Artificial neural network-guided phyto-synthesis of Pd/Pt bimetallic nanoparticles on cotton: sustainable textile functionalization with antibacterial and colorimetric properties from saffron waste

Abstract

The sustainable synthesis of palladium–platinum bimetallic nanoparticles (Pd/Pt NPs) using agricultural waste offers an eco-friendly approach to develop functional textiles with enhanced antibacterial and colorfastness properties. In this study, saffron waste comprising petals (SP) and stamens (SS) was employed as a green reducing and stabilizing agent in a microwave-assisted phyto-synthesis method to deposit Pd/Pt NPs onto cotton fabrics. To optimize and accurately predict the color strength (K/S) of the treated textiles, an artificial neural network (ANN) coupled with a genetic algorithm (GA) was implemented, outperforming traditional response surface methodology (RSM) with a high correlation coefficient (R² = 0.99). Comprehensive characterization using dynamic light scattering (DLS), UV–Visible spectroscopy, Fourier-transform infrared spectroscopy (FTIR), Field emission scanning electron microscopy with Energy Dispersive X-ray Spectroscopy (FESEM-EDX), and X-ray diffraction (XRD) confirmed the successful formation and uniform distribution of Pd/Pt NPs on cotton fibers. The treated fabrics exhibited superior antibacterial activity, achieving 99% inhibition against both Gram-positive S. aureus and Gram-negative E. coli, alongside excellent colorfastness to rubbing, washing, and light exposure. This work demonstrates the integration of green nanotechnology and machine learning for the fabrication of sustainable, smart antibacterial textiles, contributing to reduced environmental impact and advancing the development of next-generation functional fabrics.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-36565-4.

Introduction

Textiles play a vital role in protecting individuals against adverse environmental conditions. Under specific temperatures and humidity levels, these conditions can create a satisfactory situation for the growth of microorganisms¹. The growth of these microorganisms may cause damage to the fabric, leading to unpleasant odors, reduced tensile strength, and posing potential risks to the wearer². To address these issues, significant efforts have been made to develop antibacterial textiles³. In recent years, researchers have documented a wide range of antibacterial agents, including both organic and inorganic substances⁴. In particular, inorganic nanoparticles (NPs) have garnered considerable attention as antibacterial agents due to their high surface area, simple synthesis methods, remarkable thermal stability, long-term durability, scalability, and potential for commercialization⁵. Therefore, inorganic antibacterial NPs have emerged as promising candidates for coating woven fabrics⁶. The incorporation of nanomaterials into fabric substrates has proven to be an effective method for producing textiles with antibacterial, antifungal, and antiviral properties. These textiles are widely used in hygiene clothing, wound dressings, and medical equipment in hospitals⁷.

With the advancement of nanotechnology, researchers are working to create antibacterial coatings on cotton fabrics using various NPs⁸. For example, cotton fabrics treated with NPs such as Ag, TiO₂, CuO, ZnO, and graphene oxide have been investigated for antibacterial applications⁹. Additionally, various surface modification methods have been tested to enhance the antibacterial properties of cotton fabrics¹⁰. NPs are typically synthesized using a variety of chemical and physical methods. However, conventional synthesis methods have several drawbacks. They involve multi-step processes, are costly and time-consuming, often use toxic chemicals, and require expensive equipment, limiting commercialization¹¹. In contrast, green synthesis utilizes natural bioactive compounds in plant extracts as reducing/oxidizing agents and stabilizers, presenting an environmentally friendly, biocompatible, and low-cost approach¹². Moreover, the green synthesis method produces NPs with minimal waste and energy consumption. The primary motivation behind this study was to synthesize NPs through a simple process that can be easily scaled for large-scale production using a green approach¹¹. This approach not only contributes to environmental protection but also prevents the degradation of natural habitats and aligns with the Sustainable Development Goal¹³. In this context, many bio-waste extracts are considered the emerging process known for its simplicity, low cost, non-toxicity, and the production of more stable and biocompatible NPs¹⁴. In this study, saffron (Crocus sativus L.) waste extract was used to synthesize the NPs.

Saffron, known as “red gold”, is cultivated in Iran, particularly in the Khorasan region. Saffron is composed of three primary parts. Apart from the stigma, the remaining parts of the saffron plant, namely SP and SS, are considered waste¹⁵. To produce one kg of dried saffron spice (stigma), 78 kg of fresh flowers are required, a process that generates inexpensive bio-waste¹⁶. These by-products contain polyphenolic compounds. These compounds, such as flavonoids, carotenoids, and anthocyanins, are utilized in various industries, including pharmaceuticals, dyeing, traditional medicine, and food processing. Furthermore, saffron waste is recognized as a suitable source for the green synthesis of metal NPs and the reduction of pollutants¹⁶.

Prior research has established the effectiveness of plants in synthesizing NPs for biological and medical applications. For example, the gold NPs produced using Salvia plebeia extract demonstrate their promising biological activity¹⁷. The silver and gold synthesized NPs via saffron petals extract and found that these NPs had the ability to inhibit the growth of various microorganisms¹⁸. Ebrahimzadeh et al. synthesized silver NPs using Salvia plebeia through an eco-friendly, cost-effective process, showing strong antimicrobial properties¹⁹. Selenium NPs fabricated by Saffron petals extract on graphene oxide nanosheets with satisfactory antibacterial, antioxidant, and anticancer capabilities²⁰. A recent study focused on synthesizing Pd NPs from rosemary plant extract, demonstrating their effective antimicrobial and antifungal activities. Additionally, the biosynthesis of Pd NPs using different plant materials has gained attention due to its affordability, simplicity, and environmental benefits²¹. Bio-waste extracts simplify the reduction of Pd ions into Pd NPs. Moreover, Pd NPs have been synthesized from the root extract of several bio-wastes^22,23, all of which are rich in polyphenol compounds that act as stabilizers and reducing agents in the synthesis of Pd and Pt NPs. Additionally, several studies have emphasized the antimicrobial and anticancer properties of metal NPs produced using plant extracts.

