Multi-objective optimization of HDPE pyrolysis oil–diesel blends across variable compression ratios: an integrated experimental and ensemble learning approach

1 Introduction

The production and use of plastics have increased globally to a level that waste management systems were not designed to handle. To understand the contemporary relevance of waste plastic valorization, it is imperative to examine the most recent global and regional production statistics. Global plastic production has grown rapidly, reaching an estimated 460 million metric tons. According to da Costa and Desvaux et al. (2026), global plastic production increased to 460 million metric tons in 2019, simultaneously generating approximately 350 million metric tons of plastic waste annually. A 2024 study projects that global plastic use will almost double, reaching between 735 and 884 million metric tons by the year 2050 based on regression analysis of historical trends by Dokl et al. (2024). Another forecast finds that if current consumption trajectories persist, global plastic demand is projected to nearly double, reaching approximately 800 million metric tons by 2050, as depicted in the forecast trends in Figure 1a (Saxena, 2025).

Within the Indian context, Singh and Paul (2024) found that India generates an estimated 9.4 million metric tons of post-consumer polymer refuse each year; despite achieving a 60% recovery rate, more than 9,400 tons escape municipal management daily to accumulate in landfills and aquatic systems. Another study indicates that the plastic industry has experienced remarkable growth and is forecast to reach a processing capacity of 35 million metric tons per annum (MTPA) by 2030 (Gurjar et al., 2026). Compounding this challenge, plastic packaging contributes to almost 46% of the overall plastic waste in the country, and India is projected to become the 5th highest nation globally for inadequate handling of plastic waste by 2025 (Gurjar et al., 2026). Primary drivers of this demand are the packaging, tires, construction, vehicles, clothing, electrical/electronics, and textile sectors globally. Figure 1b indicates the breakdown of the Gulf Cooperation Council (GCC) polymer share. The Gulf Petrochemicals and Chemicals Association (2023) projected a compound annual growth rate (CAGR) of 2.4% and a total relative change of 168% between 2019 and 2060 for the total use of polymers.

This escalating crisis highlights the urgent need for sustainable, localized valorization strategies. Thermochemical conversion techniques such as pyrolysis offer a promising dual-benefit solution: effectively mitigating the regional plastic waste burden while transforming high-density polyethylene (HDPE) waste into a viable alternative fuel for internal combustion engines (Saxena, 2025). The production of plastic in the international trade network is increasing day by day (Houssini et al., 2025). At the same time, fossil fuels, which are limited and unevenly distributed, are used extensively in transportation and agriculture. There is a logical intersection between the two problems of plastic waste disposal and fuel insecurity. Researchers have been working on this intersection for more than two decades.

As India has the largest population, plastic consumption forms the country’s manufacturing and consumption economy. With economic expansion, plastic consumption has increased significantly (Emami et al., 2024). In contrast, the country’s recycling and formal waste processing infrastructure is not up to mark and has no recovery route for post-consumer plastics (Ministry of Chemicals and Fertilizers, 2022). From an engineering aspect, it is better to find a use for plastic waste than to manage its disposal. Pyrolysis provides one of the best solutions. In the pyrolysis process, waste plastic is burned without oxygen at approximately 300 °C–700 °C. Heat is used to break polymer chains into smaller hydrocarbons that can be used as liquid fuel (Çelik et al., 2017; Sharuddin et al., 2016). Previous studies have mentioned the feasibility of the thermolysis process to convert waste plastic into liquid fuel (Panda et al., 2010) and utilizing that fuel as an energy resource to solve the disposal problem (Tan et al., 2024; Wong et al., 2015).

Catalysts play a vital role in producing cleaner and more consistent output of plastic pyrolysis oil (Miandad et al., 2019). Kumar and Singh (2011) observed that the pyrolytic liquid fuel derived from the thermal cracking of HDPE waste exhibits physicochemical characteristics that closely align with conventional diesel specifications. Optimized cracking of waste PP produces a type of molecule similar to conventional diesel (Panda and Singh, 2014), and under controlled conditions, even PET-based feedstock yields pyrolysis oil amenable to engine use (Çelik et al., 2017). The waste plastic oil (WPO) has a calorific value in the 40–45 MJ/kg range, close enough to conventional diesel. It can be blended and used in compression-ignition (CI) engines without any hardware modification.

Properties of the WPO (viscosity, density, and flash point) vary with feedstock composition and pyrolysis conditions (Pratoomyod and Laohalidanond, 2013). It is a practical challenge to ascertain which type of plastic waste is the most suitable for producing a pyrolysis oil that is similar to diesel. Many studies show that oils from PP and PE are more suitable for the engine test (Mangesh et al., 2020a). Hydrogenation improves crude pyrolysis oil by improving its thermal stability and aromatic content, yielding clean and stable fuel (Mangesh et al., 2020b). Catalytic pyrolysis of polypropylene (PP) produces fuel with the same characteristics as diesel (Panda et al., 2016).

Mani et al. (2009) found the emissions of exhaust CO, CO₂, and unburned hydrocarbons (UHCs) from using WPO in a diesel engine to be slightly high and produced 40% less smoke. Mani et al. (2011) characterized WPO, indicating moderate use to 20%, and stated that there is a direct correlation between WPO blend ratio and exhaust emissions. Mani and Nagarajan (2009) demonstrated that delaying the injection timing to 14° before top dead center (BTDC) successfully lowered emissions of NO_*x*, CO, and hydrocarbons (HC). Furthermore, many researchers identify that when using WPO, emissions are reduced with exhaust gas recirculation techniques; however, optimizing the blend ratio is important to address inherent trade-offs in thermal efficiency (Kumar et al., 2013; Mani et al., 2011).

Rajendran et al. (2023) and Palanivelrajan et al. (2024) suggest that blend ratio variation plays an important role in the utilization of WPO in diesel engines (Damodharan et al., 2019; Faisal et al., 2023). Pure plastic oil used in diesel engines increases smoke opacity as well as unburned hydrocarbons, whereas 20%–80% blends with diesel provide performance closer to the diesel baseline (Singh et al., 2020; Jaat et al., 2021). Many studies have explored using oxygenated additives at higher concentrations of WPO and found they increase the smoke and CO. Using diethyl ether (DEE) enhances fuel atomization and ensures more thorough combustion (Devaraj et al., 2015). It also yields a notable drop in hydrocarbon and smoke emissions while maintaining similar thermal efficiency (Selvam et al., 2022). Some researchers tested blending biodiesel with WPO to improve lubricity and decrease smoke (Kumar and Sankaranarayanan, 2016; Senthilkumar & Sivakumar, 2014). Another study found that at lower load, HC and CO increase as WPO blend percentages increase, suggesting an optimum blend ratio to control the emissions (Mustayen et al., 2023; Sunaryo et al., 2021). Higher injection pressure promotes atomization and air-fuel mixing, which leads to an increase in efficiency and decreases exhaust emissions in WPO blend ratios (Heywood, 1988; Stone, 2012).

