Enhancing Engine Eﬃciency and Performance

through Machine Learning - Real-Time Engine

Tuning

Manindra de Mel

1[0009−0008−9892−8515]

Australian National University, College of Engineering, Computing and Cybernetics.

u7156805@anu.edu.au

Keywords: Machine Learning

Real-Time Engine Tuning

Engine Eﬃciency

Automotive Engineering

Predictive Model.

1 Introduction

Traditional engine tuning methods, relying on static settings, fail to adapt to

dynamic operational conditions, leading to suboptimal performance and inef-

ﬁciency. Static tuning methods are inherently limited in their adaptability to

real-time changes in engine conditions such as driving conditions, fuel quality,

and engine wear. Despite advancements in Electronic Control Units (ECUs),

which process real-time data from various sensors to optimize engine functions,

these systems operate on static maps and thresholds, limiting their adaptability

to the complex variability of real-world driving conditions. This highlights the

need for a paradigm shift towards real-time, adaptive engine tuning, leveraging

the latest in machine learning (ML) technology. [6]

This literature review will critically analyze key studies in both traditional

and ML-based engine tuning, highlighting the limitations of static methods and

the potential of ML algorithms to dynamically optimize engine parameters. The

review is structured to ﬁrst examine the principles and practices of traditional

static tuning methods, followed by an exploration of the key parameters for

real-time engine tuning. It then delves into various machine learning models

applied in engine tuning and concludes with practical applications, case studies,

and future research opportunities. By examining signiﬁcant papers, the review

will identify existing gaps and opportunities for future research, providing a

comprehensive understanding of the current state and future directions in this

ﬁeld.

2 Traditional Engine Tuning Methods

Engine tuning has historically relied on several key parameters and manual

adjustments to optimize performance and eﬃciency. Key parameters such as

air/fuel ratio (AFR), throttle position, engine load, and even driver behaviour

play critical roles in this process.

2 M. de Mel

2.1 Manual Adjustments and Fixed-Parameter Settings

The paper ”Tuning of a conventional diesel engine into a low compression ratio

diesel engine” published in the Journal of Engineering and Applied Sciences in

2015, delves into the methodologies for tuning conventional diesel engines to

operate at lower compression ratios. The study emphasizes manual adjustments,

such as using thicker head gaskets, to achieve desired engine performance. By

modifying the compression ratio from a standard 17.5:1 to 15.37:1 and 13.7:1

using thin copper spacers, the researchers could observe the eﬀects on engine

performance [1].

This research highlights the historical importance of manual adjustments

in engine tuning, detailing methods like altering head gaskets, employing low-

compression pistons, and adjusting stroke and rod lengths. These traditional

techniques are highlighted for their eﬀectiveness in controlling combustion, re-

ducing nitrogen oxides (NOx) emissions, and improving fuel eﬃciency. [1]

Key ﬁndings from the study include the reduction in NOx emissions with

lower compression ratios due to decreased peak combustion temperatures. How-

ever, this reduction comes with trade-oﬀs, such as increased emissions of hydro-

carbons (HC) and CO, lower brake thermal eﬃciency, and higher brake-speciﬁc

fuel consumption (BSFC). [1] These ﬁndings illustrate the inherent limitations

of static tuning methods, which cannot adapt to real-time changes in engine

conditions, leading to suboptimal performance.

2.2 Key Parameters for Engine Tuning

In ”Optimization of the Air/Fuel Ratio for Improved Engine Performance and

Reduced Emissions” [7], the authors investigate the impact of varying air/fuel

ratios (AFR) on engine performance and emissions. This study highlights the

crucial role of AFR in achieving optimal combustion eﬃciency, which directly

inﬂuences fuel economy and emission levels. The authors demonstrate that main-

taining a stoichiometric AFR (approximately 14.7:1 for gasoline engines) is essen-

tial for complete combustion, thereby optimizing fuel eﬃciency and minimizing

emissions. [7] However, the study also discusses the inherent trade-oﬀs: a leaner

mixture (higher AFR) can enhance fuel economy but results in higher NOx

emissions due to elevated combustion temperatures. Conversely, a richer mix-

ture (lower AFR) increases power output but leads to higher fuel consumption

and emissions of hydrocarbons (HC) and carbon monoxide (CO). [7]