A large number of metallic NPs have been successfully synthesized through biological methods, and their antibacterial and dyeing properties have been well established. However, comprehensive studies on Pd/Pt NPs synthesized from SP and SS extracts (SPE and SSE), as well as their antibacterial and dyeing properties on cotton fibers, are still lacking in the scientific literature. To address this gap, the present study focuses on the green synthesis of Pd/Pt NPs from SPE/SSE (SPE/SSE/Pd/Pt NPs). The main objectives of this work are the eco-friendly (water-based) synthesis of these NPs on cotton, their structural characterization using various analytical techniques, and the evaluation of their potential applications in biological and textile-related fields. The synthesized Pd/Pt NPs were studied by FTIR, UV–Visible absorption spectroscopy, and a DLS analyzer. The interactions within the SPE/SSE/Pd/Pt NPs – cotton fabric systems were analyzed using FTIR, XRD, and FESEM-EDX. To enhance synthesis efficiency, an ANN was utilized to model the process due to its superior capability in capturing underlying relationships compared to RSM, and the optimum conditions were determined using a GA as a robust optimization method. The modified cotton samples were then evaluated for their colorimetric properties using reflective spectrophotometry, while their antibacterial performance was tested according to the AATCC 100 standard. Additionally, the dyed fabrics underwent assessments for colorfastness against rubbing, light exposure, and repeated washing to ensure durability.

Experimental

Materials

The petals and stamens of saffron flowers, collected from Khorasan, Iran, were shade-dried and ground into a fine powder. All saffron wastes were obtained from the same local supplier to ensure consistency, and the synthesis procedure was repeated three times under identical conditions, yielding reproducible results. After sieving, the powder was used for extraction and dyeing studies. The experiments were conducted on plain-woven cotton fabric (scoured with a thread count of 16 weft/cm and 12 warp/cm) supplied by Horang. Dyeing, Iran. High-purity hexachloroplatinic acid hexahydrate (H₂PtCl₆·6 H₂O, ≥ 99%) and palladium acetate (Pd(CH₃COO)₂, ≥ 99%) were obtained from Sigma-Aldrich (USA). Additional chemicals, solvents, and distilled water were sourced from Dr. Mojallali Co., Iran. For antibacterial testing, two standard strains, S. aureus (Gram-positive, MTCC 902) and E. coli (Gram-negative, MTCC 443), were obtained from the Iran Science and Technology Research Organization (ISTRO) and cultured in Mueller-Hinton Broth (HiMedia, India).

Preparation of SPE and SSE

For the preparation of the SPE and SSE, saffron was thoroughly dried and then ground into a fine, uniform powder using a powder mill. For extraction, varying quantities (0.3, 0.5, 1, and 1.5 g) of finely ground SP and SS were individually combined with 70 mL of deionized water. The mixtures were then subjected to microwave-assisted extraction at different power settings (60, 80, and 100 W) for 5, 10, and 15 min. Following irradiation, the solutions were left to cool naturally to ambient temperature. The extracts were filtered using Whatman No. 1 filter paper to remove particulate matter and stored at 4 °C for further use in subsequent experiments.

Phyto-synthesis of Pd/Pt NPs on cotton

The synthesis process began by preparing aqueous solutions of 0.5, 1, 1.5, and 2 mM concentrations of Pd and Pt salts. The two plant extracts were then mixed with these Pt and Pd solutions at varying concentrations in different ratios, resulting in the design of 50 experiments, as shown in Table S1. After mixing the metal ion solutions and extracts under mild temperature conditions with stirring, a noticeable color change occurred in the solution.

To prepare the cotton fabric for dyeing, 1 g samples were first thoroughly cleaned by washing in a 0.5% non-ionic detergent solution. The washing process was carried out at 50 °C for 30 min using a liquor-to-good ratio (L: G) of 40:1. Following this treatment, the samples were gently rinsed with distilled water to remove any residual detergent and allowed to dry completely at room temperature before proceeding with the dyeing process. Following the pretreatment, each experimental solution (from the 50 designed experiments) was separately added to the samples at a ratio of L: G 50:1. The dyeing process was carried out under previously optimized conditions using a microwave at a power of 80 W for 15 min, which was determined as the optimal condition for the dyeing process. Following the dyeing process, the treated fabrics were thoroughly rinsed with distilled water to remove excess dye and oven-dried at 40 °C (Fig. 1). The colorimetric data of the colored textiles were quantitatively analyzed using reflectance spectrophotometry, with particular attention to the K/S expressed through the Kubelka-Munk function.

Fig. 1.

The schematic representation of the microwave-assisted phyto-synthesis process of bimetallic Pd/Pt NPs on cotton fabric.

Analysis methods

Natural dye extraction from SP and SS was carried out using a Microwave Milestone flexiWAVE system. The spectral characteristics of the synthesized Pd/Pt NPs using SPE/SSE were examined with a Shimadzu double-beam spectrophotometer (Japan). The chemical characterization of Pd/Pt NPs on cotton substrates was performed using multiple analytical techniques. Functional group analysis was conducted via FTIR spectroscopy (Nicolet Nexus 670, USA) across the spectral range of 400–4000 cm⁻¹. Surface morphology and elemental composition were examined using FESEM coupled with EDX (Zeiss Sigma VP, Germany). Particle size distribution was determined through DLS analysis (HORIBA, Japan). Crystalline structure was evaluated by XRD (Philips PW1730, Netherlands) with 2θ scanning from 10° to 80°, while elemental stability was verified by XRF spectroscopy (Philips PW1730, Netherlands). Colorimetric properties, including K/S and CIELAB coordinates (L*, a*, b*, C), were measured using an X-Rite GretagMacbeth 7000 A spectrophotometer. In the CIELAB color space, L* denotes lightness (0-100 scale), a* represents the red-green chromaticity axis, and b* indicates the yellow-blue chromaticity dimension.

