In the workshops of Nepal’s Kathmandu Valley, artisans have been casting exquisite metal sculptures for more than fourteen centuries, using techniques passed down through generations of Newari craftsmen. Now, a team of researchers has brought modern statistical optimization into this ancient art, demonstrating that careful control of three casting parameters can dramatically shrink the dimensional errors that plague traditional investment casting. The study, published in the journal Heliyon, applied a combination of the Taguchi method, grey relational analysis, and principal component analysis to gilding metal sculpture casting, achieving shrinkage reductions so large that the finishing stage of production could be shortened considerably.
Investment casting is one of metallurgy’s oldest processes, with roots stretching back to early weapons, jewelry, and religious art. It remains prized today for aerospace turbine blades and biomedical components because it delivers exceptional surface finish, dimensional accuracy, and the ability to reproduce complex shapes. In Nepal, the technique has been used since at least the sixth century A.D. to produce the bronze and gilded deities that fill temples and monasteries across the Himalayas. The process is intensely manual: a sculptor first carves a detailed wax master pattern, which is encased in a rubber mold from which multiple wax replicas can be made. These replicas are assembled with gating systems, dipped in a traditional slurry of cow dung and clay, reinforced with metal wires, dewaxed, preheated, and finally filled with molten metal. After cooling, each piece demands hours of filing, chiseling, and hand-applied gold plating.
That artisanal character comes at a cost. Defect rates in Nepali sculpture casting hover around thirty percent, and sculptures can take anywhere from one month to two years to complete depending on size. Because global economic shifts and advancing technology are squeezing demand for Nepali cast sculptures, improving casting design, modeling, and production efficiency has become essential for the industry’s survival. The research team, led by Zenisha Shrestha with Abhishek Pandey and Bijendra Prajapati, set out to determine whether systematic parameter optimization, never before applied to this traditional setting, could meaningfully improve dimensional accuracy in gilding metal, an alloy of ninety percent copper and ten percent zinc that is the most widely used sculpture material in Nepal.
The researchers chose a sword as their test specimen, selected for its cultural relevance in Nepali society and its relatively simple geometry, which makes dimensional analysis tractable. The design was modeled in SOLIDWORKS, with the gating system developed through the modulus method to reflect designs typical of sculpture manufacturing. From an Ishikawa cause-and-effect analysis of casting stability, the team identified three controllable parameters: the number of slurry coating layers, the mold preheat temperature, and the metal pouring temperature. Factors such as wax composition, alloy composition, slurry composition, cooling methods, and environmental conditions were treated as noise parameters that could not be easily controlled in the workshop.
The experimental design followed a Taguchi L9 orthogonal array, allowing nine carefully chosen experiments to explore the parameter space efficiently. Slurry coatings ranged from two to four layers, preheat temperatures spanned 500 to 600 degrees Celsius, and pouring temperatures covered the 1150 to 1200 degree Celsius range typical of Nepali sculpture foundries. Each coating choice involves a trade-off: thin shells save material and time but risk bulging and leaking under the metallostatic pressure of pouring, while thicker coatings prevent defects and improve heat retention at the cost of longer processing. Preheating the mold reduces thermal shock and premature freezing of the melt, improving fill of thin sections, though excessive preheat can accelerate mold-metal reactions and increase surface-connected porosity. Higher pouring temperatures superheat the metal above its melting point, preventing unfilled sections in intricate features.
Four response variables were measured for each casting: weight, length, breadth, and thickness. Before optimization, the team verified data quality through normality testing with probability plots and the Anderson-Darling test, confirming that all responses followed normal distributions. Pareto tests and analysis of variance at a 95 percent confidence level then established that all three process parameters significantly influenced every response variable. The number of coatings emerged as the dominant factor, contributing 44.42 percent of the variance in weight, 48.01 percent in length, 58.81 percent in breadth, and 41.82 percent in thickness. Pouring temperature exerted its greatest influence on weight at 39.31 percent and length at 31.38 percent, while preheat temperature most strongly affected breadth and thickness. Residual plots showed randomly distributed errors, confirming the reliability of the statistical model.
Because dimensional accuracy depends on optimizing all four responses simultaneously, the researchers turned to multi-response optimization. Grey relational analysis, a technique designed for systems with limited information, converts multiple responses into a single grey relational grade by normalizing the data and calculating correlation coefficients, using an identification coefficient of 0.5 consistent with prior studies. The innovation here was the coupling of grey relational analysis with principal component analysis, which uses eigenvectors to derive objective weights for each response rather than assuming they matter equally. The first principal component captured 81.9 percent of the data’s variance and identified length as the most significant response. Experiment 3, combining a 500 degree Celsius preheat, a 1200 degree Celsius pour, and four coating layers, ranked highest under both the standard and PCA-weighted grades, and both methods converged on the same optimal setting, strengthening confidence in the result.
