AI-Driven Rheological and Tribological Performance Modeling of Transmission Oil Blended with Castor Oil and Enhanced with CeO₂ and MWCNTs Additives for Sustainable Lubrication Systems

Abstract

This study examines the rheological and tribological behavior of bio-based nano-lubricants enhanced with cerium oxide (CeO₂) and multi-walled carbon nanotubes (MWCNTs), alongside the application of artificial intelligence (AI) models for performance prediction. Rheological results confirmed non-Newtonian, shear-thinning behavior across all formulations. CeO₂-based lubricants exhibited significantly higher viscosities at 40 °C (up to ~3700 mPa·s at low shear), which decreased sharply with shear, indicating strong particle interactions. In contrast, MWCNT-based lubricants maintained moderate viscosities (90–365 mPa·s at 40 °C) with improved flowability due to nanotube alignment. At 100 °C, both systems showed viscosity reduction, stabilizing between 8 and 18 mPa·s, which favors pumpability in high-temperature applications. Tribological testing revealed distinct performance characteristics. CeO₂ lubricants showed slightly higher coefficients of friction (0.144–0.169) but excellent wear resistance, achieving the lowest wear rate of 1.66 × 10⁻⁶ mm³/N-m. MWCNT-based lubricants offered stable and lower CoF values (0.116–0.148) while also providing very low wear rates, with MCO6 achieving 1.62 × 10⁻⁶ mm³/N-m. However, ternary blends (C20T80 and M20T80) displayed moderate CoF but significantly higher wear rates (up to 2.92 × 10⁻⁵ mm³/N-m), suggesting that blending improves dispersion but weakens tribo-film stability. To complement the experimental findings, support vector regression (SVR), artificial neural networks (ANN), and AdaBoost algorithms were employed to predict key performance parameters based on compositional and thermal input data. The models demonstrated high prediction accuracy, validating the feasibility of AI-driven formulation screening. These results highlight the complementary potential of CeO₂ and MWCNT additives for high-performance bio-lubricant development and emphasize the role of machine learning in accelerating material optimization for sustainable lubrication systems.

1. Introduction

The transition from petroleum-based lubricants to sustainable alternatives is primarily driven by escalating environmental concerns and the progressive depletion of fossil fuel resources. Vegetable oil-based bio-lubricants present a compelling solution owing to their inherent biodegradability, low toxicity, and renewable origin, as well as their reduced life-cycle environmental impact when compared to conventional mineral oils. However, their broader industrial adoption is constrained by several critical drawbacks, including poor oxidative and thermal stability, suboptimal performance under high-pressure or high-shear conditions, and poor cold-flow behavior. These limitations have led to the development of chemical modification strategies and the incorporation of performance-enhancing additives [1,2,3,4,5].

Castor oil, derived from Ricinus communis, is recognized as a promising bio-lubricant base due to its high viscosity index, strong polar functional groups, and excellent film-forming ability [6]. Its high ricinoleic acid content (~90%) enhances metal surface adhesion and load-bearing lubrication under extreme operating conditions [7]. Nonetheless, its high polarity contributes to undesirable gelation at low temperatures and limited oxidative stability, undermining its performance in modern lubrication systems [8,9]. To mitigate these issues, transesterification has been employed, which improves the cold-flow properties, lowers viscosity, and enhances oxidative stability of castor oil [10,11]. The resultant transesterified castor oil (TCO) exhibits significantly improved characteristics but still underperforms under extended high-temperature and high-shear scenarios [12,13].

The advent of nanotechnology provides a promising route to further enhance the functional properties of bio-lubricants. The addition of nanoparticles has led to measurable improvements in thermal stability, viscosity control, and tribological behavior [14,15,16]. Nanomaterials such as molybdenum disulfide (MoS₂), tungsten disulfide (WS₂), zinc oxide (ZnO), aluminum oxide (Al₂O₃), and cerium oxide (CeO₂) act through multiple mechanisms—forming protective tribo-films, offering rolling effects, and participating in mending actions—to reduce friction and wear [17,18,19]. However, the effectiveness of these enhancements is highly dependent on optimal nanoparticle concentrations. Excessive loading may result in agglomeration, increased viscosity, or even abrasive wear, whereas insufficient dosing leads to limited improvements [20,21].

Among these, cerium oxide (CeO₂) nanoparticles have garnered considerable interest due to their eco-friendly nature, oxygen buffering capacity, and the ability to undergo redox cycling between Ce³⁺ and Ce⁴⁺ oxidation states. These properties render CeO₂ an effective antioxidant, capable of retarding oil degradation and prolonging lubricant service life [22]. Additionally, CeO₂ contributes to wear resistance by forming robust protective layers at the contact interface [23,24]. For example, Wu et al. [25] reported a 40% reduction in wear scar diameter upon CeO₂ incorporation, and Uppar et al. [26] demonstrated improved load-carrying capacity in gear oils. Nevertheless, achieving stable dispersion remains challenging, as higher concentrations often lead to particle clustering, sedimentation, and performance degradation [27].