Chroma (C), representing color intensity, was calculated using Eq. (1):

1

The hue angle (h°), representing the dominant color tone on a circular 0-360° scale, was calculated according to the following relationship, Eq. (2):

2

Where a* and b* are the chromaticity coordinates in the CIELAB color space.

Color strength was assessed using the Kubelka-Munk function, which relates the absorption (K) and scattering (S) coefficients to the measured reflectance (R), Eq. (3):

3

Statistical analysis and modeling

A comprehensive statistical and machine learning workflow was implemented to evaluate variable relationships and develop an accurate interpolation model for predicting K/S as a function of synthesis parameters. All analyses were conducted using Python 3.11 with scikit-learn and TensorFlow.

Initial statistical screening included Pearson correlation, one-way ANOVA, Variance Inflation Factor (VIF), and Principal Component Analysis (PCA), used to assess inter-variable relationships, variance distribution, and potential multicollinearity. These analyses confirmed that the experimental design space was well-structured and suitable for multivariate modeling. Full statistical outputs, including heatmaps and loading plots, are provided in the Supporting Information.

A quadratic RSM model was initially constructed as a baseline approach. However, this model yielded limited predictive performance (R² = 0.75) and was unable to accurately capture the nonlinear behavior observed in the experimental data. Consequently, a data-driven model based on an ANN was developed to improve interpolation capability within the constrained design space.

The ANN architecture consisted of two hidden layers with 8 and 4 neurons, respectively, using tanh activation functions and L2 regularization (λ = 1 × 10⁻⁴). Inputs were standardized using StandardScaler. The model was trained using the Adam optimizer and the Huber loss function, selected for its robustness to outliers and stable convergence on small datasets. No dropout or batch normalization was applied. All training parameters and loss convergence profiles are provided in Supporting Information (Section S1).

To identify the optimal synthesis condition, the trained ANN was coupled with a GA. The GA explored a four-dimensional continuous parameter space bounded by the experimental range: 0.0–1.5 g SPE, 0.0–1.5 g SSE, 0.0–2.0 mM Pd, and 0.0–2.0 mM Pt. The GA used rank-based selection with elitism, blend-alpha crossover, and adaptive Gaussian mutation. The fitness function was defined as the predicted K/S output from the ANN. Parallel evaluation was used to accelerate optimization. Detailed mathematical formulations and hyperparameter values for the GA and ANN are provided in the Supporting Information (Section S1).

Measurements of colorfastness

The lightfastness performance of the samples was assessed in accordance with ISO 105 B02-1994 standards, employing a SOLARBOX 1500 accelerated weathering chamber to simulate light exposure. Color retention was quantitatively evaluated by comparing irradiated and protected sample sections through blue reference standards. For washfastness characterization, we followed ISO 105-C10:2006 (E) protocols, with both color change and staining effects objectively rated using standardized grayscale analysis. Crockmeter testing (ISO 105-X12:1993 (E)) determined dry and wet rubbing fastness through controlled rubbing-cotton specimens that were rigidly affixed and subjected to ten standardized strokes, with subsequent transfer to adjacent white fabric evaluated via grayscale assessment.

Measurements of antibacterial activity

The antibacterial evaluations were conducted on two sample groups: untreated raw cotton fabric and treated with SPE/SSE and Pd/Pt ions. To eliminate potential contamination, all laboratory equipment, containers, and cotton samples (1 × 1 cm², 0.1 g) underwent UV disinfection before testing. This precaution ensured that any detected antibacterial activity originated solely from the treated fabrics.

Standardized bacterial suspensions were prepared by inoculating 8 mL aliquots of broth (LB) in sterile microvials with pure cultures of representative Gram-negative (E. coli) and Gram-positive (S. aureus) strains. The cultures were subsequently incubated under optimal growth conditions (37 °C) for 18 h to achieve late-log phase populations, ensuring robust microbial viability for subsequent experimental procedures. The antibacterial efficiency was determined through optical density (OD) measurements, a widely accepted approach for quantifying bacterial reduction and assessing growth inhibition. Measurements were taken at 600 nm (OD600) using a spectrophotometer, which analyzes light absorption by bacterial cells, directly reflecting their concentration²⁴. By comparing OD values across different samples, bacterial reduction and growth inhibition were assessed. The antibacterial efficacy of the treated cotton was quantified using a standard calculation (Eq. 4) and evaluated in accordance with the AATCC 2004-100 standard method by comparing bacterial colony counts (expressed as CFU/mL) between untreated control specimens (Blank) and textile samples functionalized with SPE/SSE and Pd/Pt NPs (Test). This quantitative approach allowed precise measurement of microbial inhibition, where reduced CFU counts in Test samples directly demonstrated the antibacterial activity of the nanoparticle-treated surfaces.

4

Results and discussion

Phyto-synthesis of Pd/Pt NPs by SPE/SSE

UV–Visible spectroscopy is a powerful analytical technique for studying spectral behavior and determining the concentration of solutions in various samples. In this study, UV–Visible spectroscopy was used to optimize extraction conditions and analyze the absorption spectra of the samples (Figure S1). Additionally, the study of UV–Visible spectra was performed to confirm the synthesis of bimetallic Pd/Pt NPs using SPE/SSE (Fig. 2b). In our previous studies, the synthesis of Pd NPs using SPE and SSE was investigated, and the related analyses were fully conducted^16,25. The UV–Visible spectra of SPE/SSE revealed a distinct absorption peak at approximately λ_max ≈ 260 nm, which is likely attributed to the presence of flavonoid and polyphenol compounds²⁶. These compounds act on the inherent redox properties, which effectively coordinate metal ions and control nucleation processes without requiring synthetic capping agents. The polyphenolic compounds present in the SPE/SSE, particularly the carbonyl/carboxyl groups and flavonoid rings, played a pivotal role in facilitating the redox reactions. These reactions are responsible for converting metal ions into Pd/Pt NPs (Fig. 2a)²⁷. When Pd²⁺ and Pt²⁺ ions were added to SPE/SSE, the samples exhibited a noticeable color change, indicating Pd/Pt NP formation.This color change was fully confirmed by UV–Visible spectroscopy. The results demonstrated that SPE/SSE have the capability to reduce Pd²⁺ and Pt²⁺ ions and manufacture uniform NPs without the need for any chemical reducing agents.