The paper also discusses throttle position as another critical parameter di-

rectly tied to AFR management. It indicates how much the throttle is open,

which directly aﬀects the engine’s air intake. The throttle position sensor pro-

vides real-time data that helps adjust the AFR in real-time to suit various driving

conditions, ensuring eﬃcient operation under diﬀerent loads and speeds. [7] This

is crucial because the engine’s air intake directly impacts the AFR, and thus, the

overall eﬃciency and performance of the engine. Similarly, engine load, which

can be measured using a manifold absolute pressure (MAP) sensor, is vital for

determining the optimal AFR. Under high-load conditions, the engine requires

Real-time Engine Tuning 3

a richer mixture to produce more power, while under low-load conditions, a

leaner mixture can be used to improve fuel eﬃciency. [7] These adjustments

are necessary for maintaining the balance between performance and emissions,

demonstrating the interconnectedness of these parameters in achieving optimal

engine tuning.

The ideas presented by this paper highlight the importance of real-time AFR

adjustments and provides insights into critical considerations for developing more

optimal models, such as machine learning models, for adaptive engine tuning

systems.

Driving Style As A Key Parameter The study ”Vehicular Fuel Consump-

tion and CO2 Emission Estimation Model Integrating Novel Driving Behav-

ior Data Using Machine Learning” [8] takes a novel approach by incorporating

driving behavior data into fuel consumption and emission models. The authors

integrate data on dangerous driving behaviors, such as speeding, sudden acceler-

ation, and braking, into monthly fuel consumption models using random forest

regression. This comprehensive model accounts for real-world driving behaviors,

achieving high prediction accuracy. By factoring in driving behavior, the model

can make precise real-time adjustments to engine parameters, enhancing both

performance and eﬃciency. [8] This research emphasizes the profound impact of

driving behavior on fuel consumption and emissions, highlighting the necessity

for adaptive tuning systems that can accommodate dynamic driving conditions.

The integration of behavioral data into predictive models represents a signiﬁcant

advance in the ﬁeld, demonstrating how machine learning can be leveraged to

achieve more accurate and adaptable engine tuning.

Both papers highlight the critical parameters in engine tuning, such as the

air/fuel ratio and driving behavior, demonstrating their substantial impact on

engine performance and emissions. The SAE paper focuses on the technical op-

timization of AFR, showing its direct inﬂuence on combustion eﬃciency and

emissions. In contrast, the MDPI paper integrates behavioral data, illustrating

how real-world driving conditions can be factored into engine tuning models to

enhance prediction accuracy and adaptability. Together, these studies highlight

the importance of dynamic, real-time adjustments in engine tuning parameters

to achieve optimal performance and emissions control. This another aspect not

accounted for in the current engine operations.

2.3 Overall Insights of Static Tuning

Overall, these three papers provide a comprehensive overview of the advances

in traditional static tuning methods, highlighting their beneﬁts and current lim-

itations. While eﬀective in certain contexts, these methods are limited by their

inability to dynamically adjust to varying operating conditions, setting the stage

for more advanced, adaptive tuning approaches using machine learning. [1]

4 M. de Mel

3 Machine Learning in Engine Tuning

3.1 Overview of Diﬀerent ML Models Used for Engine Tuning

The paper ”Potential of Machine Learning Methods for Robust Performance

and Eﬃcient Engine Control Development” [3] provides a more comprehensive

overview of ML-based methods applied to engine control, focusing on reducing

the time-consuming calibration process. The paper highlights that traditional

map-based control approaches are becoming infeasible due to increasing system

complexity and real-world requirements, leading to unacceptable development

time and costs. The study identiﬁes AI-based methods as a disruptive technology

that can tackle these challenges by reducing the number of model parameters,

enabling on-line model parameter identiﬁcation, automating testing, and on-line

calibration of controllers. [3]

The paper discusses supervised learning (SL), unsupervised learning (UL),

and reinforcement learning (RL) methods. [2] SL methods, particularly regres-

sion techniques, are highlighted for their ability to model complex processes with

fewer parameters and real-time ECU implementation suitability. RL methods

are identiﬁed as promising due to their ability to learn optimal control poli-

cies through interactions with the environment, thus improving engine control

strategies. [3]

The IFAC paper highlights the transformative potential of AI-based methods

in engine control. It underscores the advantages of machine learning over tradi-

tional methods by reducing extensive calibration processes and enabling adap-

tive, intelligent engine control systems. Supervised learning streamlines engine

process modeling, while reinforcement learning ensures continuous adaptation

and optimal control. This paper is pivotal in demonstrating the critical need for

ML in engine tuning, providing a strong foundation for future advancements in

the ﬁeld.

