Dijlah Journal of Engineering Sciences

This research aims to investigate the enhancement of thermal-to
electricity conversion efficiency using integrated heat exchange
technologies in small-scale systems, in light of the growing global
interest in sustainable energy sources and their role in powering
Internet of Things (IoT) devices and self-sustained systems. The
study reviews the theoretical foundations of thermal energy
conversion through thermoelectric and pyroelectric nanogenerators,
while analyzing the contribution of integrated heat exchangers in
improving system efficiency and reducing thermal losses.
Furthermore, it highlights recent models of miniaturized heat
exchangers and performance optimization strategies based on
advanced materials and innovative structural designs. The findings
suggest that combining nanotechnology with integrated heat
exchange significantly enhances thermal-to-electric conversion
efficiency, paving the way for practical applications in smart
sensing, wearable devices, and low-power embedded systems.

This research discusses the behavior of footings with inner diameter
to outer diameter ratios of (Dinner/Douter= 0, 0.2, 0.3, and 0.4)
constructed on sandy soil taken from Tikrit. The sandy soil with a
relative density of 50% was treated with kerosene according to
footing depths with footing diameter to outer diameter ratios of
(Dfooting/Douter= 0.5Douter, 1Douter, and 1.5Douter) for time durations (2,
5.3, and 10.28) minutes. The times were determined according to a
preliminary test to determine the penetration depth of kerosene in
soil per minute. The tests were conducted on dry sandy soil
(Dfooting/Douter= 0) as a reference for comparison with the kerosene
treated sandy soil (Dfooting/Douter= 0.5Douter, 1Douter, and 1.5Douter).
Sixteen tests were conducted using a steel test setup of dimensions
900×900×600 mm. The results showed that the bearing capacity of
the ring footings was better than that of the circular footings. In the
sandy soil kerosene-treated conditions, there were better results
compared to the dry sandy soils. For both dry and kerosene-treated
soil conditions, the footing with (Dinner/Douter= 0.4) achieved the
ultimate bearing capacity. Due to the influence of kerosene on
increasing the cohesion of the soil particles, the best depth for all the
footings was (Dfooting/Douter= 1Douter).

We suggest a new method of enhancing the feature discriminability of
mammogram analysis through integrating anatomical pretext tasks in a
hybrid CNN-ViT backbone followed by improved classification using SVM.
Traditional self-supervised approaches usually use generic image transforms,
which does not necessarily reflect the subtle clinical reasoning regarding
radiologists. To tackle this, we formulate domain-specific pretext tasks,
which directly model anatomical priors, such as spatial context
reconstruction, orientation prediction conditional on ductal tree alignment
and lesion-context consistency using contrastive learning. The hybrid
backbone is a hybrid of ResNet-50 and ViT, which extracts local and global
context patterns, respectively, to produce an integrated feature representation
that links hierarchical and long-range dependencies. These characteristics
are then trimmed down using PCA, so that they can be compatible with
SVM kernels whilst retaining anatomical significance. In comparison to the
current approaches, our framework is the only one to use the pretraining
goals with clinical workflows, thus enhancing the interpretability of features
and minimizing the need of large labeled datasets. Experiments indicate that
the suggested approach performs better than the conventional deep feature
extractors in mammogram classification tasks. Integrating domain-sensitive
medical image analysis with self-directed learning through incorporating
anatomical reasoning provides new opportunities to analyze medical images,
especially in cases when few annotations are available. The contribution of
the paper is the progression of the synergy between radiologist-inspired
feature learning and up-to-date deep architectures, which gives a scalable
algorithm that enhances the accuracy of diagnostic outcomes.

This study presents a robust hybrid big data architecture that integrates both
relational (SQL) and non-relational (NoSQL) database systems to effectively
manage and classify diverse electric vehicle (EV) data. The suggested
method makes use of each database type's advantages enables efficient
storage, retrieval, and processing of heterogeneous datasets. The data was
preprocessed and evaluated using data mining techniques. followed by the
implementation of multiple machine learning techniques, such as AdaBoost,
Random Forest, K-Nearest Neighbours, Decision Tree, Support Vector
Machine, and Logistic Regression —for accurate EV type classification.
Among these, Random Forest demonstrated superior execution with a
precision of 99.99%. The trained model was deployed through a real-time,
user-friendly web interface to facilitate practical application and
accessibility. This approach highlights the advantages of hybrid data
architectures and machine learning integration, providing a scalable and
adaptable framework applicable to other domains requiring heterogeneous
data management and intelligent classification.