In a study that experimented with a variable compression ratio engine test setup, the researchers found that as the compression ratio increases, so does the in-cylinder pressure and temperature. Higher compression ratios reduce ignition delay and generally improve combustion completeness for WPO fuels with slightly different ignition characteristics than conventional diesel (Chintala et al., 2018; Kumar et al., 2013). In another study, compression ratios of 17 and 18 and P90D5E5 blends were used in diesel engines by Govinda Rao et al. (2018). According to their results, pure plastic oil at 18 CR 21° BTDC gives 38% higher brake thermal efficiency, while CO and smoke decrease, and NO_*x* increases. Such a parametric investigation in VCR engines is feasible to get a systematic evaluation of compression ratio effects without hardware changes between the test engine runs (Viswanath and Vijayabalan, 2019). WPO fuel properties were examined by Wathakit et al. (2021) for their effects in a VCR. They claim that higher CR gives better BTE, and exhaust emissions are reduced for the WPO blend. Pyrolysis oil obtained from waste packaging plastic was studied by Alawa and Chakma (2022), who showed good engine performance and balanced emissions with CRs of 17 and 18.

No study in the existing literature has as yet combined a systematic full-factorial VCR matrix (4 CRs × 5 blends × 6 loads = 120 points) with multi-algorithm ML comparison (random forest (RF), XGBoost, ANN, support vector regression (SVR)) and NSGA-II multi-objective optimization specifically for HDPE-derived WPO–diesel blends. The specific research gap addressed here is therefore the absence of a unified experimental-cum-data-driven framework that simultaneously (i) maps the full four-factor operating space of CR, blend ratio, and load, (ii) benchmarks multiple machine learning surrogates against this experimental matrix, and (iii) couples the best-performing surrogate with NSGA-II to identify a Pareto-optimal operating condition. The original contribution of this study lies in this integrated methodology and in the resulting evidence-based optimal operating point (CR = 17, B40, full load), which is validated against the directly measured experimental data rather than the surrogate model alone (Section 5.5).

2 Fuel preparation and characterization

2.1 Pyrolysis feedstock and process

The fishbone diagram in Figure 2 shows why plastic recycling is a difficult process. The seven types of plastic differ in physical characteristics, potential contamination, and cost. Only PETE (30%) and HDPE (25%) achieved some meaningful recycling rates. Although PVC, PS, and mixed plastics are still largely unrecyclable because of sorting complexity, risk of microplastics, and low market value, other plastics can be recycled. In this study, HDPE has been selected primarily because of its high hydrogen-to-carbon ratio. It has potential for better liquid yield than other plastics such as PET and PVC. The polyolefin polymer in HDPE plastic contains ethylene and polypropylene monomers, in which the linear hydrogen-to-carbon ratio is high and easy to break down. Its conversion efficiency is very good for pyrolysis oil. Polyolefin-based HDPE produces cleaner and minimally toxic pyrolysis oil. Thermochemical conversion characteristics, widespread availability, and conversion of pyrolysis oil from HDPE having similar properties to diesel make HDPE waste the ideal selection for primary feedstock for the present investigation (Al salem, 2017; Miandad, 2016).

Generally, the pyrolysis process yields high-, medium-, and low-molecular weight carbon–hydrogen fragments. High MW fractions give solid residues known as carbon black and char, medium-MW HCs provide liquid plastic oil molecules with pentane (C₅H₁₂), hexane (C₆H₁₄), octane (C₈H₁₈), decane (C₁₀H₂₂), dodecane (C₁₂H₂₆), and hydrocarbons in the range of diesel (C10 to C25), and low-MW HCs give methane (CH₄), ethane (C₂H₆), ethylene (C₂H₄), propylene (C₃H₆), and hydrogen (H₂) gases. Amongst them, liquid plastic oil, with specific alkanes (paraffins and olefins) and aromatics, has similar characteristics that are useful for diesel blending.

First, the collected waste was segregated, cleaned, and dried for 12 h at 80 °C to remove dust, moisture, and impurities from the HDPE plastic waste. The cleaned HDPE plastic waste was cut into small pieces of 5–10 mm so that heat could be transferred uniformly during the pyrolysis process.

A laboratory-scale, batch-type stainless steel reactor system was made to convert HDPE waste into liquid fuel. It is equipped with a temperature control unit, condenser assembly, and gas outlet, as shown in Figure 3. HDPE waste was converted to pyrolysis oil using thermal conversion in an oxygen-deficient atmosphere using this reactor system. A known quantity of around 20 g was loaded into the reactor for heating to 400 °C–500 °C in complete absence of atmospheric oxygen. The pyrolysis experiments were carried out in a 2-L stainless steel fixed-bed batch reactor. Prior to each experiment, the reactor was purged with nitrogen gas (N₂, 99.9% purity) for 10–15 min to establish an inert atmosphere, and nitrogen flow was maintained continuously at 200 mL/min throughout the pyrolysis run to prevent oxidation. The reactor was heated at a 12 °C–15 °C/min rate governed by an Autonics TK4S PID controller. Temperature was maintained at 400 °C–500 °C throughout the pyrolysis process. Thermal consistency within the reactor was monitored using two K-type thermocouples positioned at the top and bottom of the reactor, with a maximum observed temperature variation of ±2 °C. A natural montmorillonite clay catalyst was added (5% weight relative to feedstock) in the reactor to obtain fuel with enhanced cracking efficiency and reduce the unwanted formation of polycyclic aromatic hydrocarbons. The pyrolysis vapors were condensed using a shell-and-tube condenser with chilled water circulating at 15 °C, and the resulting liquid product was collected in a graduated glass receiver. The non-condensable gases were vented through a fume hood. Liquid yield was recorded gravimetrically at the end of each run. This study consistently yielded 65–70 wt% mass of a liquid, 15–20 wt% gas, and 5–10 wt% residual solids. Similar data were found by previous researchers (Panda et al., 2010; Singh et al., 2020).

2.2 Physicochemical characterization

The analysis of blend WPO–diesel follows ASTM and hydrometric standards. The higher calorific value of WPO was found to be 40,950 kJ/kg, which is 2.5% less than the value of diesel at 42,000 kJ/kg. The marginally lower calorific value results from the broader hydrocarbon distribution inherent to the pyrolysis process, including lighter volatile fractions (C5–C10) with lower volumetric energy density and the presence of olefinic compounds whose C=C bonds release less combustion energy per carbon atom than equivalent paraffinic chains dominant in diesel fuel. This observation is consistent with published literature reporting similar trends, where mixed plastic pyrolysis oil showed a calorific value of 41,254 kJ/kg against diesel at 42,000 kJ/kg (Singh et al., 2019). While conventional diesel exhibited a standard calorific value of 43,800 kJ/kg, HDPE-derived catalytic oil yielded a highly competitive 40,170 kJ/kg (Kumar et al., 2013).

