×
The submission system is temporarily under maintenance. Please send your manuscripts to
Go to Editorial ManagerAlternative fuels come from non-traditional sources and can partly replace fossil fuels to cut environmental impact and support sustainable energy use. A single-cylinder diesel engine with rated power 5.2 Kw at 1500 rpm was tested with American Saffron Biodiesel blends. B0 (diesel), B10 (B10D90), B20 (B20D80), and B30 (B30D70), under different loads, and the results were compared with diesel. B20 showed the best improvement, with 13.41% higher brake thermal efficiency and 30.3% lower fuel usage than conventional fuel, at 100% load. B30 reduced HC by 5.88%, while B10 gave the lowest CO, reduced by 8.51%. The main contribution is the use of decision tree machine-learning regression to predict and optimize performance and emissions. The model achieved R = 0.91 and R² = 0.83, supporting the prediction of multivariable engine responses across operating conditions.
This study develops a decision-oriented uncertainty quantification methodology for analyzing as-built surface roughness in laser powder bed fusion/direct metal laser sintering (LPBF/DMLS) SS316L. A Taguchi L9 design was used to vary laser power (300-360 W), scan speed (800-1000 mm·s-1), and layer thickness (20-80 μm), producing nine process settings with three independently fabricated specimens per setting, resulting in 27 total specimens. Surface roughness was measured by contact stylus profilometry using arithmetic mean roughness (Ra), root mean square roughness (Rq), and maximum profile height (Rz). The mean roughness varied within narrow ranges, Ra as 5.748-5.952 μm, Rq as 6.673-6.811 μm, and Rz as 28.828-28.892 μm, while within-setting scatter remained non-negligible, particularly for Rz. Probabilistic regression models were evaluated using leave-one-setting-out validation, negative log predictive density, interval coverage, calibration diagnostics, and reliability-driven accept/reject analysis. For Ra and Rq, a low-capacity linear mean model with pooled variance achieved the best predictive density, indicating limited transportable heteroscedastic structure under setting-wise extrapolation. For Rz, a nonlinear mean model with pooled variance performed best. Unregularized two-stage variance learning produced unstable uncertainty estimates, whereas shrinkage regularization improved calibration and reduced spurious setting-dependent variance effects. The decision analysis showed that calibration strongly influences process acceptance, reliability thresholds sharply reduced the number of accepted settings, and shrinkage-stabilized uncertainty produced a conservative and consistent decision frontier. The main contribution of this work is the integration of grouped validation, probabilistic calibration, variance-shrinkage modelling, and reliability-aware decision analysis for surface roughness qualification in LPBF/DMLS SS316L.