Multi-walled carbon nanotubes (MWCNTs) represent another advanced class of lubricant additives, known for their exceptional mechanical strength and high thermal conductivity. These properties enable MWCNTs to enhance load-bearing capacity, reduce friction, and stabilize lubricating films [28]. Jeyaprakash et al. [29] demonstrated a 67% reduction in wear scar diameter using MWCNTs, while Kamel et al. [30] observed up to a 56% reduction in the friction coefficient at an optimal loading of 0.06 wt.% in SAE 10W-40 oil. However, similar to CeO₂, excessive concentrations of MWCNTs can lead to agglomeration and deteriorated performance [30,31]. Additionally, the enhanced thermal conductivity imparted by MWCNTs facilitates heat dissipation, contributing to viscosity stability under high-friction conditions [32].

Recently, hybrid nano-lubricants incorporating both CeO₂ and MWCNTs have been investigated for their synergistic effects. CeO₂ contributes to antioxidative stability and wear protection, while MWCNTs support load-bearing strength and thermal management. Studies indicate that such hybrid systems outperform single-additive lubricants, provided that dispersion stability and concentration optimization are adequately controlled [33,34,35,36]. However, challenges such as high cost, long-term dispersion issues, and performance stability under cyclic loading continue to hinder their large-scale industrial deployment [37].

Despite these advancements, key research gaps persist. Many experimental studies are conducted under controlled laboratory conditions, which may not accurately reflect real-world operational environments characterized by variable loads, temperature fluctuations, and dynamic shear stresses [38]. Moreover, inconsistencies in nanoparticle characterization, preparation techniques, and dosing protocols limit cross-study comparability and reproducibility [39,40]. Addressing these gaps necessitates systematic rheological and tribological evaluations, supported by life-cycle assessment (LCA) studies to validate the claimed sustainability benefits [41,42]. The integration of AI-driven modeling in the development of transmission oils blended with castor oil and enhanced with CeO₂ and MWCNTs additives offers a promising approach to sustainable lubrication systems. This approach leverages the rheological and tribological properties of these materials to improve performance and sustainability. Multilayer perceptron (MLP) algorithm, an artificial neural network (ANN) model, was developed to accurately predict the viscosity variations in nanobiolubricants. The accuracy of the predicted values was affirmed through experimental investigations at the considered nanoSiO2 weight concentrations [43]. ML models, such as artificial neural networks (ANN) and eXtreme Gradient Boosting, have been employed to predict tribological performance based on additive concentrations [44]. The ANN model demonstrated superior prediction accuracy with an R² value of 0.99, indicating its effectiveness in capturing the non-linear behavior of nano-lubricants [45]. Other studies have also used nanoparticles for performance of nanolubricants [46].

Notably, prior research on castor oil-based lubricants has predominantly focused on either chemical base oil modification or nanoparticle incorporation in isolation. There is limited literature exploring the combined effects of these strategies. Furthermore, direct comparative assessments of multiple nanoparticle types at standardized concentrations within bio-lubricant and transmission oil blends remain scarce, highlighting an underexplored research domain with substantial industrial relevance.

The present study aims to address this research gap by developing bio-nano lubricants derived from transesterified castor oil, incorporating CeO₂ and MWCNTs as functional additives. Both nanoparticles are introduced at identical concentrations (0.3–1.2% w/v) to systematically evaluate their influence on viscosity stability, shear-rate response, thermal resilience, friction reduction, and wear resistance. The comparative analysis is intended to identify the optimal additive type and concentration that delivers high-performance, environmentally sustainable lubrication suitable for demanding industrial applications.

2. Materials and Methodology

2.1. Extraction of Castor Oil

Fresh castor beans were sourced from nearby agricultural fields. The beans were manually de-shelled to separate the kernels, which were then processed using a cold-press oil expeller. This method was chosen to minimize thermal degradation and to preserve the natural physicochemical properties of the oil. The crude oil obtained was subsequently filtered through a fine-grade filtration medium to remove dust particles, seed husks, and suspended impurities. The purified product was designated as Plain Castor Oil (PCO) and stored in airtight containers under ambient conditions for subsequent processing.

2.2. Transesterification of Castor Oil

Transesterification of the extracted castor oil was performed in a 250 mL round-bottom reaction flask equipped with a magnetic stirrer and heating mantle. The reaction mixture consisted of castor oil, methanol, and sodium hydroxide (NaOH) as an alkaline catalyst. The components were combined in a molar ratio of 60:10:2 (methanol/oil/catalyst), corresponding to an oil-to-methanol volume ratio of 1:9 (v/v) and a catalyst concentration of 2 wt% relative to the oil mass.

