factor, three-level experiments with fewer runs than full factorial designs.
Focusing on a fixed tilt angle of 5° and P80 abrasive (to explore a more aggressive polishing regime), they varied spindle speed (3,000–5,000 rpm), pressure (30–50 N), and edging distance (10–30 mm) across 17 experimental runs, including five center points for error estimation. Analysis of variance (ANOVA) confirmed the model’s statistical robustness: the overall p-value was <0.0001, indicating extreme significance, while the lack-of-fit p-value (0.073) suggested the model adequately represented the true response surface.
All linear terms (spindle speed, pressure, edging distance) and quadratic terms (speed², pressure²) were highly significant (p < 0.0001). Interestingly, the interaction between spindle speed and edging distance, as well as between pressure and edging distance, also proved significant (p < 0.01), whereas the speed–pressure interaction was not—implying these two factors act largely independently on surface roughness.
The resulting second-order regression equation achieved an R² value of 0.9878, demonstrating exceptional goodness-of-fit. Validation through 15 additional test points yielded a root mean square error (RMSE) of just 0.01447 µm between predicted and observed Ra values—confirming the model’s predictive power in real-world conditions.
Using this equation, the researchers employed numerical optimization to identify the parameter combination that minimizes Ra. The optimal solution was found at: spindle speed = 4,158.9 rpm, polishing pressure = 38.4 N, and edging distance = 30 mm. Under these conditions, the predicted surface roughness reached a minimum of 3.3 µm—a substantial improvement over baseline runs that exceeded 6 µm.
This level of precision is transformative for aerospace tooling. By integrating the GRA-RSM model into robotic control software, manufacturers can dynamically adjust polishing parameters based on real-time feedback or pre-defined geometry, ensuring consistent rubber surface quality across large, complex molds. Moreover, the methodology is transferable: with minor recalibration, it could be applied to other elastomers or even soft polymers used in automotive, medical device, or consumer electronics manufacturing.
The implications extend beyond surface finish. A smoother rubber die reduces dust adhesion, which in turn enhances the reliability of automated inspection systems. It also prolongs abrasive life by mitigating clogging, lowering consumable costs. Most importantly, it ensures tighter tolerances in composite assembly—critical for structural performance in flight-critical components.
The research team emphasizes that their approach exemplifies the convergence of data-driven modeling and physical experimentation. While machine learning models are increasingly popular in manufacturing, they often function as “black boxes” with limited interpretability. In contrast, the GRA-RSM framework offers both predictive accuracy and engineering insight—revealing not just what works, but why.
Looking ahead, the group plans to expand the model to include additional variables such as abrasive wear progression, temperature effects, and multi-axis toolpath strategies. They are also exploring real-time adaptation using in-situ surface sensors, which could enable closed-loop polishing control—a key step toward fully autonomous finishing systems.
This work underscores a broader trend in advanced manufacturing: the shift from empirical trial-and-error to physics-informed, statistically validated process optimization. As industries demand higher precision, repeatability, and sustainability, such hybrid methodologies will become indispensable.
In an era where every micron counts, the ability to predict and control surface morphology with sub-micron confidence is not just an engineering achievement—it’s a competitive necessity.
Authors: Qiangsheng Shi¹,², Xiaojian Zhang¹,² (corresponding author), Wei Chen³, Zeyuan Yang¹,², Sijie Yan¹,²
Affiliations:
¹ School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074, China
² State Key Laboratory of Digital Manufacturing Equipment and Technology, Wuhan 430074, China
³ Wuxi CRRC Times Intelligent Equipment Co., Ltd., Wuxi 214174, China
Published in: China Mechanical Engineering, Vol. 32, No. 24, pp. 2967–2974, December 2021
DOI: 10.3969/j.issn.1004-132X.2021.24.009