Beyond this paper, other ML models such as least squares support vector ma-

chines (LS-SVM) combined with genetic algorithms (GA), and adaptive memory

online sequential extreme learning machines (AMOSELM), as well as RL models

have also been employed for real-time engine tuning.

LS-SVM and Genetic Algorithms In ”Automotive Engine Idle Speed Con-

trol Optimization Using Least Squares Support Vector Machine and Genetic

Algorithm” [10], the authors propose a novel approach to optimize engine idle

speed control using LS-SVM combined with GA. This study addresses the sig-

niﬁcant impact of the electronic control unit (ECU) parameters on engine idle

speed performance, which is crucial for fuel eﬃciency, driveability, and emission

control. The LS-SVM model is used to predict engine performance based on

various input parameters, while GA optimizes these parameters to achieve the

best performance. [10]

The study demonstrates that the LS-SVM model, trained using data from

dynamometer tests, accurately predicts engine performance. The GA then ef-

ﬁciently searches for the optimal ECU settings within user-deﬁned constraints.

Real-time Engine Tuning 5

This integrated approach signiﬁcantly improves idle speed performance, reducing

fuel consumption and emissions. The methodology is generic and can be applied

to various vehicle control optimization problems, making it highly relevant for

real-time engine tuning applications.

The practical implications of this study are profound, as it demonstrates

how combining LS-SVM and GA can streamline the optimization process, saving

time, fuel, and human resources while achieving superior performance compared

to traditional empirical tuning methods.

Eﬃcient Point-by-Point Engine Calibration Using Machine Learning

and Sequential Design of Experiment Strategies The study ”Eﬃcient

Point-by-Point Engine Calibration Using Machine Learning and Sequential De-

sign of Experiment Strategies” [9] introduces an innovative approach aimed at

optimizing engine calibration by signiﬁcantly reducing the number of experi-

ments required. Traditionally, engine calibration involves a labor-intensive trial-

and-error process to determine the optimal settings for the electronic control unit

(ECU) parameters. [1] This method becomes increasingly impractical as modern

engines incorporate more advanced technologies and require precise calibration

to meet stringent performance and emissions standards. [7]

To address these challenges, the authors propose a method that combines

sequential design of experiments (DoE) with an initial-training-free online ex-

treme learning machine (ITF-OELM). The sequential DoE strategy is used to

collect data iteratively, allowing for the model to be updated continuously with

new measurements. This approach ensures that only the most informative ex-

periments are conducted, thereby minimizing the number of tests needed. [9]

The ITF-OELM model is particularly suited for this application due to its

ability to learn incrementally without requiring a large initial dataset. This model

can start the learning process as soon as the ﬁrst data point is available and can

update its predictions as more data is gathered. [9] This is crucial for point-by-

point calibration, where the model must be accurate at each operating point

before moving to the next.

The study validates the proposed approach through simulations using a com-

mercial engine software, GT-Power, and real engine tests on a bench. The results

demonstrate that the method can achieve high calibration accuracy with signiﬁ-

cantly fewer experiments compared to traditional methods. [9] The approach not

only reduces the time and resources required for calibration but also improves

the precision of the calibration process by continuously reﬁning the model with

new data.

Overall, this study highlights the potential of combining advanced machine

learning techniques with eﬃcient experimental design strategies to enhance en-

gine calibration processes. The ITF-OELM model’s ability to learn and update

in real-time makes it a powerful tool for modern engine calibration, ensuring

optimal performance and compliance with environmental standards.