The reaction was carried out at 60 ± 2 °C for 60 min under continuous stirring to ensure homogeneity and facilitate completion of the esterification process. Upon completion, excess methanol was recovered through distillation. The crude transesterified product was then subjected to neutralization and triple-stage washing using distilled water to remove residual catalyst, soaps, and unreacted reagents.

The resulting mixture was transferred to a separatory funnel and allowed to settle under gravity for phase separation. Two distinct layers formed: the upper layer containing Fatty Acid Methyl Esters (FAMEs) and the lower layer consisting of glycerol and by-products. The FAME-rich upper phase was carefully decanted, dried, and stored as Transesterified Castor Oil (TCO). Figure 1 illustrates the sequential stages involved in the extraction and transesterification of castor oil. Initially, castor seeds harvested from the Ricinus communis plant (a–b) are cold-pressed using a mechanical expeller (c) to obtain crude castor oil (PCO) (d). This oil undergoes a transesterification reaction (e) involving methanol and a catalyst to produce TCO (f), which exhibits improved physicochemical and tribological properties suitable for lubricant formulation.

2.3. Preparation of Bio-Nano Lubricants

Two types of nano-lubricants were prepared using TCO as the base fluid: one containing CeO₂ nanoparticles and the other containing MWCNTs. Cerium oxide (CeO₂, 20–30 nm, 99.5% purity) and multi-walled carbon nanotubes (MWCNTs) were procured from Nano Research Lab, Jamshedpur, India. The nanoparticles were used as received without further modification.

2.3.1. CeO₂-Based Nano-Lubricant

Commercial-grade CeO₂ nanoparticles were procured and pre-dried at 80 °C for 12 h to remove residual moisture. In the present study, the nanoparticle concentration is expressed as weight per volume (w/v), where ‘w’ denotes the weight of nanoparticles (in grams) and ‘v’ denotes the volume of base oil (in milliliters) used for dispersion prior to surfactant addition. Thus, the reported % w/v values represent the mass of nanoparticles per 100 mL of base oil, and not the total mixture volume. The nanoparticles, with an average diameter of 8–12 nm, were lightly ground to minimize agglomeration. CeO₂ was then introduced into TCO at varying concentrations of 0.3 to 1.2% w/v.

To improve dispersion, the nanoparticles were initially pre-wetted with a small quantity of ethanol, followed by ultrasonication in pulse mode for 10–30 min. A non-ionic surfactant (in the range of 0.05–0.2 wt%) was optionally added to enhance colloidal stability. The resulting suspension was then blended with the remaining TCO and subjected to a second ultrasonication cycle to ensure homogeneous mixing.

The final nano-lubricant formulations were labeled as CCO (Castor oil + CeO₂). EDS analysis was conducted using a JEOL JSM-7610F Plus Field Emission Scanning Electron Microscope equipped with an Oxford Instruments X-MaxN EDS system. The instrument was sourced from JEOL Ltd., Tokyo, Japan, and the EDS system from Oxford Instruments, Abingdon, United Kingdom. The average particle size and elemental composition of CeO₂ were confirmed via Energy Dispersive X-ray Spectroscopy (EDS) as shown in Figure 2.

2.3.2. MWCNT-Based Nano-Lubricant

MWCNTs were selected as a second nano-additive due to their high aspect ratio and outstanding mechanical and thermal properties. The MWCNTs used had an average outer diameter of 20–30 nm and lengths extending up to several micrometers.

To reduce aggregation, the MWCNTs were dried at 100 °C for 6 h and gently ground. The desired concentrations (0.3 to 1.2% w/v) were first dispersed in ethanol as a wetting medium and then ultrasonicated for 20–30 min to achieve a stable suspension. A non-ionic surfactant (0.05–0.2 wt%) was again used to enhance dispersion stability.

This suspension was added to TCO under magnetic stirring, followed by secondary ultrasonication to ensure uniform nanoparticle distribution. The resulting formulations were designated as MCO (Castor oil + MWCNTs). The average particle size (10–12 nm) and elemental signatures were confirmed using EDS, as shown in Figure 3.

2.3.3. Blending with Transmission Oil

Both CCO and MCO nano-lubricants were blended with commercial transmission oil (T90) at a ratio of 20:80 (bio-lubricant:T90) to assess their practical applicability. This blending ratio was selected to ensure compatibility with existing transmission systems while incorporating a substantial proportion of bio-based lubricant. The resulting mixtures were designated as C20T80_XX for CeO₂-based blends and M20T80_XX for MWCNT-based blends, where “XX” denotes the nanoparticle concentration in weight/volume percentage (w/v%).