According to CHNOS elemental analysis data from Kumar et al. (2013), the ultimate analysis of catalytic HDPE WPO yields an elemental composition of 84.4% carbon, 12.7% hydrogen, 0.79% nitrogen, 0.12% sulfur, and 1.99% oxygen. The WPO contains less than 0.01% sulfur, which reduces the emissions of SO₂ and sulfate particles into the atmosphere. Moreover, the high level of purity of WPO allows it to work in conjunction with the existing exhaust gas treatment systems in place to control emissions. At temperatures above ambient, waste plastic oil contains more vapor than diesel fuel, due to its lower flash point (36 °C) than diesel, which is 52 °C. This higher vapor content of WPO makes it easier to initiate the combustion process when the engine is cold started. It also demands stricter precautions in storage and transfer. Its flow resistance at 40 °C measures 2.1 mm²/s, somewhat less than diesel’s 2.6 mm²/s, leading to subtly smaller fuel droplets when injected. Such refinement plays a role in how carbon monoxide levels shift under light engine load.

The physical and chemical characteristics of WPO were measured and compared with those of commercial diesel oil. The measured density of the WPO is 860 kg/m³, which is slightly higher than standard commercial diesel oil at 830 kg/m³ (per the ASTM D4052 standard). Based on bomb calorimeter tests, WPO has a heating value of 40,950 kJ/kg and is thus 2.5% lower than diesel oil, which has a heating value of 42,000 kJ/kg. Therefore, slightly more mass of WPO would be required to deliver the same quantity of energy.

The temperature of the ignition flash point was 36 °C (ASTM D93 test method), and the fire point was 41 °C (ASTM D92 test method). Even though it is lower than that of diesel fuel (52 °C flash point and 55 °C fire point), the chemical reactions that occur at ignition start to overcome stored risks rather quickly. Measured via the ASTM D445 method, the kinematic viscosity at 40 °C was 2.1 mm²/s. This value sits slightly below the 2.6 mm²/s typically observed in standard diesel, allowing the WPO to flow with slightly less resistance through the injection system. What stands out above all else is the sulfur level. Based on previous research of pyrolysis oil derived from HDPE, the sulfur content of WPO is estimated to be less than 0.01 wt%, which is only a fraction of the sulfur present in standard commercial diesel. Because of this difference, combusting WPO would release significantly lower amounts of SO₂ compared to the same quantity of diesel. This naturally provides an obvious ecological benefit to the process without any further refining. It should be noted that the sulfur content values are explicitly cited from prior literature dealing with HDPE-derived pyrolysis oil of similar density and energy content (Stone, R. (2012); Gungor (2015)). Figure 4 represents a normalized radar comparison of key fuel properties, visualizing the near equivalence of WPO to diesel across all tested parameters.

3 Experimental setup and instrumentation

3.1 Engine platform

The experimental investigations were conducted utilizing a single-cylinder, four-stroke multi-fuel VCR diesel engine coupled with an eddy current dynamometer setup. The CI research engine is equipped with several special features, including a tilting cylinder block that allows step increases in compression ratio (CR) from 14 to 18 without halting the engine or removing any of the combustion components from the cylinder liner. The details of the baseline geometry of the test engine rig and operating conditions are presented in Figure 5 and Table 1. The open engine control unit (ECU) module controlled the injection coil, fuel injector, fuel pump, and idle air control valve and maintained the engine speed at 1500 rpm (±5 rpm) by going up/down while varying the fuel blend ratio and load. Experimental tests were repeated three times to verify the data consistency for each measurement and minimize random errors. The final values reported in the figures and tables represent the calculated mean of these replicates. An uncertainty and error analysis was conducted for all measured parameters, with data expressed as mean ± standard deviation (SD). The relative uncertainties for major parameters, including brake thermal efficiency (BTE) and brake-specific fuel consumption (BSFC), were maintained within acceptable limits as detailed in the uncertainty analysis section.

TABLE 1. Complete specification of the VCR research engine.

3.2 Emission measurement

The exhaust levels of carbon monoxide (CO), carbon dioxide (CO₂), and nitrogen oxides (NO_*x*) were evaluated through an AVL DiTEST CDS 440 multi-gas analyzer. The measurement technique used for CO and CO₂ is non-dispersive infrared (NDIR) absorption, while an electrochemical sensor is used for NO_*x*. Prior to each test day, the analyzer was calibrated using certified reference gas mixtures. Details of the characteristics of the analyzer and associated measurement uncertainties are given in Table 2.

TABLE 2. Instrumentation details and accuracy specifications.

Unburned hydrocarbon (THC) emissions were not measured in the present investigation, as the AVL DiTEST CDS 440 analyzer used in this study was configured for CO, CO₂, and NO_*x* channels only and did not include a dedicated flame ionization detector (FID)-based HC measurement channel.

3.3 Test matrix and procedure

A full-factorial test matrix consisting of 4 × 5 × 6 = 120 operating points was tested. Four compression ratios (CR = 15, 16, 17, 18) and five fuel configuration blends (0%, 10%, 20%, 30%, and 40%) were tested at six load steps (2 kg, 4 kg, 6 kg, 8 kg, 10 kg, and 12 kg) under a constant 1500 rpm operating condition. Zero-load data points were excluded from analysis to avoid no-load anomalies in efficiency calculations. For all test conditions, a constant amount of coolant of 230 L/h was supplied to the engine jacket, and 90 L/h went to the exhaust heat exchanger. Five-minute or greater duration averages were recorded from 10 consecutive engine cycles. Instead of preparing the WPO–diesel premixes days ahead of testing, the diesel fuel and WPO–diesel mixtures were measured by volume inside sealed containers immediately prior to engine testing. For the mixtures, each fuel blend was thoroughly homogenized using a digital magnetic stirrer at 500 rpm for 20 min immediately prior to engine testing to ensure absolute mixture uniformity and prevent phase separation; no chemical stabilizers or surfactants were added. The test fuel and compression ratio sequences for the fuels were randomly varied at each load step; the purpose being to avoid confounding equipment creep with differences with real fuel effects.