The confirmatory experiment delivered striking improvements. The weighted grey relational grade rose from 0.357 under initial conditions to 0.966 under optimal conditions, closely matching the predicted value of 0.9677. Weight deficit fell from 3.160 percent to 0.682 percent, length shrinkage dropped from 4.860 percent to 0.545 percent, breadth shrinkage plummeted from 6.156 percent to 1.067 percent, and thickness shrinkage declined from 4.200 percent to 1.800 percent. In practical terms, a cast sword produced under the optimized parameters now deviates from its wax pattern by barely one percent in its principal dimensions, meaning far less manual filing and correction before gold plating. Given that finishing work is among the most labor-intensive stages of sculpture production, even modest dimensional improvements translate into significant reductions in lead time and cost.
The implications extend well beyond Nepal’s foundries. The authors note that the GRA-PCA methodology can be adapted to any complex manufacturing process where multiple conflicting objectives must be balanced, including precision casting of aerospace turbine blades, biomedical device manufacturing where dimensional control is critical, and automotive component casting where strength, weight, and tolerances compete. For Nepal, the study represents the first application of advanced multi-response optimization to traditional sculpture casting, and it arrives at a pivotal moment for an industry whose economic viability depends on competing with industrialized producers. Future work, the researchers suggest, could examine surface roughness, mechanical properties, and additional process parameters. But the central message is already clear: fourteen centuries of artisanal wisdom and twenty-first-century statistical rigor are not adversaries. When the number of coatings, the preheat temperature, and the pouring temperature are tuned together, the ancient art of Himalayan metal sculpture can achieve a precision its original masters could scarcely have imagined, preserving both a cultural heritage and the livelihoods of the craftsmen who sustain it.
Subject of Research: Multi-response optimization of investment casting parameters to improve the dimensional accuracy of gilding metal sculptures in traditional Nepali manufacturing.
Article Title: Improvement of dimensional accuracy in gilding metal sculpture manufacturing using grey relational analysis coupled with principal component analysis
Article References: Shrestha, Z., Pandey, A., & Prajapati, B. (2026). Improvement of dimensional accuracy in gilding metal sculpture manufacturing using grey relational analysis coupled with principal component analysis. Heliyon, 12(14), Article e45412. https://doi.org/10.1016/j.heliyon.2026.e45412
Image Credits: AI Generated
DOI: Not provided
Keywords: investment casting, grey relational analysis, principal component analysis, Taguchi method, dimensional accuracy, gilding metal, Nepal sculpture manufacturing, shrinkage reduction, ANOVA, multi-response optimization, Kathmandu Valley, metal casting
Cite Scienmag News
APA MLA Chicago
Drew Townsend. (September 13, 2026). Ancient Nepali Sculpture Casting Gets a Modern Statistical Upgrade. Scienmag. https://scienmag.com/ancient-nepali-sculpture-casting-gets-a-modern-statistical-upgrade/
Drew Townsend. “Ancient Nepali Sculpture Casting Gets a Modern Statistical Upgrade.” Scienmag, 13 September 2026, https://scienmag.com/ancient-nepali-sculpture-casting-gets-a-modern-statistical-upgrade/. Accessed 13 September 2026.
Drew Townsend. “Ancient Nepali Sculpture Casting Gets a Modern Statistical Upgrade.” Scienmag. September 13, 2026. https://scienmag.com/ancient-nepali-sculpture-casting-gets-a-modern-statistical-upgrade/
Copy citation Download RIS
Tags: ANOVAdimensional accuracydimensional accuracy in traditional metalworkgilding metalgrey relational analysisgrey relational analysis for sculpture precisionhistorical Nepalese bronze and gilded sculpturesimproving manual sculpture production processesintegration of modern analytics in ancient craftsinvestment castinginvestment casting techniques in NepalKathmandu ValleyKathmandu Valley ancient artmetal castingmodern statistical optimization in metallurgymulti-response optimizationNepal sculpture manufacturingNepalese metal sculpture castingPrincipal Component Analysisprincipal component analysis in art restorationshrinkage reductionTaguchi methodTaguchi method in metal castingtraditional Nepalese craftsmanship