6 M. de Mel

3.2 Practical Applications and Case Studies Illustrating ML in

Real-Time Engine Tuning

AMOSELM The study ”An Adaptive On-Board Real-Time Model with Resid-

ual Online Learning for Gas Turbine Engines Using Adaptive Memory Online

Sequential Extreme Learning Machine” [11] proposes an adaptive on-board real-

time model (AORM) for gas turbine engines. This model combines a component-

level model (CLM) with a residual online learning model (ROLM) utilizing the

AMOSELM algorithm. The AMOSELM algorithm eﬀectively learns the resid-

uals between the engine and the CLM during engine service, improving the

accuracy of the model. [11]

The AMOSELM algorithm stands out for its ability to handle both grad-

ual performance degradation and sudden changes in engine performance. The

algorithm employs a bell-type membership function to adjust the memory coef-

ﬁcient during the online learning stage. The shape parameters of this membership

function are optimized oﬄine using particle swarm optimization (PSO), which

ensures robust real-time learning and accurate model updates.

Veriﬁcation through virtual engine ﬂight data simulation and actual ground

tests conﬁrms that the AMOSELM algorithm signiﬁcantly reduces modeling

errors, [11] outperforming traditional online learning algorithms like OSELM

and MOSELM. The study shows that the AORM framework, enhanced by

AMOSELM, achieves better alignment between the real-time model and the

actual engine, thereby improving the overall accuracy and reliability of engine

performance predictions

Reinforcement Learning for PID Tuning The ”Reinforcement Learning

Approach to Autonomous PID Tuning” [2] study explores the use of reinforce-

ment learning (RL) for the autonomous tuning of proportional-integral-derivative

(PID) controllers, enhancing the control of dynamic systems, including engines.

PID controllers are widely used in industrial control systems, such as automotive

cruise control, where they maintain a car’s speed by adjusting the throttle based

on speed sensor input. [2]

Reinforcement Learning (RL) formulates the tuning problem as a decision-

making process, allowing an agent to learn optimal PID settings through smart

trial-and-error. This approach reduces the need for extensive on-line training

and minimizes equipment wear by allowing initial training to be conducted oﬀ-

line using an approximate step-response model. The RL agent interacts with

the environment, receives feedback, and updates its control policy to improve

performance. [2]

The study highlights RL’s eﬀectiveness in handling non-linear systems, with

the agent adapting to real-time process dynamics to improve setpoint tracking

and disturbance rejection. The proposed RL-based tuning scheme demonstrates

its applicability through a pilot-scale multi-modal tank system, showcasing its

potential for real-time engine tuning by enhancing performance, reducing train-

ing time, and ensuring safe operation. [3] This approach ensures that PID con-

Real-time Engine Tuning 7

trollers can adapt to changing conditions in real-time, providing a robust solution

for dynamic and complex engine control scenarios. [2]

Overall, these studies collectively highlight the signiﬁcant advancements made

in ML-based engine tuning. The integration of LS-SVM and GA has shown

practical improvements in optimizing engine parameters, while the AMOSELM

algorithm has proven eﬀective in real-time model updating and accuracy. The

use of RL for PID tuning further exempliﬁes the potential of ML in enhancing

control strategies and adapting to dynamic conditions. Together, these papers

highlight the transformative impact of ML in developing more eﬃcient, adaptive,

and intelligent engine control systems. However, despite these advancements, im-

plementing ML in real-time engine tuning presents several technical challenges

that must be addressed to fully realize its potential. The following section dis-

cusses these challenges in detail, providing insight into the complexities and con-

siderations necessary for the successful deployment of ML-based engine tuning

solutions in the automotive and aerospace industries.

3.3 Technical Challenges in Implementing ML for Real-Time

Tuning

Implementing machine learning (ML) for real-time engine tuning involves over-

coming several technical challenges highlighted in recent studies. One primary

challenge is ensuring data quality and preprocessing. The study on AMOSELM

emphasizes that the performance of ML algorithms heavily depends on the ac-

curacy and reliability of the input data. Poor data quality, including noise and

outliers, can signiﬁcantly aﬀect model accuracy. Thus, robust data preprocess-

ing steps, such as ﬁltering and normalization, are crucial to ensure eﬀective ML

model performance during real-time operations.