4 Results and discussion

4.1 Brake power

Brake power (BP) increased as a function of engine load for all blends and across all compression ratios from 0.55–0.68 kW for 2 kg to 3.32–3.48 kW for 12 kg, driven by the increased fuel delivery rates and higher in-cylinder pressures that accompany an increasing torque load. B30 achieved the most favorable full-load BPs (up to 3.0% higher) at CR 15–17 compared to neat diesel. This is a localized improvement due to the special physical characteristics of WPO from HDPE. The WPO fractions have a much lower kinematic viscosity (2.1 mm² vs. 2.6 mm² of diesel) and flash point (36 °C vs. 52 °C of diesel). These specific parameters of WPO properties provide a faster evaporation rate from the fuel spray and early mixing of charge in ignition delay with CR 15 to 17. Neat diesel was recovered at CR 18 BP of 3.45 kW. The vaporization benefits provided by the WPO blends were completely offset by the increased in-cylinder temperatures and pressures at CR 18, which ramped up the chemical reactions of the base fuel. The BP penalties for the lower heating values (LHV) of B30 and B40 were 2.0% and 0.3%, respectively. The B40 fuel had slightly poorer performance than B30 at most intermediate loading rates; this is related to a slight shift in droplet size distribution, which resulted in some saturation at high fuel delivery rates. The lower concentration blends at 10% and 20%, however, stayed within 1.5% of the baseline diesel curve for all the tested parameters, revealing their excellent acceptability as true drop-in alternative fuels, as displayed in Figure 6. The B40 blend at CR 17 produced the full-load brake power of 3.46 kW, which is quite competitive and as high as the highest waste-utilization ratio reported by Alawa and Chakma (2023).

4.2 Brake thermal efficiency

Figure 7a demonstrates a steady, upward trend in brake thermal efficiency alongside engine load for all blends across all compression ratios. Specifically, the highest BTE of 38.70% was achieved by the reported full-load brake power of 3.44 kW achieved by the B40 blend at CR 17 outpaces the maximum blend performance documented by Alawa and Chakma (2023). In their study, a maximum brake power of 3.37 kW was recorded, though it is worth noting their parameters were limited to a maximum 20% blend ratio and a peak compression ratio of 16.7. At 2–4 kg, inter-blend BTE differences remained within 2%–3%, as lean in-cylinder conditions produced comparable combustion quality regardless of blend concentration. With increasing load, B40 progressively diverged from baseline diesel fuel and lower blend ratios (Govinda Rao et al., 2018). At full load (12 kg), B40 recorded BTE values of 38.18%, 37.76%, 38.47%, and 38.70% at CR 15, 16, 17, and 18, respectively, exceeding neat diesel by 13.8%, 13.7%, 16.0%, and 16.1% in the same sequence. Under low-to-intermediate loads, B30 underperformed B20 because of a change in its spray configuration, but it quickly caught up to 35.5%–37.0% at full load when the thermal environment had stabilized.

Although CR = 18 with B40 achieves the absolute maximum BTE 38.70%, CR = 17 with B40 delivers 38.47%, a difference of only 0.23 percentage points while producing substantially lower NO_*x* emissions (310 ppm vs. 470 ppm at CR = 18). This emissions advantage at CR = 17, combined with near-equivalent thermal efficiency, identifies CR = 17 with B40 at full load as the optimal operating condition when both efficiency and emission targets are considered together.

Figure 7b shows the percentage difference of the brake thermal efficiency (BTE) of the different blends from the standard diesel in a range of engine loads (2 kg–12 kg). The comparative analysis shows that as the percentage of WPO blends increases, BTE also increases (Travadi and Subhedar, 2025; Modi et al., 2025). A key observation from the results shows a clear trend where the efficiency of the blends improves as the load increases. B20 provides a consistent, modest efficiency gain of 2–4% across most of the mid-range loads. B30 starts as a low-performing blend at a low load (−18.8%) but recovers to being highly efficient (+8.8%) at maximum load. This suggests that B30 requires high cylinder temperatures and pressures to atomize and burn effectively. The B40 blend is clearly the most efficient across nearly all load conditions. At the maximum load of 12 kg, B40 is 15% more efficient than standard diesel.

4.3 Brake-specific fuel consumption

The consumption of fuel declined sharply with increasing load across all blends and compression ratios, falling from 0.60–0.80 kg/kWh at 2 kg to 0.24–0.30 kg/kWh at 12 kg. As illustrated in Figure 8a, this improvement is a well-established consequence of improved combustion efficiency, reduced friction-dominated losses, and higher mean effective pressure utilization at increased thermal loads. At low load, the 10% blend measured the highest BSFC across all CRs, reaching 0.80 kg/kWh at CR15, which is a 25.0% penalty relative to diesel that is attributed to its inferior atomization under lean, thermally underdeveloped in-cylinder conditions. The marginal cetane contribution is insufficient to offset combustion inefficiency at light loads. As load increased, combustion completeness progressively improved with increasing load, and B30 delivered the lowest full-load BSFC of 0.24 kg/kWh at CR 15, representing an 11.1% reduction over neat diesel 0.27 kg/kWh, driven by WPO’s improved spray atomization enabling near-complete fuel energy conversion and advancing combustion phasing toward the optimum crank angle (Kalargaris et al., 2017b).

B20 achieved equally competitive full-load values of 0.24 kg/kWh at CR 17 and CR 18, confirming that intermediate blend concentrations balance atomization efficiency gains against calorific value penalty most effectively at higher compression ratios. At CR = 17 and full load, the B40 blend achieved a BSFC of 0.26 kg/kWh. This value is equivalent to B30 at CR = 15 and is within 3.7% of neat diesel under the same conditions. This confirms that the higher compression ratio increases combustion completeness, effectively compensating for the lower calorific value of B40. Although B40’s calculated calorific value-weighted blend (LHV = 41.0 MJ/kg) is slightly lower than that of neat diesel (42.0 MJ/kg), necessitating marginally greater volumetric fuel delivery, the accompanying improved atomization and shortened ignition delay at CR = 17 more than offset this thermodynamically, yielding competitive full-load BSFC (Alawa and Chakma, 2023).

Figure 8b presents the BSFC heat map at full load (12 kg) across compression ratios 15–18 and WPO blend ratios B0–B40. The range of BSFC is 0.240–0.300 kg/kWh, with high consumption at low compression ratios and the pure diesel (B0) condition. The notable, and expected, reduction in BSFC is clearly observed with increasing WPO blend fraction, especially in the B20–B30 range. The optimum operating condition (CR 15) and the global minimum BSFC (0.240 kg/kWh) are obtained with B30 and also at CR 17 and CR 18 with B20. This is because the moderate proportions of blended WPO have proved to be suitable for partial substitution of diesel fuel across the compression ratios tested and improve combustion efficiency.

4.4 Carbon monoxide emissions

CO emissions decreased steadily with increasing engine load for all fuel blends and compression ratios, with higher in-cylinder temperatures at higher loads leading to more complete oxidation of CO to CO₂. At CR = 15, diesel had the highest minimum load 2 kg CO of 0.14%, whereas B40 had 0.10%, a 28.6% reduction, which is credited to the WPO’s ability to perform better in spray atomization and lower viscosity, allowing more complete oxidation in locally fuel-rich regions. At CR = 16, B20, B30, and B40 achieved 0.02% at full load, representing a 33.3% reduction over diesel’s 0.03%. B20 had the lowest value of 0.01% at 12 kg, which is depicted in Figure 9. It corresponds to a synergistic effect between the decrease in ignition delay and increase in combustion completeness with high compression.