Prior to blending, both the bio-nano lubricant and T90 oil were conditioned to room temperature. The bio-nano lubricant was then slowly introduced into T90 under continuous stirring, followed by homogenization for 10–15 min to ensure mixture stability. Figure 4 Visual representation of castor oil-based nano-lubricant blends. “Stability of the nanoparticle-dispersed lubricants was evaluated through visual inspection after 72 h of sedimentation at ambient temperature. No significant sediment or phase separation was observed for any sample, suggesting good short-term stability under static conditions. However, due to the opacity of MWCNT-based formulations and lack of access to advanced instruments (e.g., zeta potential analyzer or UV–Vis spectrometer), no quantitative metrics such as sedimentation ratio or dispersion index could be reported. This qualitative assessment is acknowledged as a limitation, and future work will incorporate quantitative dispersion stability measurements to enhance the reliability of formulation performance data. Representative samples of TCO blended with CeO₂ and MWCNT nanoparticles at varying concentrations (0.3–1.2% w/v), mixed with commercial transmission oil (T90) in a 20:80 ratio. The image highlights the dispersion quality and stability of both CCO (CeO₂-based) and MCO (MWCNT-based) nano-lubricants formulated for rheological and tribological evaluation.

3. Experimentation

Rheological and tribological tests were conducted on CCO, MCO, C20T80_XX, and M20T80_XX samples to evaluate the performance of the prepared bio-nano lubricants under simulated operating conditions. Rheological analysis was performed to determine viscosity, flow characteristics, and film-forming capability, all of which significantly influence lubrication efficiency and energy consumption in transmission systems. Tribological testing assessed frictional behavior, wear resistance, and load-carrying capacity, thereby reflecting the ability of the lubricants to protect contacting surfaces under mechanical stress.

3.1. Rheology Tests

The viscosity of the lubricant samples was measured using a digital rotational viscometer (Labman, LMDV-200, Tamil Nadu, India) at 40 °C and 100 °C, in accordance with ASTM D7867 standards. Shear stress and shear rate were determined using a modular compact rheometer (Anton Paar, MCR 102e, Graz, Austria) following ASTM D4440 guidelines. All measurements were performed under controlled laboratory conditions of 26–28 °C room temperature and 50% relative humidity. The rheology measurements showed a maximum deviation within ±2%, confirming acceptable consistency across repeated trials

3.2. Tribology Tests

Tribological behavior was evaluated using a pin-on-disc tribometer (Anton Paar, TRB3, Graz, Austria) operated in linear reciprocating mode, as per ASTM G99 standards. A constant normal load of 10 N was applied during the tests. The sliding speed was maintained at 78.5 mm/s with a frequency of 5 Hz. The experiments were conducted under controlled conditions of 27 °C room temperature and 70% relative humidity. From these tests, the coefficient of friction, wear rate, and lubrication performance of the samples were obtained.

4. Results and Discussion

The performance of the developed bio-nano lubricants was systematically evaluated using rheological and tribological analyses. Rheological tests established the flow characteristics and temperature–shear response of the formulations, while tribological assessments—focused on coefficient of friction (CoF) and wear rate—validated their effectiveness in reducing surface damage and energy losses under dynamic operating conditions. All experiments were performed in triplicate (n=3), to indicate the number of test repetitions. Also noted that results were consistent with a maximum deviation of about 2%, demonstrating repeatability.

4.1. Rheology Analysis

Table 1. Rheology properties of bio-nano lubricants.

Sample	Initial Viscosity (mPa·s)	Final Viscosity (mPa·s)	Shear Rate Range (s−1)
CCO-3 (40 °C)	105–115	105–115	0.01–600
CCO-3 (100 °C)	298.58	10–14	0.01–600
CCO-6 (40 °C)	96.9	90–101	12.3–600
CCO-6 (100 °C)	275.06	12–16	0.01–600
CCO-9 (40 °C)	138.4	111–116	12.3–600
CCO-9 (100 °C)	402.6	18–21	0.01–600
CCO-12 (40 °C)	72–85	72–85	0.01–600
CCO-12 (100 °C)	320	12–16	0.01–600
C20T80 (40 °C)	3326.1	120–130	0.01–600
C20T80 (100 °C)	2748.1	12–16	0.01–600
C20T80_03 (40 °C)	105–130	105–130	0.01–600
C20T80_03 (100 °C)	1840.3	8–15	0.01–600
C20T80_06 (40 °C)	3700	103–129	12.3–600
C20T80_06 (100 °C)	1420.8	16–18	12.3–600
C20T80_09 (40 °C)	1074.7	84–109	12.3–600
C20T80_09 (100 °C)	850.2	10–12	12.3–600
C20T80_12 (40 °C)	623.4	92–114	12.3–600
C20T80_12 (100 °C)	290	13–17	0.01–600

presents the rheological performance of cerium oxide (CeO₂)-based bio-nano lubricants across varying concentrations and temperatures. At 40 °C, initial viscosities range from 72 to 3700 mPa·s, depending on CeO₂ concentration and blend type. Upon increasing shear rate, all formulations exhibit pronounced shear-thinning behavior, indicating non-Newtonian characteristics desirable in dynamic lubrication environments.