Fig. 2.

a Schematic representation of the redox process of Pd/Pt ions mediated by flavonoid compounds in SPE/SSE, b UV–Visible spectra of SPE/SSE and SPE/SSE/Pd/Pt ions, c FTIR spectra of SPE, SSE, SPE/SSE, and SPE/SSE/Pd/Pt ions, d DLS analysis of Pd/Pt NPs synthesized using SPE/SSE.

FTIR, a highly sensitive technique for identifying chemical bonds and surface functional groups on NPs, was performed. The goal was to study the probable functional groups in the SPE/SSE involved in the reduction, capping, and stabilization of Pd/Pt NPs. Figure 2c presents the FTIR spectra of SSE, SPE, and SPE/SSE. In the extract’s spectrum, a broad peak at 3450 cm⁻¹ is observed, which is attributed to phenolic hydroxyl groups present in flavonoids²⁸. Distinct bands at 2931, 1768, 1640, and 1042 cm⁻¹ were also detected, corresponding to aliphatic C–H bonds, ester carbonyl groups, C=C double bonds, and C–O stretching, respectively. The FTIR spectra of the extracts exhibit no significant differences compared to the spectrum of the sample containing Pd/Pt ions; a noticeable decrease in the absorption band at 1768 cm⁻¹ was observed²⁹. Additionally, a new band appeared at 600 cm⁻¹, indicating the presence of Pd/Pt ³⁰. The analysis of these spectra confirms that the functional groups present in SPE/SSE play an important role in the production of Pd/Pt NPs.

DLS measurements were conducted to determine the size of the synthesized Pd/Pt NPs using SPE/SSE. As illustrated in Fig. 2d, the average diameter of the Pd/Pt NPs was approximately 53.4 nm, well within the standard range of less than 100 nm. This confirms their suitability for NP production³¹. Furthermore, the polydispersity index (PDI) of 0.52 indicates a relatively uniform particle distribution and stability within the colloidal system.

Analysis of ANN-GA modeling results

Statistical analysis

The correlation analysis revealed several key trends. K/S exhibited a strong positive correlation with Pt ions (r = 0.71), indicating a significant influence on K/S. While Pd ions and SSE showed no strong linear correlation with K/S (r ≈ 0), the presence of an increasing K/S suggests potential non-linear effects. SPE demonstrated a moderate correlation with K/S (r = 0.4), implying an indirect influence³². Figure 3 presents the correlation matrix and coefficient values. Figure 4 further illustrates this indirect influence, showing different variable combinations plotted against K/S³³.

Fig. 3.

Pearson’s correlation coefficient matrix.

Fig. 4.

Plots of a K/S vs. Pt ions and Pd ions, b K/S vs. SPE and SSE, c K/S vs. SPE and Pd ions, d K/S vs. SPE and Pt ions, e K/S vs. SSE and Pd ions, and f K/S vs. SSE and Pt ions.

The ANOVA results (F-statistic = 3.1877, p-value = 0.02486) indicated that at least one independent variable had a notable influence on the response, underscoring the importance of certain predictors in determining K/S. The VIF analysis confirmed the absence of significant multicollinearity, as all values remained below the threshold of 5. Specifically, SPE and SSE exhibited a VIF of 2.45, while Pt and Pd ions had a VIF of 1.96. Although SPE and SSE were somewhat correlated, they did not present a substantial issue for regression modeling. PCA results provided further insight into the dataset structure. Pd and Pt dominated the first principal component (PC1) with loading scores of approximately 0.7, indicating their significant and similar contributions to the main source of variation. SSE and SPE contributed minimally to PC1 (on the order of 10⁻¹⁵). However, they exhibited high loadings of around 0.707 on the second principal component (PC2), suggesting a secondary dimension of variation not directly correlated with K/S.

A baseline model using second-order polynomial regression (RSM) was initially applied to assess the feasibility of response surface modeling. However, RSM failed to adequately capture the system’s complexity, with a test R² of 0.75 and MSE of 0.26, indicating underfitting and limited ability to model nonlinear interactions (Fig. 5a–c). These shortcomings highlighted the need for a more flexible modeling strategy^34,35.

Fig. 5.

RSM performance a on the dataset; ANN performance b on the dataset.

Figure 5a shows the performance of quadratic RSM on the dataset. To address the limitations of RSM, a compact ANN was developed to capture the complex, non-linear relationships between the synthesis parameters and K/S. The model was trained on the full dataset without data splitting, enabling accurate interpolation across the experimental design space. Hyperparameters and architecture were optimized using a GA, which also served to identify the input conditions that maximized the ANN-predicted K/S. This integrated GA–ANN framework provided a reliable and data-efficient approach to both modeling and optimization.

ANN model architecture

To model the relationship between synthesis parameters and K/S, a compact ANN architecture was implemented. The network consisted of an input layer with four neurons (corresponding to SPE, SSE, Pt, and Pd), followed by two hidden layers with 6 and 3 neurons, respectively, and a single-neuron output layer. Hidden layers employed the hyperbolic tangent (tanh) activation function to capture non-linearities while maintaining gradient stability. The output layer used a Leaky ReLU activation to handle positive-predictive skew while preserving responsiveness near zero.