Higher compression translates to a greater combustion efficiency benefit of WPO blends, as B40 achieved 0.015% at 10 kg load at CR = 18, which is 10% lower than diesel. On average, the overall decrease in mean no-load CO was approximately 47.5% when the CR was increased from 15 to 18, implying that the combustion completeness is strongly governed by the compression ratio. The optimum condition is CR = 17 with B30, as reported by Damodharan et al. (2019). The enhanced atomization dynamics, triggered by the 19% lower kinematic viscosity of WPO, yield a significantly reduced Sauter mean diameter (SMD) of the fuel spray. These fine spray structures accelerate droplet evaporation and expand the local oxygen availability within the fuel-rich core, driving the complete chemical oxidation of CO to CO₂ under high thermal loads.

4.5 NO_*x* emissions

The Zeldovich thermal NO_*x* mechanism, which suggests higher peak combustion temperatures and more oxygen available at higher loads, increases the oxidation rate of nitrogen, resulting in NO_*x* emissions that monotonically increased with all applied engine load conditions for all compression ratios and blends. An analysis of the recorded results for this test, shown in Figure 10, indicates that diesel recorded 320 ppm at full load, while the B20 and B30 blends recorded 310 ppm and 315 ppm, respectively, and the B40 blend recorded 285 ppm, which is a 10.9% reduction due to the slightly lower adiabatic flame temperature of the higher WPO blend fractions. The 10% blend exhibited marginally elevated NO_*x* at several load points, consistent with the advancing heat release and increasing local flame temperatures. With increasing compression ratio, NO_*x* increased substantially; at CR = 18 and full load, diesel reached 475 ppm while B20 and B30 peaked at 520 ppm, reflecting the higher end-of-compression temperatures intensifying premixed combustion and prolonging residence time in high-temperature zones. Conversely, B40 at CR = 18 (470 ppm) is comparable to diesel, suggesting that the dilution effect of the plastic-derived fraction moderates thermal NO_*x* formation despite increased compression. The increased NO_*x* at B20 and B30 at CR = 18 (520 ppm) compared to neat diesel (475 ppm) is attributed to the slightly higher adiabatic flame temperature of intermediate blend fractions under peak compression conditions, where the dilution effect of the plastic-derived fraction is insufficient to moderate thermal NO_*x* formation.

Critically, B40 at CR = 17 produced only 310 ppm at full load, which is 34.7% lower than B40 at CR = 18 (470 ppm) and 3.1% lower than diesel at CR = 17 (320 ppm). This confirms CR = 17 with B40 as the optimal condition for the NO_*x*–BTE trade-off: maximum efficiency gain (38.47% BTE, +16.0% vs. diesel) is achieved with the lowest NO_*x* among high-efficiency operating points. The results are consistent with those of Faisal et al. (2023), who found that an increase in the WPO blend fractions at a lower compression ratio always resulted in lower NO_*x* emissions because the heat release rate was lowered.

The unexpected increase in NO_*x* emissions for the B20 and B30 blends at the peak compression ratio (CR = 18) and maximum load conditions, reaching 520 ppm against the 475 ppm of standard diesel, demands a deeper mechanistic evaluation. The heavier molecular weight fractions present in the waste plastic oil prolong the physical and chemical ignition delay. This extended delay allows a greater volume of the combustible mixture to accumulate within the cylinder, which then ignites rapidly, substantially increasing the severity and fraction of the premixed combustion phase. Additionally, these intermediate blend fractions distinctly alter the local equivalence ratio (Φ) distribution, forming localized fuel-rich pockets that mix optimally with the highly compressed air. When this accumulated charge combusts, it generates intense, localized heat release that pushes the peak adiabatic flame temperature significantly higher than that of standard diesel. Because the extended Zeldovich mechanism for thermal NO_*x* formation is exponentially dependent on these elevated peak flame temperatures and sufficient oxygen availability, this severe premixed combustion pulse directly drives the NO_*x* kinetics higher. Therefore, the combination of extended ignition delay and intense localized heat release perfectly explains why intermediate blends produce more thermal NO_*x* than baseline diesel under increased thermal loading conditions.

4.6 CO₂ emissions

Figure 11 shows that CO₂ emissions increased monotonically with load for all fuel blends and compression ratios, as carbon oxidation efficiency increases with increasing thermal state as combustion is more complete. B40 consistently produced the least amount of CO₂ because its carbon-to-hydrogen ratio is lower, which resulted in less mass of CO₂ created per combustion cycle. At CR = 17 and full load, B40 achieved 1.80%, compared to diesel at 2.0%, which is a 10% reduction, and thus, the carbon structure of WPO based on HDPE gives a small greenhouse gas (GHG) benefit. As expected, as a result of the systematic increase in CO₂ from CR = 15 to 18, the combustion completeness is increasingly improved, in agreement with the results reported by Wathakit et al. (2021).

The steady increase in CO₂ emissions fundamentally tracks the heightened combustion temperatures and optimized air–fuel mixing dynamics. As the advanced atomization sequence interacts with the high in-cylinder thermal state, the local equivalence ratio homogenizes, ensuring that the chemical energy release is shifted entirely toward complete carbon conversion.

4.7 NO_*x*–CO trade-off

Figure 12 shows the experimental assessment of the emission profiles at maximum load (12 kg). For all compression ratios (CR), an inverse classical relationship was found between NO_*x* and CO emissions. As CR increases from 15 to 18, the NO_*x* content of neat diesel increases by 48.4% (320–475 ppm) due to higher peak temperatures in the combustion chamber that promote Zeldovich NO_*x* formation, while the CO content is lowered by 0.04 vol% to 0.01 vol%. In all the blends and CRs tested, the B40 blend with CR = 17 presents the most favorable combination of NO_*x* and CO: CO 0.020 vol% (close to diesel), and NO_*x* 310 ppm (low compared with B40 at CR = 18, 470 ppm and B40 at CR = 18, 320 ppm). This trade-off space for all 120 operating conditions, and at full load for each blend, is shown in Figure 10. The optimum compression ratio for B40 blends with respect to emissions, completeness, and moderate heat release rate of the WPO blends is found to be CR = 17, with no severe NO_*x* surges at this point. Thus, CR = 17 is identified as the optimum balance for WPO–diesel blends to reduce high NO_*x* emissions and achieve complete combustion, according to Faisal et al. (2023).