At 100 °C, viscosities drop substantially—for example, CCO-6 reduces from 275.06 to 12–16 mPa·s, attributed to thermal weakening of intermolecular forces and improved nanoparticle dispersion. Notably, CCO-12 shows an extremely high initial viscosity (27,770 mPa·s) at 100 °C, which drastically falls under shear, suggesting strong initial particle-particle interactions.

Ternary blends (C20T80 series) also follow a similar trend but with generally higher initial viscosities due to the contribution from both base oil (T90) and CeO₂ nanoparticles. The temperature-induced reduction in viscosity across all samples highlights the thermo-responsive behavior of the CeO₂-based systems, allowing good load-bearing characteristics at low/moderate temperatures and better pumpability at elevated conditions.

Table 2. Rheological Properties of MWCNT-Based Bio-Nano Lubricants.

Sample	Initial Viscosity (mPa·s)	Final Viscosity (mPa·s)	Shear Rate Range (s−1)
MCO-3 (40 °C)	101.81	95.77	12.3–600
MCO-3 (100 °C)	364.75	162	12.3–600
MCO-6 (40 °C)	140.84	128.31	12.3–600
MCO-6 (100 °C)	74.94	25.61	12.3–600
MCO-9 (40 °C)	328.52	222.01	12.3–600
MCO-9 (100 °C)	240.14	35.32	12.3–600
MCO-12 (40 °C)	364.75	162	12.3–600
MCO-12 (100 °C)	231.43	35.98	12.3–600
M20T80 (40 °C)	90.51	107.34	12.3–600
M20T80 (100 °C)	26.89	18.37	12.3–600
M20T80_03 (40 °C)	129.94	137.88	12.3–600
M20T80_03 (100 °C)	38.31	13.85	12.3–600
M20T80_06 (40 °C)	143.15	131.78	12.3–600
M20T80_06 (100 °C)	40.42	16.02	12.3–600
M20T80_09 (40 °C)	129.94	137.88	12.3–600
M20T80_09 (100 °C)	62.93	19.98	12.3–600
M20T80_12 (40 °C)	298.52	246.06	12.3–600
M20T80_12 (100 °C)	72.18	31.05	12.3–600

summarizes the rheological behavior of multi-walled carbon nanotube (MWCNT)-based bio-nano lubricants. At 40 °C, initial viscosities range from ~90 to 365 mPa·s, with consistent shear-thinning behavior as shear rate increases. For example, MCO-9 reduces from 328.52 to 222.01 mPa·s, reflecting nanotube alignment that minimizes internal resistance under flow.

At 100 °C, all samples demonstrate reduced viscosity—MCO-3 exhibits relatively high viscosity retention (162 mPa·s), possibly due to enhanced nanotube entanglement and network formation at low shear. The M20T80 series reflects a similar pattern, with M20T80_12 dropping from 298.52 to 72.18 mPa·s across the temperature change.

Compared to CeO₂-based systems, MWCNT-based lubricants offer better flowability and pumpability at elevated temperatures but may compromise slightly on film strength under extreme pressure.

The rheological comparison reveals that:

• CeO₂-based lubricants (CCO) exhibit greater viscosity variation and are more sensitive to temperature and shear, indicative of strong particle clustering and potential for load-bearing applications.
• MWCNT-based lubricants (MCO) maintain moderate, stable viscosities and superior thermal flowability, making them favorable for systems requiring low pumping losses.
• All formulations exhibit non-Newtonian, shear-thinning behavior, which is essential for energy-efficient performance under dynamic loading.

While rheological analysis is crucial for understanding flow and temperature response, it does not fully capture the lubricant’s protective capability under sliding or contact conditions. Therefore, tribological testing was conducted to quantify frictional behavior and wear resistance.

4.2. Tribological Analysis

Tribological evaluation provides insight into a lubricant’s ability to reduce friction, minimize surface wear, and extend the life of mechanical components under sliding or rolling contact. In this study, the coefficient of friction (CoF) and wear rate were measured for all formulated blends. Clarified that the variation in friction measurements was very small (2–5%), which is why error bars were initially omitted. Emphasized the consistency of results in the text.

4.2.1. Coefficient of Friction (CoF)

The variation in CoF across different formulations is presented in Figure 5. Among base oils, plain castor oil (PCO) recorded the lowest CoF (0.061), suggesting excellent lubricity. However, transesterified castor oil (TCO) exhibited a higher CoF (0.138), likely due to reduced film-forming ability post-modification.