Another signiﬁcant challenge is the computational complexity and require-

ments of these models. The AMOSELM algorithm, for instance, introduces con-

siderable computational demands, necessitating high processing power to handle

continuous data ﬂow and perform rapid computations. [11] This is particularly

critical for real-time applications where models must run eﬃciently on existing

hardware, such as Full Authority Digital Electric Control (FADEC) systems in

aerospace. Balancing model accuracy with real-time performance requires op-

timized algorithms that can provide timely updates without causing excessive

computational delays.

Handling non-linear relationships between engine parameters is also a com-

plex task. The study on LS-SVM and GA addresses this by integrating genetic

algorithms to search for optimal settings. However, this process of tuning hyper-

parameters for both LS-SVM and GA is computationally intensive and requires

meticulous calibration to ensure the model converges to optimal solutions eﬃ-

ciently. [10] This adds another layer of complexity to the implementation.

Moreover, adaptive learning and model updating are essential for maintain-

ing model accuracy over time. The AMOSELM algorithm’s capability to adapt

to both gradual performance degradation and sudden changes is vital for real-

time applications. This adaptability, however, demands frequent updates and

8 M. de Mel

recalibration, which can be challenging. [11] The oﬄine optimization strategy

using particle swarm optimization (PSO) ensures robust real-time learning but

also necessitates a solid framework for periodic updates and maintenance. [11]

Finally, integrating ML models into existing engine control systems presents

signiﬁcant hurdles. Ensuring compatibility with current electronic control units

(ECUs) and seamless communication between the ML model and control systems

is critical. This integration must enhance performance without disrupting exist-

ing operational capabilities or introducing new vulnerabilities or failure points.

Ensuring that ML models can be integrated smoothly and provide added value

without compromising system integrity is a key challenge. [3]

Overall, while ML models like AMOSELM and LS-SVM with GA oﬀer sub-

stantial potential for optimizing engine operations, their implementation in real-

time environments requires addressing signiﬁcant technical challenges. Ensuring

data quality, managing computational complexity, handling non-linear relation-

ships, maintaining adaptive learning capabilities, and achieving seamless inte-

gration with existing systems are essential for the successful deployment of these

advanced ML models in real-world engine tuning applications.

3.4 Adapting To The Changing Automotive Industry

In recent years, the automotive industry is witnessing a signiﬁcant shift towards

more sustainable and eﬃcient powertrains, with hybrid electric vehicles (HEVs)

and fully electric vehicles (EVs) gaining prominence. This transition is driven

by increasing regulatory pressures to reduce emissions and improve fuel econ-

omy, along with growing consumer demand for greener transportation options.

In response, researchers are focusing on advanced technologies and strategies to

optimize the performance and eﬃciency of these vehicles. A key area of research

is the development of robust energy management strategies (EMSs) for HEVs,

which aim to optimize power distribution between the internal combustion en-

gine and the electric motor. Recent studies, such as the one discussed below,

illustrate the potential of machine learning methods in enhancing these EMSs.

The paper ”Optimal Energy Management Strategies for Hybrid Electric Ve-

hicles: A Recent Survey of Machine Learning Approaches” [4] explores the po-

tential of machine learning methods in developing robust and eﬃcient EMSs

for HEVs. This comprehensive review emphasizes the critical role of EMSs in

optimizing power distribution between the engine and motor, thereby maximiz-

ing fuel economy and minimizing emissions. The traditional approaches, mainly

based on optimal control theory, are becoming increasingly inadequate due to

the complexity and real-time requirements of modern HEVs. Machine learning

techniques oﬀer a promising alternative, capable of handling non-linear behaviors

and adapting to dynamic operating environments.