5 Machine learning and multi-objective optimization

A machine learning approach was utilized to forecast the performance and emission behavior of a CI engine operating on diesel blended with WPO. This predictive modeling aims to deepen the understanding of the empirical results and clarify the impact of key operating variables. Experimental data were used to develop predictive models, which could predict engine behavior for various operating conditions. This study combines experimental testing with machine learning modeling and multi-objective optimization to determine the best operating parameters for a VCR compression-ignition engine running on diesel blends mixed with WPO. The methodology is presented sequentially in Figure 13. Post-consumer HDPE waste was pyrolyzed at 400 °C–500 °C, and the product was mixed with diesel 0%–40% by volume (w/w) and called WPO. A full-factorial experimental campaign across four compression ratios (CR 15–18), six load levels (2–12 kg), and five blend configurations yielded 120 data points measuring BP, BTE, BSFC, CO, NO_*x*, and CO₂. Data were partitioned 70:30 for training and testing, with MinMaxScaler normalization applied for ANN and SVR models.

Four algorithms (random forest (RF), XGBoost, ANN, and SVR) were trained with hyperparameters tuned via 10-fold cross-validated search. Model performance was assessed using R ², root mean squared error (RMSE), mean absolute error (MAE), and 10-fold cross-validation. RF achieved BTE cross-validation R ² = 0.990 ± 0.008, the highest stability among all models.

Multi-objective optimization via RF surrogate grid search and NSGA-II Pareto analysis converged at CR = 17, B40, and load = 12 kg. This configuration was confirmed experimentally to yield BTE = 38.70%, NO_*x* = 310 ppm, and BSFC = 0.260 kg/kWh, representing the emission-balanced optimal condition of this study.

5.1 Dataset preparation

In the present work, a dedicated dataset was generated by operating a single-cylinder VCR engine under the diverse conditions outlined in Table 3 to train and develop the computational model. A total of 120 data points were gathered through an experimental campaign, increasing the compression ratio from 15 to 18 bar and the engine load from 2 kg to 12 kg with increments of 2 kg (six load levels in total) while changing the fuel blend of the diesel from 0% to 40% at 10% increments. The model incorporates three distinct input parameters: the blend percentage of WPO, the engine load, and the compression ratio of the single-cylinder VCR engine. In the numerical problem, the fuel blend labels were transformed into a numerical value representing the WPO in the diesel mixture (0%, 10%, 20%, 30%, or 40%). The model yields six distinct outputs derived from the experimental data, encompassing BTE and BSFC alongside CO, nitrogen oxides, and carbon dioxide emission profiles.

TABLE 3. Experimental design matrix and factor constraints.

The statistical framework evaluates the absolute parameters, including mean, minimum, maximum, skewness, and kurtosis for both performance parameters (BP, BTE, and BSFC) and exhaust emissions (CO, NO_*x*, and CO₂) in Table 4. The low skewness factors across the core variables confirm a highly stabilized, unbiased distribution devoid of extreme experimental outliers (0 kg load anomalies). This robust, clean data distribution establishes a mathematically sound foundation for the subsequent hyperparameter tuning and multi-objective machine learning modeling.

TABLE 4. Descriptive statistical analysis of distribution dynamics of input–output framework.

5.2 Hyperparameter optimization

To avoid underfitting or overfitting, hyperparameters were carefully tuned before deploying the machine learning predictive pipeline. A systematic procedure was used to implement a grid search framework integrated with a grid search CV 10-fold cross-validation over the basic training subset found. Brake thermal efficiency (BTE) was chosen as a representative target variable to identify the optimum hyperparameters for tuning all the algorithms. The investigated search space of the algorithms and the corresponding optimal configurations are summarized in Table 5. The grid search was found to be effective in setting architecture boundaries for each learner. The parameters controlling the complexity of the model and voting variances were carefully tuned for the RF and XGBoost (XGB) ensemble techniques, including the number of estimators n_estimators and tree depth max_depth.

y ^ = 1 K ∑ k = 1 K T k x ,

TABLE 5. Hyperparameter optimization architecture and best selected parameters via GridSearchCV.

where T_k(x) represents the discrete performance or emission mapping function of the kth independent decision tree trained on a bootstrapped data partition.

A MinMaxScaler was employed to normalize the input features, ensuring optimal performance for the support vector machine (SVR) and multilayer perceptron (MLP-ANN) architectures, both of which are heavily influenced by feature scaling. The best activation function, number of hidden layers, and penalty coefficient (C) were chosen as they give the lowest mean squared error (MSE) in the execution of the grid. The rigorous foundation setup for all four models is at their highest mathematical level prior to comparative validation.

5.3 Predictive performance evaluation and cross-validation framework

A dual-validation approach of an independent train-test split (70:30) and a robust 10-fold cross-validation scheme was followed to comprehensively assess and evaluate the forecasting accuracy of finely-tuned machine learning models, including RF, XGB, ANN, and SVR. The multi-criteria statistical performance measures presented in Table 6 (coefficient of determination (R2), RMSE, and MAE) are calculated for both the training set and the test set for all six output parameters of the engine.

R 2 = 1 − ∑ y i − y ^ i 2 ∑ y i − y ¯ i 2 ,

RMSE = 1 n ∑ y i − y ^ i 2 ,

MAE = 1 n ∑ y i − y ^ i ,

TABLE 6. Comparative predictive performance matrix of machine learning algorithms across core targets.

where y i represents the experimentally measured engine performance or emission output and y ^ i is the corresponding value predicted by the machine learning algorithm. The dataset average is denoted by y ¯ i. It is derived from (n = 120) operating test runs.

The empirical results indicate that the algorithms are highly predictive, especially the ensemble-based ones. The best learner was the random forest model, which achieved an outstanding test R2 of 0.9972 for BP and 0.9911 for BTE. This means that about 99.7% and 99.1% of the total experimental variances of engine power and thermal efficiency can be explained by the RF network, respectively. These metrics of testing errors were also very low for the localized test set, with root mean squared error (RMSE) of 0.0464 kW for BP and 0.7503% for BTE. XGBoost (XGB) was the next most competitive model, with greater model optimization on emission vectors such as NO_*x* (Test R2 = 0.9451). Most configurations achieved satisfactory testing accuracy. However, standard architectures, namely, the MLP-ANN framework and support vector regression (SVR), demonstrated a marginally lower capacity to generalize, owing to the highly non-linear nature of combustion chemical kinetics. However, “unacceptably high cross-validation variance” is observed for the EXCEPTION targets (BSFC, CV R ² = 0.552 ± 0.438) and (NO_*x*, CV R ² = 0.556 ± 0.122), which are definitely not reliable surrogate targets.

To evaluate the model for potential overfitting and verify the numerical stability of the predictive framework, a 10-fold cross-validation approach was implemented. Table 7 provides the mean cross-validated R2, its standard deviation (±SD), and mean RMSE across the 10 independent folds.

TABLE 7. 10-fold cross-validation stability and generalization analysis with ML model performance comparison.