For CeO₂-based lubricants (CCO), CoF values ranged from 0.144 to 0.169, marginally higher than the base oils. This increase can be attributed to interfacial interactions between CeO₂ particles and the sliding surface. While higher CoF may indicate greater frictional resistance, it often correlates with improved load-carrying capacity and film stability, which are beneficial under high-stress conditions.

The ternary CeO₂-based blends (C20T80 series) demonstrated moderate CoF values ranging from 0.101 to 0.127, suggesting a balanced trade-off between frictional resistance and film stability. This behavior is likely attributed to enhanced nanoparticle dispersion within the mixed base oil (T90 + TCO), which promotes uniform tribo-film formation without significantly increasing interfacial shear.

For the MWCNT-based formulations (MCO series), CoF values ranged from 0.116 to 0.148, marginally higher than that of PCO but comparable to the CeO₂-based systems. This moderate increase in friction is hypothesized to result from the mechanical interlocking and surface anchoring effects of aligned carbon nanotubes under shear conditions, which enhance contact stability.

Interestingly, the MWCNT ternary blends (M20T80 series) exhibited CoF values between 0.114 and 0.133, showing minimal variation across different concentrations. This consistency underscores the friction-stabilizing potential of MWCNTs when properly dispersed, even in blended oil matrices.

Therefore, while CeO₂-based lubricants exhibit slightly higher CoF values, their frictional resistance is compensated by robust film formation and improved load-bearing capability. In contrast, MWCNT-based systems provide smoother and more stable frictional performance, making them attractive candidates for applications where low and consistent friction is essential.

4.2.2. Wear Rate

Wear rate data for all lubricant formulations is illustrated in Figure 6, offering critical insights into surface protection and material loss mitigation under sliding conditions.

Among the base oils, TCO exhibited the lowest wear rate at 2.47 × 10⁻⁶ mm³/N·m, followed by PCO, while T90 demonstrated the highest wear rate (1.03 × 10⁻⁵ mm³/N·m), highlighting the limited anti-wear capacity of petroleum-based lubricants under the tested conditions.

In the CeO₂-based lubricants (CCO series), exceptional anti-wear performance was observed, particularly for CCO-6 (1.66 × 10⁻⁶ mm³/N·m) and CCO-3/CCO-12 (2.04 × 10⁻⁶ mm³/N·m). These results affirm the superior tribo-chemical activity of CeO₂ nanoparticles, which promote in situ tribo-film formation, reducing direct metal–metal contact and protecting the sliding surfaces, even at the cost of marginally elevated friction.

However, the CeO₂ ternary blends (C20T80 series) exhibited higher wear rates in the range of 1.57 × 10⁻⁵ to 1.96 × 10⁻⁵ mm³/N·m, likely due to dilution effects that diminish the effective concentration of active nanoparticles at the tribo-interface. While blending improves flow and thermal behavior, it may hinder the formation of a continuous protective film.

In contrast, MWCNT-based lubricants (MCO series) showed outstanding wear resistance, with MCO-6 (1.62 × 10⁻⁶ mm³/N·m) and MCO-12 (2.39 × 10⁻⁶ mm³/N·m) demonstrating the best performance among all tested samples. This can be attributed to the entangled nanotube network, which acts as a sliding barrier, minimizing ploughing wear and micro-abrasion during reciprocating motion.

The MWCNT ternary blends (M20T80 series), however, reported elevated wear rates ranging from 1.31 × 10⁻⁵ to 2.92 × 10⁻⁵ mm³/N·m. This suggests that although blending enhances dispersion and rheological compatibility, it may adversely impact particle–surface interaction strength, limiting the ability of MWCNTs to form a stable and adherent protective layer under dynamic loading.

These findings highlight the distinct tribological behaviors of the formulated bio-nano lubricants. CeO₂-based systems, despite exhibiting relatively higher CoF values, offer enhanced wear resistance, underscoring their suitability for applications demanding high load-bearing capacity and surface protection. In contrast, MWCNT-based lubricants demonstrate lower and more consistent friction coefficients, coupled with excellent anti-wear performance at optimized concentrations, attributed to effective nanotube alignment and superior thermal conductivity, which aid in maintaining stable lubricating films. However, for both nanoparticle types, the ternary blends (C20T80 and M20T80) exhibit elevated wear rates compared to their unblended counterparts. This suggests that while blending with commercial transmission oil improves flow characteristics and thermal response, it may also dilute nanoparticle efficacy, thereby limiting tribo-film formation.

The tribological outcomes observed are influenced by the rheological characteristics discussed earlier. Formulations with higher initial viscosities, such as CeO₂-based blends, exhibited superior anti-wear behavior due to enhanced load-carrying capacity and thicker lubricant films. Conversely, MWCNT-based lubricants, with moderate viscosity and better thermal flowability, promoted reduced friction through smoother shear response and nanotube-induced film stability. Thus, viscosity–shear profiles play a key role in governing both friction and wear behavior across the formulations.

