Open Access

ARTICLE

Hybrid Taguchi and Machine Learning Framework for Optimizing and Predicting Mechanical Properties of Polyurethane/Nanodiamond Nanocomposites

Markapudi Bhanu Prasad¹, Borhen Louhichi², Santosh Kumar Sahu^1,*

1 School of Mechanical Engineering, VIT-AP University, Besides A.P. Secretariat, Amaravati, 522237, India
2 Deanship of Scientific Research, Engineering Sciences Research Center (ESRC), Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, 11432, Saudi Arabia

* Corresponding Author: Santosh Kumar Sahu. Email:

Computer Modeling in Engineering & Sciences 2025, 145(1), 483-519. https://doi.org/10.32604/cmes.2025.069395

Received 22 June 2025; Accepted 24 September 2025; Issue published 30 October 2025

Abstract

This study investigates the mechanical behavior of polyurethane (PU) nanocomposites reinforced with nanodiamonds (NDs) and proposes an integrated optimization–prediction framework that combines the Taguchi method with machine learning (ML). The Taguchi design of experiments (DOE), based on an L9 orthogonal array, was applied to investigate the influence of composite type (pure PU, 0.1 wt.% ND, 0.5 wt.% ND), temperature (145°C–165°C), screw speed (50–70 rpm), and pressure (40–60 bar). The mechanical tests included tensile, hardness, and modulus measurements, performed under varying process parameters. Results showed that the addition of 0.5 wt.% ND substantially improved PU performance, with tensile strength increasing by 117%, Young’s modulus by 10%, and hardness by 21% at optimal conditions of 145°C, 70 rpm, and 50 bar. SEM analysis revealed ductile fracture in pure PU and brittle fracture in the optimized PU/ND composite. ANOVA confirmed that composite type was the most influential factor, contributing 70.27%, 87.14%, and 74.16% to tensile strength, modulus, and hardness, respectively. Regression modeling demonstrated a deviation of less than 10% between predicted and experimental values, validating the framework. To further strengthen predictive capability, computational modeling and analytical procedures were employed through machine learning frameworks. Random Forest achieved R²/MSE values of 0.95/0.53 (tensile), 0.95/4.03 (modulus), and 0.94/2.44 (hardness). XGBoost performed better, with 0.98/0.12, 0.98/0.77, and 0.98/0.60, while Gradient Boosting provided the highest accuracy with 0.99/0.03, 0.99/0.02, and 0.99/0.01. Residual plots supported these results, showing wide fluctuations for RF and tightly clustered residuals near zero for GB and XGB, highlighting their superior accuracy, precision, and generalization. Overall, the integrated Taguchi–ML framework demonstrates a robust and efficient strategy for optimizing processing parameters and accurately predicting the performance of high-strength PU–ND nanocomposites.

---

Keywords

---