Handwritten Text Recognition (HTR) in low-resource languages, such as
Hungarian, faces persistent challenges due to the limited availability of high
quality datasets. This paper introduces the HuHTR-Toolkit, an open-source
framework for generating and augmenting Hungarian HTR datasets. The
toolkit generates realistic handwritten text images from open-source corpora,
enabling the creation of large-scale datasets and offering a wide range of
augmentation techniques including mathematical, chromatic, and style-based
transformations to enhance model robustness. Using the Belval HuHTR
Toolkit, we produced over three million synthetic text-image line pairs and
evaluated their impact on transformer-based models. Data augmentation, the
second approach to expanding data, applies image-processing techniques.
Experimental results indicate that synthetic data with touch augmentation
greatly improves character and word-level accuracy, reducing reliance on
costly human-annotated datasets. The toolkit and generated datasets are
publicly available to support further research in low-resource HTR and
cultural heritage digitization.

Background: The non-linear convection-diffusion equation is one of the cornerstones
of computational fluid dynamics and applied mathematics, particularly in applications
such as environmental modeling, heat transfer or reactive flows. The finite volume
method (FVM) is known for its good conservation property and robustness, but
usually has difficulty in accurately simulating the sharp gradients and discontinuities
without high computational cost. PINNs also provide a mesh-free data-driven
technique however it may suffer from the slow convergence rate and poor
generalization. To address these shortcomings, in this manuscript we introduce
Learning-Enhanced Finite Volume Methods (LE-FVM), a new class of approach that
bridges the scientific computation and machine learning communities using advanced
deep learning models — including PINNs, FVGNs, and KANs — into an FVM
framework.
Methods: We have developed the physics-constrained hybridization of machine
learning models assigned directly within a FVM context to improve flux
determination, support in situ adaptive mesh refinement optimization and end-to-end
solvers constrained by integral conservation laws. Important contributions include
twice-message aggregation in FVGN for irregular mesh, and adaptive loss weighting
using the Neural Tangent Kernel to achieve a good balance between PDE residuals
and boundary conditions. The methodology was extensively tested on standard 1D and
2D nonlinear convection-diffusion-reaction test-cases (Burgers’, Fisher’s, Burgers
Huxley, Newell–Whitehead–Segel) employing the metrics of error norms (L₁, L₂,
L∞), statistical tests (Wilcoxon signed-rank coefficient variation) and generalization
tests on unseen geometries.
Results: The LE-FVM framework achieved a 95%+ success rate in delivering high
fidelity solutions across all benchmarks. For the 1D Burgers’ equation at Re=1, PINN
based LE-FVM reduced the maximum absolute error by a factor of 173 and the RMSE
by 60x compared to GFEM. In 2D unsteady flow simulations, FVGN demonstrated a
77% improvement in prediction accuracy for velocity fields and a 56% reduction in
training time compared to purely data-driven graph networks. The framework also
exhibited superior generalization, accurately predicting flow around unseen elliptical
and airfoil geometries without retraining. Mean residual errors for LE-FVM solutions
were consistently on the order of 10⁻⁴ to 10⁻⁵, outperforming traditional methods by
one to two orders of magnitude.
Conclusions: The proposed LE-FVM framework successfully bridges the gap
between physical fidelity and data-driven adaptability, offering a robust, efficient, and
highly accurate solver for complex nonlinear systems. By embedding FVM’s
conservation principles into the loss functions and architectures of modern neural
networks, LE-FVM ensures physically consistent solutions while dramatically
accelerating computation for parametric studies and real-time applications.

Emotion and feeling recognition by analyzing Electroencephalogram (EEG)
signal consider a huge leap in the development in the aspects of human
computer interaction and affective computing. This paper suggests an
innovative hybrid method that combines Residual Network (ResNet50) and
Long Short-Term Memory (LSTM) deep learning architectures for emotion
classification from EEG signals. Our methodology utilizes the spatial feature
extraction by ResNet50 with the temporal sequence modeling strengths of
LSTM networks. The proposed method was tested and validated on a
complete dataset consists of EEG recordings from 28 subjects during
emotionally-exciting computer game sessions, achieving a very well
performance metrics that reaching approximately 0.99. Comparative study
shows and illustrated that our framework or approach outperforms individual
models or networks, considered as a new ideal system for EEG-based
emotion recognition. This research consider a noticeable achievement add to
the aspects of affective computing and the related aspect such as human
computer interaction and many other potential applications.