To prevent overfitting in the small dataset (n = 50), L2 weight regularization (λ = 1 × 10⁻⁴) was applied to all dense layers. Input variables were standardized using StandardScaler, and the network was trained using the Adam optimizer (learning rate = 0.01) with the Huber loss function. Model selection was performed through k-fold cross-validation (k = 5) to ensure robustness and mitigate variance due to sample size constraints.

The final ANN model achieved an R² of 0.99 and MSE of 0.01 on the data (Fig. 5b), substantially outperforming the RSM model. Hyperparameters are summarized in Table 1.

Table 1.

Table of ANN Hyperparameters.

Hyperparameter	Value/description
Input variables	SPE (g), SSE (g), Pt (mM), Pd (mM)
Layer sizes	4, 6, 3, 1 (output)
Activation (hidden)	tanh
Activation (output)	Leaky ReLU
Regularization	L2 (λ = 1 × 10⁻⁴)
Optimizer	Adam
Learning rate	0.01
Loss function	Huber
Evaluation metric	Mean absolute error (MAE)

ANN-GA optimization

The trained ANN model was integrated with a GA to identify the optimal synthesis conditions that maximize color strength. The GA was configured to explore a bounded four-dimensional space defined by the experimental design: SPE (0.0–1.5 g), SSE (0.0–1.5 g), Pt (0.0–2.0 mM), and Pd (0.0–2.0 mM). Fitness was evaluated as the predicted K/S value generated by the ANN (Fig. 6).

Fig. 6.

A typical schematic of an ANN with multiple hidden layers.

Using rank-based selection, blend-alpha crossover, and adaptive mutation, the GA converged to an optimal solution after 83 generations. The optimal condition was identified as SPE = 1.03 g, SSE = 1.18 g, Pt = 2.00 mM, and Pd = 0.71 mM, corresponding to a predicted K/S of 6.27. Experimental validation under these conditions yielded a K/S value of 6.22, closely matching the model’s prediction and confirming the reliability of the GA–ANN framework. The optimization curve is shown in Figure 7.

Fig. 7.

a Optimize the output of each generation, b Monte Carlo sensitivity analysis of the GA–ANN model.

To evaluate the stability of the GA-ANN model, a Monte Carlo sensitivity analysis was performed by introducing random perturbations (± 5%) to the optimized input parameters (Fig. 7b). Across 1000 trials, the baseline fitness value (6.27) showed an average deviation of − 0.28 ± 0.39, corresponding to a relative variability of only 6.24%.

This narrow distribution and slight negative shift indicate that the optimized solution lies in a broad and stable region of the search space rather than a sharp local extremum. Consequently, the GA–ANN framework demonstrates robust predictive behavior, maintaining consistent performance under minor fluctuations of the input parameters.

Characterization of Pd/Pt NPs on cotton

SPE/SSE are rich in bioactive constituents, including flavonoids, polyphenols, alkaloids, and terpenoids, which play a dual role as reducing and stabilizing agents during the synthesis of metallic NPs³⁶. When cotton fabric is treated with a solution containing metal ions and plant extract, these bioactive compounds trigger the reduction of metal ions, resulting in the in situ formation of NPs on the cellulose fibers. This direct deposition occurs due to two primary factors. First, the hydroxyl (–OH) and carbonyl (C=O) functional groups in the cellulose structure act as active sites for metal ion adsorption. Second, plant-derived compounds can form complexes with metal ions, bringing them closer to the fabric surface and enhancing in situ reduction³⁷. Following the fabrication of NPs, their stabilization on the cotton fabric is governed by multiple mechanisms:

(i) NPs can form covalent bonds with functional groups on the cellulose surface (e.g., –OH and –COOH). Additionally, residual metal ions may establish ionic interactions with anionic functional groups on the fabric³⁸. (ii) Functional groups on both NPs and plant-derived compounds can engage in hydrogen bonding with cellulose, further enhancing NP adhesion. (iii) If NPs and the fabric surface carry opposite charges, electrostatic attractions improve their adhesion (Fig. 8a)³⁸. Additionally, organic compounds present in the plant extract (e.g., polyphenols) act as a protective layer around the NPs, preventing their aggregation and enhancing their long-term stability by mitigating oxidation³⁹.

Fig. 8.

a Schematic representation of the interaction between synthesized Pd/Pt NPs and cotton fabric using SPE/SSE, b FTIR spectra of Pd/Pt NPs deposited on cotton fibers via SPE, SSE, and SPE/SSE, c XRD patterns of untreated cotton fabric, Pd/Pt NPs deposited on cotton fibers via SPE, SSE, and SPE/SSE.

FTIR spectroscopy was performed to examine the attachment of dual Pd/Pt NPs onto cotton fibers using the SPE, SSE, and SPE/SSE (Fig. 8b). The spectra of fabrics dyed with SPE, SSE, and SPE/SSE revealed a distinct absorption band at 3443 cm⁻¹, confirming the stretching vibrations of the -OH group⁴⁰. The absorption band at 2900 cm⁻¹ corresponded to the C-H stretching vibrations of alkane groups in cellulose glucose. The band at 1443 cm⁻¹ was assigned to C–H bending vibrations or CH₂ bonds⁴¹. The absorption band at 1245 cm⁻¹ was associated with the C-O-C stretching vibrations of cellulose ether bonds, and the peak at 1002 cm⁻¹ corresponded to the C–O stretching vibrations in the polysaccharide structure of cellulose⁴¹.

A comparison of the SPE/SSE spectrum with those of SPE and SSE indicated an increase in intensity in certain regions; however, no new bands were formed. A medium-intensity band emerged at 610 cm⁻¹ in fibers dyed with Pd/Pt NPs, regardless of the extract used¹⁶. Additionally, changes in the band at 1638 cm⁻¹ confirmed the successful attachment of Pd/Pt NPs onto and within the surface of cotton fibers.