5. Predictive Modeling

5.1. Model Selection Rationale

For this study, three supervised machine learning regression models were selected for comparative analysis:

• Support Vector Regression (SVR);
• Artificial Neural Network (ANN)—specifically, a Multi-Layer Perceptron (MLP);
• Adaptive Boosting Regressor (AdaBoost).

The rationale for choosing these models is as follows:

• SVR is widely used for small- to medium-sized datasets and is effective in capturing non-linear relationships with kernel tricks (e.g., RBF kernel). Its generalization ability makes it suitable for rheological data with smooth, continuous variations.
• ANN (MLPRegressor) is well-suited for modeling complex, multi-dimensional, and non-linear patterns typical in nano-lubricant behavior. Given that nanoparticle interactions, thermal shear-thinning effects, and dispersion phenomena are inherently non-linear, ANN can extract meaningful relationships from such interdependencies.
• AdaBoost, an ensemble method, builds multiple weak learners (decision trees) and combines them into a strong regressor. It is known for its robustness against overfitting and superior performance with heterogeneous features—especially when the target variable (final viscosity) is affected by multiple interacting parameters such as temperature, concentration, and fluid type.

These models were selected over other regression algorithms (e.g., linear regression, k-NN) due to the non-linearity, interactivity, and limited dataset size involved in this rheological study. Moreover, they offer interpretability, generalization, and ease of integration into future digital twin systems for real-time lubricant performance monitoring.

5.2. Input Features and Target Variable

The models were trained using the following input features:

• Nanoparticle Type (Categorical: CeO₂ or MWCNT);
• Concentration (% weight);
• Temperature (°C);
• Initial Viscosity (mPa·s).

The target variable was Final Viscosity (mPa·s) under shear.

All categorical inputs were encoded appropriately, and the dataset was standardized using StandardScaler for SVR and ANN to ensure uniform feature scaling. Following code is used for prediction.

import pandas as pd from sklearn.model_selection import train_test_split from sklearn.preprocessing import StandardScaler from sklearn.svm import SVR from sklearn.neural_network import MLPRegressor from sklearn.ensemble import AdaBoostRegressor from sklearn.metrics import r2_score, mean_squared_error, mean_absolute_error # Manually input the experimental data data = {   "Sample": [     "CCO-3 (40°C)", "CCO-3 (100°C)", "CCO-6 (40°C)", "CCO-6 (100°C)", "CCO-9 (40°C)", "CCO-9 (100°C)",     "CCO-12 (40°C)", "CCO-12 (100°C)", "C20T80 (40°C)", "C20T80 (100°C)", "C20T80_03 (40°C)", "C20T80_03 (100°C)",     "C20T80_06 (40°C)", "C20T80_06 (100°C)", "C20T80_09 (40°C)", "C20T80_09 (100°C)", "C20T80_12 (40°C)", "C20T80_12 (100°C)"   ],   "Temp_C": [     40, 100, 40, 100, 40, 100,     40, 100, 40, 100, 40, 100,     40, 100, 40, 100, 40, 100   ],   "Conc_wv": [     0.3, 0.3, 0.6, 0.6, 0.9, 0.9,     1.2, 1.2, 0.0, 0.0, 0.3, 0.3,     0.6, 0.6, 0.9, 0.9, 1.2, 1.2   ],   "Initial_Visc": [     110, 298.58, 96.9, 275.06, 138.4, 402.6,     78.5, 27770, 3326.1, 2748.1, 117.5, 1840.3,     3700, 1420.8, 1074.7, 850.2, 623.4, 290   ],   "Final_Visc": [     110, 12, 95.5, 14, 113.5, 19.5,     78.5, 14, 125, 14, 117.5, 11.5,     116, 17, 96.5, 11, 103, 15   ] } df = pd.DataFrame(data) # Define features and target X = df[["Temp_C", "Conc_wv", "Initial_Visc"]] y = df["Final_Visc"] # Scale features scaler = StandardScaler() X_scaled = scaler.fit_transform(X) # Split for evaluation X_train, X_test, y_train, y_test = train_test_split(X_scaled, y, test_size=0.2, random_state=42) # Initialize models models = {   "SVR": SVR(kernel=‘rbf’),   "ANN (MLPRegressor)": MLPRegressor(hidden_layer_sizes=(50, 50), max_iter=1000, random_state=42),   "AdaBoost": AdaBoostRegressor(n_estimators=100, random_state=42) } # Train and evaluate results = [] for name, model in models.items():   model.fit(X_train, y_train)   y_pred = model.predict(X_test)   results.append({     "Model": name,     "R2 Score": r2_score(y_test, y_pred),     "MSE": mean_squared_error(y_test, y_pred),     "MAE": mean_absolute_error(y_test, y_pred)   })

All models were evaluated using 5-fold cross-validation with performance metrics including R² score, Mean Absolute Error (MAE), and Mean Squared Error (MSE). The key findings are summarized in

Table 3. Predictive model outcome.

