Optimising Ester-Cured phenolic resin sand moulds using Taguchi–TLBO approach to improve mechanical properties for AZ31 magnesium Biomedical castings

• Validated models quantify how sand temperature, resin content, and curing agent jointly govern compressive strength, hardness, and collapsibility in NBRBS moulds for AZ31, with hardness and collapsibility captured by quadratic terms and R 2 above 94%. • A clear factor hierarchy is established in which resin content most strongly influences compressive strength and collapsibility, sand temperature controls hardness, and the curing agent has a minor effect within the studied range. • Teaching and Learning Based Optimisation converged rapidly for single and multi objective cases, and the equal weighted solution stabilised by iteration two at 25 °C, 4 wt% resin, and 38 wt% curing agent, reproducing experimental trends with low to moderate absolute errors. • Overall error metrics with mean absolute error 1.24, mean absolute percentage error 18.04 percent, and root mean square error 1.99 support process planning and parameter adjustment within this study, while collapsibility merits targeted refinement. No-bake resin-bonded sand moulds are widely used for magnesium casting, yet the coupled effects of sand temperature Ts, resin content Rw, and curing-agent concentration CAw on mould properties are not fully quantified. This study computed main and interaction effects on compressive strength CS, hardness HB, and collapsibility CB, developed predictive models, and identified single- and multi-objective optima for process planning. A Taguchi L9 design was executed on a SMART Foundry line, and regression models were evaluated by ANOVA and validation. CS was represented adequately by a first-order model, whereas HB and CB required quadratic forms with coefficients of determination above 94 %. Error analysis gave mean absolute error 1.24, mean absolute percentage error 18.04 percent, and root mean square error 1.99. HB showed the highest predictive fidelity, CS moderate agreement, and CB the largest relative variability. Factor effects indicated Rw as the strongest influence on CS and CB, Ts as the primary control for HB, and CAw had minor influence. Teaching–Learning-Based Optimisation converged rapidly in single- and multi-objective modes. The multi-objective solution stabilised by iteration two at Ts 25 °C, Rw 4 wt% and CAw 38 wt%. The integrated Taguchi–TLBO method supports parameter selection within this study’s scope.