Figures 14a–e compare the experimental measurements against the predicted values, revealing a tight alignment along the diagonal for BTE, BSFC, CO, NO_*x*, and CO₂. With R2 scores bounded between 0.942 and 0.996, the model demonstrates a robust capability to generalize the core performance and emission trends. Residual plots in Figures 14f–j show scattered errors around zero with no apparent pattern. These findings prove the absence of bias and confirm predictive consistency for the model across all operating conditions. This behavior arises because fuel injection and energy throughput scale directly with rising engine torque demands. The 10-fold cross-validation framework confirms the superior generalization of the ensemble regressors (RF and XGBoost) over conventional algorithms. The cross-validated mean R ² values for BP (0.996 ± 0.002) and BTE (0.990 ± 0.008) for random forest exhibit a narrow standard deviation, confirming prediction stability across random data partitions. For emission targets, RF and XGBoost maintained cross-validation R ² above 0.850 for CO and CO₂. The ANN showed unacceptable instability for NO_*x* (CV R ² = 0.556 ± 0.122), and the SVR failed for BSFC (CV R ² = 0.552 ± 0.438); these model-target combinations are excluded from the surrogate-based optimization. This statistical consistency validates the use of RF and XGBoost as the surrogate engine models for the NSGA-II optimization phase.

5.4 Feature importance and parametric sensitivity analysis

The random forest method was utilized to rank the impact of the operational parameters (compression ratio, load, and blend ratio) on exhaust emissions and engine performance. These priority rankings are detailed in Table 8 and visualized in Figure 14. The sensitivity analysis reveals that engine load is the most dominant governing factor dictating the overall system behavior, particularly for brake power (BP) and nitrogen oxide (NO_*x*) emissions, contributing 91.38% and 72.86%, respectively. This is attributed to the direct scaling of fuel injection and energy throughput with increasing engine torque requirements.

TABLE 8. Relative feature importance contribution matrix (%) of operational variables.

Conversely, the CR exhibits a substantial evolutionary influence on CO emissions (32.15%) and brake thermal efficiency (BTE) (12.45%), owing to its direct control over the peak cylinder temperature and air-fuel mixing dynamics. The WPO blend percentage carries localized impact by creating significant sensitivity regarding fuel consumption metrics, specifically BSFC at 16.83% and carbon emissions such as CO at 22.67%. These shifts stem from modified thermochemical properties and viscosity levels in WPO blends. Figure 15 illustrates how load manages the engine thermodynamic scale while variations in CR and fuel blending refine combustion efficiency along with overall emission output.

Figure 16 displays the predicted 3D response surfaces for BTE and BSFC when CR reaches 17 throughout the complete load domain. The BTE surface exhibits monotonic improvement under load for every blend fraction. B40 blend consistently reaches the highest surface peak. The BSFC surface demonstrates nearly equal performance for B30 and B40 when operating under high-load conditions. The minimum values systematically occur at the maximum engine load for the B40 blend, thereby validating the empirical observations.

5.5 Multi-objective optimization via NSGA-II

To resolve the inherent trade-offs between engine performance and environmental emissions, a multi-objective optimization problem was formulated using the trained random forest ensemble as a surrogate engine platform. The optimization framework is designed to simultaneously maximize performance parameters while minimizing fuel consumption and hazardous exhaust emissions. Mathematically, the multi-objective problem vector is defined as: Optimize F x = max ⁡ . BTE x , min ⁡ . BSFC x , ⁡ min ⁡ NOx x , ⁡ min ⁡ CO x T

Where the decision variable vector x = C R , L o a d , B l e n d T is bounded by operational constraints of the experimental matrix.

• CR Bounds: 15 ≤ C R ≤ 18
• Engine load bounds: 2 ≤ L o a d ≤ 12
• WPO blend bounds: 0 ≤ B l e n d ≤ 40

To execute the NSGA-II framework, the genetic algorithm parameters were configured with a population size of 100, a crossover probability of 0.9 using simulated binary crossover (SBX), and a mutation probability of 0.1 using polynomial mutation. The convergence criteria were strictly defined based on two conditions described below, ensuring that the global optimal front was robustly stabilized.

• A maximum generation limit of 500.
• When the average change in the spread of the Pareto front remained below a tolerance threshold of 10⁻⁶ over 50 consecutive generations.

The genetic algorithm determined an optimal configuration to be CR = 17, engine load of 12.00 kg, and a WPO of 40.00% (B40). At these specific data points, validated BTE reached 38.47% with BP of 3.44 kW using B40 fuel during full engine operation. While increased alternative fuel blends caused slight increases in BSFC = 0.260 kg/kWh and also NO_*x* output (310 ppm), maximum cylinder pressure plus thermal performance at this compression ratio successfully countered standard efficiency losses historically associated with raw plastic-derived fuels during high-load testing protocols. CO emissions at this condition were 0.020 vol%, and CO₂ was 1.80%; both are within acceptable limits compared to neat diesel. This confirms that operating the engine at its upper mechanical threshold enables efficient compatibility of high-volume (40%) plastic waste oil without significant performance degradation.

It should be noted that the NSGA-II surrogate model predicted BP = 3.44 kW, BSFC = 0.260 kg/kWh, NO_*x* = 310 ppm, and CO₂ = 1.80% at the converged optimum. These surrogate predictions are inconsistent with experimentally measured values under the same conditions, indicating that the RF surrogate model extrapolated beyond the training data boundary for certain outputs. The experimentally validated values at CR = 17, B40, and 12 kg load, as reported in Sections 4.1–4.6, are summarized in Table 9 and are therefore used as the confirmed optimal outcome of this study.

TABLE 9. Engine optimization condition at CR = 17, blend = B40, and load = 12 kg.

Figure 17 shows the BTE vs. NO_*x* space: (a) identifies the optimal operating condition at CR = 17, load = 12 kg, and 40% WPO blends, achieving a BTE of 38.7% at a relatively controlled NO_*x* of 310 ppm among all high-load points. The blend sensitivity plot (b) confirms that the 40% blend at 12 kg load yields the highest BTE. The optimal output summary (c) validates that this condition delivers strong performance metrics: BP of 3.44 kW and BTE of 38.47% with acceptable emission levels, with CO at 0.02 vol% and CO₂ at 1.80 vol%, although NO_*x* at 310 ppm warrants further attention for emission compliance.