The study reviews various machine learning methods, including supervised

learning, reinforcement learning, and unsupervised learning, and their appli-

cations in EMS development. Supervised learning algorithms such as Random

Forests, Support Vector Machines (SVM), and Artiﬁcial Neural Networks (ANN)

are highlighted for their ability to predict vehicle behavior and optimize power

Real-time Engine Tuning 9

distribution eﬀectively. Reinforcement learning, particularly deep reinforcement

learning (DRL), is noted for its capacity to develop optimal control policies

through continuous interaction with the environment. This method has shown

signiﬁcant improvements in fuel eﬃciency and adaptability to diﬀerent driving

conditions. Unsupervised learning techniques, including clustering and dimen-

sionality reduction, are also discussed for their potential in identifying patterns

in driving behavior and optimizing EMS accordingly. [4]

Adding to the discussion, the paper ”A Deep Reinforcement Learning Based

Energy Management Strategy for Hybrid Electric Vehicles in Connected Traf-

ﬁc Environment” [5] proposes a novel deep reinforcement learning (DRL) based

EMS for HEVs within a connected traﬃc environment. The study employs a

deep deterministic policy gradient (DDPG) algorithm to integrate vehicle ref-

erence speed planning with energy management strategies. The upper layer of

this approach plans the vehicle’s reference speed, while the bottom layer em-

ploys an adaptive equivalent consumption minimization strategy (A-ECMS) to

manage power split control. This dual-layer strategy considers factors such as

distance headway, fuel consumption, and terrain information, resulting in a 3.5%

improvement in fuel consumption compared to traditional proportional-integral

(PI) controller-based strategies. [5]

The integration of machine learning in EMSs represents a transformative

shift in HEV technology, allowing for real-time, adaptive control that signiﬁ-

cantly enhances vehicle performance and eﬃciency. As the automotive market

continues to evolve towards more sustainable solutions, the application of ad-

vanced machine learning methods in EMS development is likely to play a crucial

role in meeting future energy eﬃciency and emission reduction goals.

4 Conclusion

This literature review has provided a comprehensive analysis of traditional and

machine learning-based engine tuning methods. The ﬁrst section on traditional

engine tuning highlighted key issues brought up in the reviewed papers, such as

the limitations of static tuning methods and the critical role of parameters like

air/fuel ratio (AFR), throttle position, and engine load. These studies identiﬁed

essential parameters that are crucial for further research in implementing ML

methods into engine tuning.

In the section on machine learning in engine tuning, the paper ”Potential of

Machine Learning Methods for Robust Performance and Eﬃcient Engine Con-

trol Development” [3] was pivotal in outlining the potential ML has in engine

control. This paper, along with others that followed, showcased signiﬁcant ad-

vancements in ML models for engine tuning. The integration of LS-SVM and GA

demonstrated practical improvements in optimizing engine parameters, while the

AMOSELM algorithm proved eﬀective in real-time model updating and accu-

racy. Additionally, the use of RL for PID tuning highlighted the adaptability

and eﬃciency of ML in enhancing control strategies and adapting to dynamic

conditions.

10 M. de Mel

The section on adapting to the changing automotive market emphasized the

shift towards hybrid electric vehicles (HEVs) and the development of energy

management strategies (EMSs) using ML methods. Papers like ”Optimal En-

ergy Management Strategies for Hybrid Electric Vehicles: A Recent Survey of

Machine Learning Approaches” [4] and ”A Deep Reinforcement Learning Based

Energy Management Strategy for Hybrid Electric Vehicles in Connected Traﬃc

Environment” [5] illustrated the application of ML in optimizing power distri-

bution between the internal combustion engine and electric motor, achieving

signiﬁcant improvements in fuel eﬃciency and emission reductions.

Despite these advancements, there remains a notable gap in the literature

concerning the implementation of constantly learning ML models speciﬁcally tai-

lored for internal combustion engines (ICEs) in the automotive industry. Current

research predominantly focuses on hybrid and electric powertrains, often over-

looking the majority of vehicles on the road today, which are still powered by

ICEs. The absence of real-time, adaptive ML models that can be seamlessly inte-

grated into existing ICEs presents a signiﬁcant opportunity for future research.

The proposal on ”Real-time Engine Tuning” [6] aims to address this gap by

developing ML models that can learn continuously and adapt in real-time to op-

timize engine eﬃciency for ICEs. By creating a framework that can be plugged

into the current ﬂeet of ICE vehicles, this research will pave the way for sub-

stantial improvements in fuel economy and emissions reduction. The proposed

solution will utilize real-time data to dynamically adjust engine parameters, en-

suring optimal performance under varying operating conditions.