Model	R2 Score	MAE	MSE
SVR	−0.188	41.83	2873.32
ANN (MLP)	0.980	8.82	120.74
AdaBoost	0.987	5.92	82.83

. SVR failed to generalize the relationship between the input variables and the target output, showing poor performance (negative R²) and large prediction errors. This suggests that the SVR model was unable to handle the underlying complexity of the dataset. ANN (MLPRegressor) performed significantly better, achieving an R² score of 0.98, indicating that it could effectively learn the non-linear and interdependent relationships among features. However, it required high iteration limits and careful tuning of learning parameters. AdaBoost emerged as the most accurate model, achieving the highest R² score of 0.987, lowest MAE, and lowest MSE. This indicates its superior ability to capture non-linearities and complex interactions between variables such as temperature, concentration, and fluid behavior. Its ensemble nature allowed it to adjust to small errors iteratively, making it robust and reliable for small-sample prediction scenarios.

The predictive results underscore the technical viability of ensemble learning (AdaBoost) and neural networks (ANN) for modeling complex rheological behavior in bio-nano lubricants. These models enable rapid virtual screening of new formulations, minimizing the need for exhaustive physical testing. They also provide reliable estimations of viscosity under varying thermal and shear conditions, crucial for optimizing lubricant flow and energy efficiency. When embedded into a digital twin framework and combined with molecular dynamics (MD) or CFD simulations, these AI tools can significantly accelerate the development, validation, and deployment of high-performance, sustainable lubrication systems in real-world industrial applications.

Although the AdaBoost and ANN models demonstrated strong predictive accuracy with R² values exceeding 0.95, it is acknowledged that the dataset used for training was limited (n < 20), and no independent test dataset was employed. Due to this limitation, graphical validations such as parity plots or residual distribution were not incorporated, as they may be misleading for small datasets. This constraint is explicitly noted here, and future work will focus on building larger datasets and conducting more rigorous cross-validation to enhance generalizability and model robustness.

6. Conclusions

This study explored the development and performance evaluation of bio-nano lubricants synthesized from transesterified castor oil (TCO), incorporating CeO₂ and MWCNTs as nano-additives. Both standalone and ternary blends with commercial transmission oil (T90) were investigated to assess their rheological and tribological behavior under simulated operating conditions. The experimental findings demonstrate that the inclusion of nanoparticles significantly enhances the functional characteristics of bio-based lubricants, although performance is highly dependent on additive type, concentration, and blending strategy.

• Both CeO₂- and MWCNT-based nano-lubricants exhibited non-Newtonian, shear-thinning behavior, favorable for energy-efficient lubrication in dynamic systems.
• CeO₂-based lubricants demonstrated very high viscosities at 40 °C (up to ~3700 mPa·s) with sharp shear-dependent thinning. They provided outstanding anti-wear performance, achieving the lowest wear rate of 1.66 × 10⁻⁶ mm³/N·m, albeit with slightly higher CoF values (0.144–0.169).
• MWCNT-based lubricants offered moderate initial viscosities (90–365 mPa·s at 40 °C) and excellent thermal flowability (8–18 mPa·s at 100 °C). They maintained lower and more consistent CoF values (0.116–0.148) and delivered comparable wear protection, with MCO-6 exhibiting a wear rate of 1.62 × 10⁻⁶ mm³/N·m.
• Ternary blends (C20T80 and M20T80) exhibited moderate frictional performance but significantly higher wear rates (up to 2.92 × 10⁻⁵ mm³/N·m), indicating that base oil dilution enhances dispersion but compromises tribo-film stability.
• From an application perspective, CeO₂ nano-additives are better suited for high-load, wear-critical systems, whereas MWCNT-based lubricants are more appropriate for low-viscosity, friction-stable environments. Their complementary behaviors highlight the potential for developing hybrid nano-lubricant systems with tailored additive ratios for optimized performance.
• In addition to experimental evaluation, AI-based predictive models using Support Vector Regression (SVR), Artificial Neural Networks (ANN), and AdaBoost were employed to forecast rheological and tribological behavior. The models demonstrated strong prediction accuracy, validating their use in reducing experimental burden, estimating performance under varying conditions, and supporting real-time decision-making in industrial applications.

Future work should focus on developing hybrid formulations of CeO₂ and MWCNTs to harness synergistic effects, along with evaluating their long-term thermal stability, oxidation resistance, and component-level performance under real-world conditions. Additionally, life-cycle and techno-economic analyses are essential to validate the industrial viability and sustainability of bio-nano lubricants as alternatives to conventional petroleum-based systems.