The fabrication of Pd/Pt bimetallic NPs on cotton fibers is studied by XRD analysis. As shown in Figure 8C, the XRD analysis of cotton fabric reveals distinct diffraction peaks at 2θ angles of 14.7°, 16.3°, 22.7°, and 34.6°, which are attributed to the crystallographic planes (110), (110), (200), and (004), respectively. These peaks confirm the characteristic crystalline structure of cellulose within the fabric⁴². Following the deposition of NPs on cotton fibers using SSE, SPE, and SPE/SSE, four additional diffraction peaks were observed at 2θ values of 40°, 47°, 68°, and 80.5°. These peaks are related to the (111), (200), (220), and (311) planes, respectively, signifying the presence of Pd and Pt NPs with a face-centered cubic (fcc) crystal structure⁴³. As illustrated in Figure 8C, in spite of the presence of Pd/Pt NPs, the characteristic peaks of cotton fibers remain visible. This observation suggests that the synthesis of Pd/Pt NPs on cotton using SPE/SSE has not altered the crystalline structure of the cotton fibers. Overall, the crystallographic characteristics of the Pd/Pt NPs deposited on cotton are confirmed by the XRD analysis.

The synthesis of dual Pd/Pt NPs using SPE/SSE, SPE, and SSE, along with their particle size distribution on cotton fibers, was examined by FESEM analysis (Fig. 9). For this purpose, FESEM images were taken from three dyed fabric samples prepared using the same extract concentration and ion solution. The raw cotton fabric (Fig. 9a) exhibited a smooth surface; however, after the deposition of NPs, the morphology of the dyed samples became rough (Fig. 9b–d). As observed in the images, the formation of Pd/Pt NPs with various shapes, predominantly spherical, is evident on the fabric surface, confirming the successful synthesis. However, differences in NP distribution across the fabric were noted. Based on the results and images, the NPs synthesized using the SPE/SSE method demonstrated superior performance compared to SPE and SSE. Additionally, the average particle diameter was measured at 70 nm.

Fig. 9.

FESEM images and Line-scan EDX analysis for a untreated cotton fabric, Pd/Pt NPs deposited on cotton fibers via b SSE, c SPE, and d SPE/SSE.

Line-scan EDX analysis was performed on the same samples, confirming the presence of Pd and Pt in all treated fabrics. In contrast, the raw cotton fabric, used as a control, exhibited no signals other than those of carbon and oxygen, which correspond to the cellulose structure of the fibers. In the Pd/Pt-treated samples, Pd and Pt signals were detected, with their intensities varying according to the extract used. Notably, the SPE/SSE-synthesized Pd/Pt NPs exhibited the highest Pd and Pt signal intensities.

Evaluation of colorimetric properties

The color coordinates (L*a*b*) of 50 dyed cotton fabric samples are reported in Table S2, and their corresponding images are presented in Figure S2. The best-performing samples from each category, dyeing with SPE and SSE, SPE/SSE combination, and SPE/SSE dyeing with each of the NPs are presented in Table 2.

Table 2.

Colorimetric properties of cotton fabrics dyed with SPE/SSE and treated with Pd and Pt NPs.

As shown in Table 2, the deposition of NPs on the dyed fibers induces noticeable changes in color coordinates. Among the samples, the fabric dyed SPE/SSE/Pd/Pt NPs exhibits the lowest L* value (61.22), indicating the darkest shade. The reduction in L* upon the existence of Pd and Pt NPs confirms the enhancement of color depth. Moreover, the incorporation of these NPs leads to a decrease in the b* value, signifying a reduction in yellowness. This effect likely arises from the structural transformation of hydroxyl groups within the SPE/SSE constituents. The enol form (–C–OH) undergoes tautomerization to the more stable keto configuration (–C=O), facilitating the reduction of Pd²⁺ and Pt²⁺ ions to Pd and Pt NPs⁴³. Additionally, Pd/Pt ions can form complexes with functional groups present in SPE/SSE and cotton fibers. Consequently, the chemical characteristics of SPE/SSE components are altered by Pd/Pt ions, resulting in a color shift in the dyed cotton fabric.

Notably, the K/S also exhibits a significant increase. A comparison of K/S values between fabrics dyed with SPE and SSE versus those dyed with SPE/SSE reveals a marked enhancement. Furthermore, the deposition of Pd/Pt NPs on SPE/SSE-dyed fabrics increases the K/S value from 1.93 to 5.48, indicating a substantial improvement in color strength. The pronounced enhancement in K/S values for the fabric treated with Pd/Pt nanoparticles suggests the involvement of nanoparticle-specific optical phenomena beyond the inherent color of the dyes. While the plasmonic response of palladium and platinum is distinct from that of strongly plasmonic metals like gold and silver, their bimetallic nanostructures can exhibit unique electronic interactions at the interface. This interfacial coupling can modulate their collective optical behavior, leading to enhanced extinction coefficients in the visible spectrum. The observed color deepening is likely a consequence of this engineered nanomaterial acting as a light harvesting medium. The nanoparticles facilitate intensified light matter interactions through a combination of absorption and Mie scattering, effectively increasing the optical path length within the fibrous matrix and amplifying the absorbance of the bound dye molecules. Thus, the nanoparticles function not merely as a passive additive but as an active component that optimizes the fabric’s optical performance. These findings suggest that the concentration of Pd/Pt ions has a significant influence on the depth of color in the samples⁴⁴.

Evaluation of color fastness properties

Color fastness is a crucial characteristic when selecting dyes for textiles, as it determines the fabric’s ability to resist color fading and maintain its appearance over time^45,46. This property is particularly important for ensuring that dyed fabrics retain their vibrancy when exposed to environmental factors such as washing, light, and rubbing⁴⁷.

Color fastness ratings of cotton fabrics dyed with SPE/SSE and treated with Pd and Pt NPs, for washing, light, and rubbing, are shown in Table 3 (all samples in Table S3). Fabrics dyed solely with SPE and SSE exhibited relatively poor rubbing and washing fastness. However, when these extracts were combined (SPE/SSE), a slight improvement in fastness properties was observed. Notably, a significant enhancement in all three fastness parameters was achieved upon incorporating Pd and Pt NPs into the SPE/SSE-dyed fabrics, with an improvement of approximately 1–2 units in fastness ratings.