5.6 Revised mechanistic basis for BTE interpretation

In an oxygen-free environment, high-density polyethylene (HDPE) is thermally treated, transforming it entirely to a liquid product (WPO), which is mainly composed of paraffinic and olefinic hydrocarbons. At peak operating load, the BTE trends of the WPO blends indicate that the governing mechanisms are:

• Low flash point and early vaporization: The 36 °C flash point of WPO is notably lower than the 52 °C baseline of conventional diesel. Consequently, a higher proportion of the fuel charge vaporizes ahead of schedule within the compression stroke, promoting a more uniform fuel–air blend. The resulting reduction in ignition delay improves combustion efficiency, especially at high compression ratios where the in-cylinder charge temperature is high (Kalargaris et al., 2017a).
• Comparable hydrogen to carbon (H/C) ratio: The WPO consists predominantly of straight-chain aliphatic hydrocarbons with an H/C ratio of 1.81 that is marginally lower than diesel’s 1.83. This comparable ratio confirms similar combustion stoichiometry, meaning the air–fuel ratio requirements for complete combustion remain close to diesel, supporting stable engine operation across all blend ratios (Kumar and Singh, 2011).
• Better fuel atomization dynamics: The WPO demonstrates a 19% reduction in 40 °C kinematic viscosity when measured against standard diesel fuel (2.6 mm²/s). This results in a considerably lower Sauter mean diameter (SMD) of the injected fuel spray, which causes finer droplets, higher specific droplet surface area, and quicker evaporation rates, consequently increasing the overall intensity of air-fuel mixing (Mani et al., 2009).
• Compression ratio conversion efficiency: At high compression ratios (CR = 18.00), the high end-of-compression temperatures can overcome the chemical ignition delay of the heavier molecular fractions (C15–C25) in WPO. This brings the heat release more into the direction of top dead center (TDC), which gives maximum heat release thermodynamic work conversion efficiency (Govinda Rao et al., 2018).

5.7 Experimental accuracy and reliability analysis

To identify the statistical reliability of the experimental trends, the uncertainty propagation (UP) was carried out on the computed performance parameters, according to the standard law of error propagation. The calculated absolute and relative uncertainties for the various core engine parameters under the optimized B40 full-load matrix, along with the source instrumentation specifications, are summarized in Table 10.

TABLE 10. Propagated experimental uncertainty matrix of performance and emission metrics.

As summarized in Table 11, the present study achieves the highest reported peak BTE (38.47%) among directly comparable VCR -WPO studies while maintaining the lowest NOX at the optimal condition (310 ppm at CR = 17, B40), attributable to the combination of a controlled full-factorial test matrix, HDPE-specific feedstock selection, and data-driven multi-objective optimization. None of the comparable studies employed machine learning or systematic optimization frameworks alongside VCR testing.

TABLE 11. Comparative summary of key findings from directly related VCR engine studies using WPO.

6 Conclusion

This study presented an empirical assessment, data-driven prediction, and multi-objective optimization of a compression-ignition engine fueled with HDPE-derived waste plastic pyrolysis oil (WPO) blends (up to 40% on a mass basis) across varying compression ratios (CR 15–18) and engine loads. The principal quantitative findings are summarized as follows.

• ⁃ Fuel synthesis and energy density: Thermal pyrolysis of post-consumer HDPE waste (at 400 °C–500 °C with a clay catalyst) yielded 65–70 wt% of liquid fuel. The isolated WPO exhibited a high gross calorific value (40,950 kJ/kg), a density of 860 kg/m³, and an ultra-low sulfur content (<0.01 wt%), establishing its basic suitability as a low-emission drop-in component.
• ⁃ Major findings:
• Brake power (BP) for WPO blends up to B40 remained statistically equivalent to baseline mineral diesel across all tested loads and CRs.
• Brake thermal efficiency (BTE) scaled positively with both CR and WPO concentration, peaking at 38.70% for the B40 blend at CR 18. This represents a 16.1% relative improvement over neat diesel (33.33% BTE) under identical conditions, governed by WPO’s lower kinematic viscosity promoting finer spray atomization and more rapid, near-TDC heat release.
• Brake-specific fuel consumption (BSFC) for the B40 blend exceeded standard diesel by 0.260 kg/kWh at peak load. This moderate increment correlates directly with the marginally lower heating value of WPO relative to mineral diesel.
• For neat diesel specifically, advancing the compression ratio from 15 to 18 suppressed the tailpipe CO content from 0.04 vol% down to 0.02 vol% for 17 CR and 0.01 vol% with 18 CR.
• NO_*x* emission increased by 48.4%, increasing from 320 ppm to 475 ppm for the compression ratio ranges 15 to 18, driven by higher peak temperatures in the combustion chamber.
• Tailpipe CO₂ emissions remained practically identical across both fuels due to the comparable H/C ratio of the polyethylene-derived oil.
• ⁃ Predictive modeling and global optimization: For data-driven output mapping, the random forest algorithm outperformed the artificial neural network (ANN) in predictive accuracy. Addressing the fundamental BTE–NO_X trade-off, NSGA-II optimization established the global Pareto-optimal operating point at CR = 17 using the B40 blend at 100% load (12 kg). This condition achieved an ideal operational balance:
• BTE: 38.47%
• Brake power: 3.44 kW
• CO: 0.020 vol%
• NO_*x*: 310 ppm
• ⁃ Recommendations: B40 is validated as the upper-limit optimal blend for high-compression CI operation, while B10 to B20 are recommended as conservative, zero-modification drop-in fuels. Ultimately, this work demonstrates that integrating empirical matrix testing with machine learning (RF) and genetic algorithms (NSGA-II) provides a highly reliable framework for calibrating alternative fuel engines.

In summary, these quantitative findings transition HDPE-derived WPO from a conceptual waste-valorization strategy into a validated, high-efficiency drop-in fuel for standard compression-ignition engines.

7 Future scope

There are several logical extensions to this study.

• Crank-angle resolved combustion diagnostics: Acquiring high-speed in-cylinder pressure and heat release rate (HRR) data across the VCR matrix to explicitly quantify ignition delay and the premixed combustion phase duration. This would provide definitive kinetic validation for the proposed high-molecular-weight NO_*x* divergence mechanism.
• Parametric pyrolysis synthesis studies: Expanding the thermal cracking window beyond the fixed 400 °C–500 °C range utilized herein to systematically map how distinct pyrolysis temperatures alter WPO distillation curves, molecular speciation (characterized via GC-MS and FT-IR), Sauter mean diameter (SMD) spray atomization, auto-ignition characteristics, and the overall engine energy-exergy destruction.
• Long-term tribological and endurance trials: Executing extended mechanical durability testing (e.g., 100+ hour continuous cycles) to evaluate real-world operating impacts under high WPO blend ratios, with specific focus on fuel injector coking/fouling, cylinder liner wear rates, and lubricating oil degradation.
• Techno-economic analysis (TEA) and life-cycle assessment (LCA): Conducting a comprehensive cradle-to-grave evaluation to establish the commercial viability of decentralized WPO production within the Indian municipal framework, factoring in plastic waste collection logistics, localized reactor CAPEX/OPEX, and net carbon credits.
• Hybrid oxygenated multi-blending: Investigating binary or ternary hybrid fuel formulations incorporating oxygenated renewable additives (such as second-generation bio-ethanol or Karanja biodiesel) to chemically suppress the elevated CO emissions observed under low-load and low-CR regimes, thereby synthesizing a superior multi-feedstock alternative fuel.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Evgeny Rebrov, University of Warwick, United Kingdom

Reviewed by: Divyansh Singh, National Institute of Technology, Silchar, India

Maulik A. Modi, Kadi Sarva Vishwavidyalaya, India