In conclusion, while the path towards ML in real-time engine eﬃciency man-

agement is well-paved for hybrid and electric vehicles, There is a pressing need for

solutions tailored to ICEs. This gap presents a valuable opportunity for future re-

search to create innovative ML-based models that can enhance the eﬃciency and

sustainability of the existing automotive ﬂeet. By focusing on real-time adapt-

ability and continuous learning, my research will contribute to the development

of cutting-edge technologies that support the transition to more sustainable and

eﬃcient automotive practices.

References

1. Bridjesh, P., Gopal, A., Mohanamurugan, S.M.: Tuning of a conventional diesel

engine into a low compression ratio diesel engine. International Journal of Applied

Engineering Research (2015)

2. Dogru, O., Velswamy, K., Ibrahim, F., Wu, Y., Sundaramoorthy, A.S.,

Huang, B., Xu, S., Nixon, M., Bell, N.: Reinforcement learning ap-

proach to autonomous pid tuning. Computers & Chemical Engineer-

ing (2022). https://doi.org/https://doi.org/10.1016/j.compchemeng.2022.107760,

https://www.sciencedirect.com/science/article/pii/S0098135422001016

3. Garg, P., Silvas, E., Willems, F.: Potential of machine learning methods for

robust performance and eﬃcient engine control development. IFAC-PapersOnLine

(2021). https://doi.org/https://doi.org/10.1016/j.ifacol.2021.10.162, https:

//www.sciencedirect.com/science/article/pii/S2405896321015640

Real-time Engine Tuning 11

4. Jui, J.J., Ahmad, M.A., Molla, M.I., Rashid, M.I.M.: Optimal en-

ergy management strategies for hybrid electric vehicles: A recent sur-

vey of machine learning approaches. Journal of Engineering Research

(2024). https://doi.org/https://doi.org/10.1016/j.jer.2024.01.016, https:

//www.sciencedirect.com/science/article/pii/S2307187724000166

5. Li, J., Wu, X., Hu, S., Fan, J.: A deep reinforcement learn-

ing based energy management strategy for hybrid electric vehi-

cles in connected traﬃc environment. IFAC-PapersOnLine (2021).

https://doi.org/https://doi.org/10.1016/j.ifacol.2021.10.156, https://www.

sciencedirect.com/science/article/pii/S2405896321015585

6. de Mel, M.: Enhancing engine eﬃciency and performance through ma-

chine learning - real-time engine tuning (proposal). Realtime Engine Tun-

ing (2024). https://doi.org/https://doi.org/10.1016/j.ifacol.2021.10.156, https://

realtimeenginetuning.web.app/paper.html

7. SAE: Optimization of the air/fuel ratio for improved engine performance and

reduced emissions. SAE Technical Paper Series (1996), https://www.sae.org/

publications/technical-papers/content/961714/

8. Wang, Z., Mae, M., Nishimura, S., Matsuhashi, R.: Vehicular fuel consumption

and co2 emission estimation model integrating novel driving behavior data using

machine learning. Energies (2024). https://doi.org/10.3390/en17061410

9. Wong, P.K., Gao, X.H., Wong, K.I., Vong, C.M.: Eﬃcient point-by-

point engine calibration using machine learning and sequential de-

sign of experiment strategies. Journal of the Franklin Institute (2018).

https://doi.org/https://doi.org/10.1016/j.jfranklin.2017.02.006, https:

//www.sciencedirect.com/science/article/pii/S0016003217300674

10. Wong, P.K., Tam, L., Li, K., Wong, H.: Automotive engine idle speed con-

trol optimization using least squares support vector machine and genetic algo-

rithm. International Journal of Intelligent Computing and Cybernetics (2008).

https://doi.org/10.1108/17563780810919140

11. Xu, M., Wang, K., Li, M., Geng, J., Wu, Y., Liu, J., Song, Z.: An adaptive on-board

real-time model with residual online learning for gas turbine engines using adaptive

memory online sequential extreme learning machine. Aerospace Science and Tech-

nology (2023). https://doi.org/https://doi.org/10.1016/j.ast.2023.108513, https:

//www.sciencedirect.com/science/article/pii/S1270963823004108