Table 3.

Color fastness of cotton fabrics dyed with SPE/SSE and treated with Pd and Pt NPs.

Sample	Light fastness	Wash fastness			Rub fastness
		Color change	Staining on the cotton	Staining on the wool	Dry	Wet
SPE	5–6	3	3	3	2–3	2
SSE	5	2	3	2–3	3	2–3
SPE/SSE	6	3–4	3	3	3–4	3
SPE/SSE/Pt NPs	7	5	5	5	5	5
SPE/SSE/Pd NPs	6	4	4–5	4	4–5	4
SPE/SSE/Pd/Pt NPs	7–8	4–5	5	5	5	4–5

Among the tested samples, the fabric treated with SPE/SSE/Pd/Pt NPs demonstrated the highest stability, exhibiting excellent light and rubbing fastness and achieving the highest ratings. The observed improvement in light fastness appears to stem principally from the inherent oxidative resistance of the dye molecules themselves, rather than from dye-fiber interactions. This enhanced stability may result from multiple factors: (1) reduced surface adsorption of dye molecules, (2) stronger bonding between SPE/SSE constituents and cotton fibers facilitated by Pd/Pt ions, and (3) formation of a stable coordination complex involving metal ions, dye molecules, and cellulose fiber²⁵.

Evaluation of antibacterial properties

The antibacterial activity of the best-performing fabrics dyed with SPE, SSE, SPE/SSE, and SPE/SSE treated with Pd, Pt, or Pd/Pt NPs is presented in Table 4 (all samples in Table S4). As shown in Table 4, fabrics dyed solely with plant extracts exhibited notable antibacterial properties. For instance, in the SPE/SSE sample, the decrease of E. coli and S. aureus was 45.0% and 48.0%, respectively. This antibacterial effect is related to the existence of bioactive ingredients, for instance phenols, flavonoids, alkaloids, terpenoids, tannins, and essential oils, which disrupt the bacterial cell membrane, inhibit vital pathways, and induce oxidative stress⁴⁸. The combination of two extracts enhanced the antibacterial activity, likely due to the synergistic effect of a broader range of bioactive compounds, which interact more effectively to inhibit bacterial growth.

Table 4.

Antibacterial activity evaluation of cotton fabrics dyed with SPE/SSE and treated with Pd and Pt NPs.

Sample	Reduction of bacteria (%)		Inhibition zone (mm)
	E. coli	S. aureus	E. coli	S. aureus
SPE	44.0	45.0	2.2	2.3
SSE	30.0	33.0	1.6	1.6
SPE/SSE	45.0	48.0	2.4	2.5
SPE/SSE/Pt NPs	99.0	99.0	4.8	4.9
SPE/SSE/Pd NPs	70.0	75.0	3.6	3.8
SPE/SSE/Pd/Pt NPs	99.0	99.0	5.0	5.0

To further enhance antibacterial performance, Pd and Pt NPs were deposited onto the SPE/SSE-dyed fabrics. After Pd NPs deposition, the antibacterial activity increased to 70.0% and 75.0% for E. coli and S. aureus, respectively, while the inhibition zone reached 3.6 mm and 3.8 mm. In contrast, Pt NPs exhibited significantly higher antibacterial efficiency, achieving 99.0% bacterial reduction for both strains, with an inhibition zone of 4.8 mm and 4.9 mm.

The results demonstrate that Pd/Pt NPs synthesized using SPE/SSE exhibit outstanding antibacterial performance. The SPE/SSE/Pd/Pt NPs sample achieved the highest bacterial reduction (99.0%) with an inhibition zone of 5.0 mm. This strong antibacterial effect is related to the interaction of Pd/Pt NPs with the bacterial cell wall, which disrupts cellular processes, prevents nutrient absorption, and ultimately leads to bacterial cell death⁴⁹. Additionally, the antibacterial efficacy of Pd/Pt NPs is believed to be amplified via catalytic generation of reactive oxygen species (ROS). These ROS induce oxidative stress, damaging cell membranes and intracellular components, thereby enhancing antimicrobial activity⁵⁰. Although direct measurement of ROS was not performed in this study, this mechanism is supported by previous literature, and further mechanistic studies are suggested for future work^51,52.

These findings suggest that SPE/SSE extracts can reduce and stabilize Pd²⁺ and Pt²⁺ into Pd/Pt NPs. Their intrinsic bioactive compounds further enhance antibacterial properties. This technology holds great potential for the medical, healthcare, and textile industries, particularly for medical masks, hospital garments, protective clothing, and antimicrobial functional textiles¹⁶. Overall, these results highlight that the combination of plant-based extracts and metallic NPs offers an eco-friendly and effective strategy for providing an antibacterial textile.

Additionally, the Pd and Pt content in the SPE/SSE/Pd/Pt NPs sample was determined using XRF after 5 and 10 washing cycles. As presented in Table 5, the initial amounts of deposited Pd and Pt on the cotton fabric were approximately 442 and 436 ppm, respectively. After 10 washing cycles, the remaining Pd and Pt contents in the fabric were 384 and 379 ppm, respectively. These results indicate that the loss of Pd/Pt due to washing was minimal, confirming the strong adhesion and stability of the NPs on the cotton fabric. The high retention of Pd/Pt NPs suggests their successful binding to the textile fibers through the phyto-synthesis process using SPE/SSE. The biosynthesized Pd/Pt NPs on textiles demonstrate excellent safety and stability, as the measured release concentrations remain well below established human toxicity thresholds. This green synthesis approach utilizing SPE/SSE not only ensures minimal environmental impact but also maintains the functional benefits of the NPs, making it a sustainable and biocompatible textile modification strategy.

