Investigation and machine learning-based prediction of mechanical properties in hybrid natural fiber composites

Abstract

With the growing emphasis on sustainability and circular economy principles in industrial sectors, the effective utilization of waste-derived and abundantly available natural resources has become imperative. Renewable lignocellulosic fibers such as jute fiber (JF), banana fiber (BaF), coconut coir (CC), and pineapple leaf fiber (PALF) serve as viable reinforcements for developing eco-friendly polymer composites. In this study, hybrid natural fiber (NF) composites were fabricated using the hand lay-up method with varying weight percentages of alkaline-treated JF, CC, BaF, and PALF. Alkaline treatment improved fiber-matrix interfacial bonding by reducing hydroxyl groups and enhancing surface roughness, which significantly contributed to improved mechanical and thermal performance. Mechanical testing revealed that the composite containing 20% JF, 20% CC, and 10% PALF exhibited superior properties, achieving a tensile strength of 85.8 MPa, flexural strength of 134.5 MPa, impact strength of 23.3 J/m², and hardness of 72.6 Shore D. Fourier transform infrared spectroscopy (FTIR) confirmed successful alkaline treatment through reduced hydroxyl group intensity and improved fiber-matrix compatibility. Atomic force microscopy (AFM) further revealed increased surface roughness (Ra = 31.43 nm), indicating enhanced interfacial adhesion. To predict mechanical performance, machine learning (ML) regression models including linear regression, multilinear regression, decision tree, and random forest were developed using experimental data. Among these, the random forest model demonstrated the highest predictive capability, achieving R² values of 0.968 for tensile strength, 0.939 for flexural strength, 0.941 for impact strength, and 0.962 for hardness, along with the lowest error metrics (mean absolute error and root mean squared error). These results highlight its suitability for accurate prediction of mechanical properties in fiber-reinforced composites. Furthermore, dynamic mechanical analysis (DMA) revealed enhanced viscoelastic behaviour in treated fiber composites, while thermogravimetric analysis (TGA) confirmed excellent thermal stability up to 690 °C in JF/CC/PALF-reinforced samples.

Keywords: Composites; Machine learning model; Mechanical behaviour; Natural fibers.

Fig. 1

Manufacturing process of hybrid composite samples.

Fig. 2

Mechanical properties of neat epoxy samples.

Fig. 3

(a) Tensile strength and (b) stress–strain curves of hybrid composite samples.

Fig. 4

(a) Flexural strength and (b) impact strength of hybrid composite samples.

Fig. 5

Tensile and flexural modulus of hybrid composite samples.

Fig. 6

Hardness of hybrid composite samples.

Fig. 7

Morphological images of hybrid composite samples.

Fig. 8

FTIR spectra of sample S-5.

Fig. 9

(a) 3-D roughness surface texture, (b) 2-D roughness surface texture, and (c) 2-D line diagram for roughness measurement of NF.

Fig. 10

Data flow for machine learning in fiber-reinforced hybrid composites.

Fig. 11

Linear regression model for (a) tensile strength and (b) flexural strength.

Fig. 12

Linear regression model for (a) impact strength and (b) hardness.

Fig. 13

Tensile strength comparison between actual and predicted ML models.

Fig. 14

Flexural strength comparison between actual and predicted ML models.

Fig. 15

Impact strength comparison between actual and predicted ML models.

Fig. 16

Hardness comparison between actual and predicted ML models.

Fig. 17

DMA for sample S-5 (Selected as based on the mechanical features).

Fig. 18

TGA and DTG for sample S-5 (Selected as based on the mechanical features).