Table 5.

Retention of Pd and Pt NPs on cotton fabrics dyed with SPE/SSE after 5 and 10 washing cycles.

Washing cycles	Content (ppm)		Reduction of bacteria (%)
	Pd	Pt	E. coli		S. aureus
			Pd	Pt	Pd	Pt
–	442 ± 10	436 ± 10	99	99	99	99
5	427 ± 10	409 ± 10	95	91	92	88
10	384 ± 10	379 ± 10	86	80	81	76

Table 6 presents a comparative summary of various green synthesis methods for metallic NPs using different plant-based extracts. As seen, most reported approaches employ single metal systems such as Ag or Pd under relatively harsh conditions or without systematic optimization. In contrast, the present work utilizes saffron waste as a sustainable bioreductant for the in situ synthesis of bimetallic Pd/Pt NPs on cotton fabric under mild, environmentally friendly conditions. Furthermore, the integration of an ANN-GA enabled a data-driven optimization of experimental parameters, leading to reproducible results and superior dual functionality in terms of coloration and antibacterial activity. These distinctive features highlight the novelty and advantage of the present approach over previously reported green synthesis strategies.

Table 6.

Comparison of the present synthesis route with other green methods for metallic NPs fabrication.

No.	Biological source extract	Metal NPs	Substrate	Key findings	Refs.
1	R. tuberosa leaf	CuO	Cotton fabric	Bactericidal on cotton Photocatalytic dye degradation	53
2	Pistachio	Ag	Cotton fabric	RSM optimized Antibacterial Excellent durability Significant increase in K/S	54
3	Ash of burnt Seidlitzia rosmarinus	CuO	Polyester fabric	Antibacterial Photocatalytic & self-cleaning UV protection Improved color, wettability & mechanical properties	55
4	Root of Scutellaria baicalensis	Ag	Silk fabric	Spherical, uniform NPs Antibacterial Moderate antioxidant activity	56
5	Mulberry leaves	TiO2	Cotton fabric	UV protection UPF 195 Antibacterial activity Good colorfastness and durability	42
6	Coriandrum sativum	ZnO	Cotton/polyester fabrics	Better UV-blocking efficiency	57
7	Salvadora persica root	ZnO	Cotton fabrics	Antibacterial Anti-crease, anti-UV Durable to washing	58
8	Spirulina	Ag	Cotton fabric	RSM optimized Photocatalytic 100% Congo red degradation Antimicrobial	59
9	Neem leaves	Ag	Cotton fabric	Taguchi optimization Even natural dyeing Enhanced color fastness Thermal stability Antimicrobial activity	60
10	Saffron waste	Pd/Pt	Cotton fabric	Synthesis of bimetallic NPs Using mixed plant extracts ANN-GA optimized Antibacterial Excellent durability Significant increase in K/S	Present study

Conclusion

In this study, Pd/Pt NPs were synthesized using saffron waste, and the dyeing process of cotton fabrics was conducted through microwave-assisted technology, offering an environmentally friendly and sustainable approach. The ANN method for optimizing and predicting K/S values demonstrated higher accuracy than RSM, exhibiting a strong correlation with experimental data (R² = 0.99). Integrating ANN with GA optimization further enhanced prediction performance and outperformed conventional statistical models. Comprehensive characterization via DLS, UV–Visible, FTIR, FESEM with elemental analysis, and XRD confirmed the successful formation and uniform distribution of Pd/Pt NPs on cotton substrates. The modified fabrics exhibited remarkable antimicrobial performance, achieving up to approximately 99% reduction of both Gram-negative and Gram-positive bacterial strains. In addition, the treated textiles retained strong colorfastness under rubbing, light, and washing according to standard tests.

While these findings highlight considerable potential for practical applications, certain limitations of the current methodology should be acknowledged. The phyto-synthesis process, though environmentally friendly, may result in variations in nanoparticle size and morphology due to the natural variability of saffron waste composition. Moreover, further optimization of synthesis parameters is necessary to ensure consistent nanoparticle deposition. Future research should focus on scaling up this approach for industrial applications and exploring other agricultural wastes as alternative biobased resources. Expanding the use of advanced machine learning models beyond ANN could further enhance predictive precision and process optimization.

Overall, this study provides a sustainable and efficient alternative to conventional textile processing, leveraging agricultural waste for the production of antibacterial cotton fabrics. These findings underscore the potential of green nanotechnology to advance environmentally responsible textile manufacturing and contribute to the development of smart and functional fabrics.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

SPE

Saffron petal extract

SSE

Saffron stamen extract

SPE/SSE

A blended extract of saffron petals and stamen

NPs

Nanoparticles

Pd/Pt NPs

Palladium–platinum bimetallic nanoparticles

SPE/SSE/Pd NPs

Palladium nanoparticles synthesized using the blended extract of saffron petals and stamen

SPE/SSE/Pt NPs

Platinum nanoparticles synthesized using the blended extract of saffron petals and stamen

SPE/SSE/Pd/Pt NPs

Palladium–platinum bimetallic nanoparticles synthesized using the blended extract of saffron petals and stamen

ANN

Artificial neural network

GA

Genetic algorithm

Author contributions

Mousa Sadeghi-Kiakhani: Conceptualization, Supervision, Funding acquisition, Resources, Validation, Project administration, Data analysis, Review & editing. Mohammad-Mahdi Norouzi: Methodology, Experimental investigation, Data collection, Formal analysis, Visualization, Writing – original draft, Project management. Amir Hossein Ramezani: Data optimization, Artificial intelligence analysis, Writing – review & editing. Elaheh Hashemi: Writing – review & editing, Investigation.

Funding

This research has been financially supported by the Saffron Institute, university of Torbat Heydarieh. The grant number was 160979.

Data availability

No data were used for the research described in this article.

Declarations

Competing interests

The authors declare no competing interests.