Electroencephalography (EEG) signals are complicated and noisy, but the
are a useful tool for monitoring brain activity and detecting disorders. This
paper examines how we can use hybrid deep learning models to boost the
performance of EEG tasks by enhancing the signal and extracting
informative features. We start by talking about the difficulties with raw EEG
data (such as blinking eye or muscle movement artefacts) and the
requirement for substantial improvement. Next, we describe how hybrid
deep learning architectures, such as recurrent networks or convolutional
neural networks with transformers, may better extract significant features
from EEG data than conventional techniques. We highlight evidence from
recent studies that applied these hybrid models to critical clinical areas:
detecting epileptic seizures, identifying mental health conditions like
depression, aiding motor-impaired patients via brain–computer interfaces,
and monitoring emotional states. The results from prior research show
significantly improved accuracy and reliability – for instance, hybrid models
achieving above 95–98% accuracy in certain tasks – which marks a
substantial improvement over earlier approaches.

Arabic language processing with artificial intelligence has evolved
significantly in the past decades, from traditional rule- and dictionary-based
techniques, through statistical models to modern deep and transformer
models. This review intends to present an overview of the most well-known
Arabic models as well as datasets used for its training, and the main practical
applications such as sentiment analysis, machine translation, speech
recognition, and smart assistant.AI-based Arabic NLP has had good progress
in the previous decades, from rule and dictionary-based approaches to
statistical methods and deep transformative learning models nowadays. In
addition to it, the most popular state-of-the-art models that are fine-tuned for
the Arabic language and their corpora of training data will be considered as
well as major applications such as Sentiment Analysis, Machine Translation,
Speech Recognition, and Virtual Assistant.This article review outlines the
necessity for investment in language resources and advanced models to
improve AI systems’ ability to accurately understand Arabic natural
language, a contribution that will support real-life applications and smart
services associated with its present formalized variant of AI model
capabilities.

This study investigates the experimental effect of openings and varying steel
section heights on the structural ‎behavior of concrete-filled cold-formed U
shaped composite beams. Four specimens were fabricated and ‎tested under
four-point loading: one reference beam without openings (CB1) and three
with circular openings ‎at different section heights (300, 350, and 400 mm).
The results showed a 16% reduction in the ultimate ‎flexural strength with
openings compared to the reference beam, with a significant increase in
deflection and ‎slippage due to the loss of effective section area and the
formation of stress concentrations around the ‎openings. Increasing section
height, on the other hand, gradually improved strength and stiffness, with
the ‎ultimate load reaching 305 kN at a height of 400 mm, indicating the
ability of the larger section to ‎redistribute stresses and reduce slippage. The
specimens also demonstrated a transition in failure mode from ‎localized
failure around the openings to a more uniform flexural failure at higher
heights. These results ‎emphasize the importance of incorporating geometric
considerations of openings and height into the design ‎to ensure efficient
performance and safety of modern composite beams.‎

In many industries today, including healthcare monitoring, environmental
monitoring, and intelligent infrastructure, wireless sensor networks (WSNs)
serve as a crucial base for cutting-edge applications. However, their
openness and lack of resources make them especially susceptible to security
risks like illegal access, data manipulation, and eavesdropping. In these
networks, trust in systems that frequently work in highly regulated and
mission-critical contexts depends on safe data transmission, which is more
than just a technological need. The hybrid encryption scheme, which
combines the speed of symmetric cryptography with the security of
asymmetric methods, is the subject of this study. The suggested model seeks
to strike a reasonable balance between optimum security and maximum
resource usage by taking advantage of the strengths of both. The study
examines the impact of hybrid encryption on data integrity, transmission
latency, and energy efficiency in WSNs through simulation and performance
analysis. The main goal is to develop a solution that is lightweight and safe,
enabling wireless sensor networks to run continuously even in harsh real
world environments.

This paper describes an applied–experimental study on smart suspension for
small passenger cars, using micro-electromechanical systems (MEMS)
accelerometers to provide real-time adaptive damping. The system combines
the multi-axis vibration sensing with digital signal processing and semi
active MR dampers, for active control of ride comfort, road handling, and
tire–road contact stability. A small hatchback platform was equipped with an
in-house developed sensor actuator network and driven over specifically
prepared road excitations, such as fine asphalt, speed bump or rough surface
at velocities between 20 and 60 km/h at different loading conditions.
Performance was assessed in terms of ISO 2631-1 ride comfort indices,
damping ratios, suspension travel and dynamic tire load fluctuation. The
proposed system has shown that RMS body acceleration is reduced by up to
33% while dynamic tire load variation is reduced by 30–40 %, compared
with the conventional passive suspension. This paper also investigates the
practical problems of sensor preciseness, processing latency and system
integration on an automotive platform with limited resources. The insights
gained here yield practical guidance for the implementation of smart
suspensions on electrically powered and/or self-driving city vehicles, which
must overcome stringent space, weight and energy efficiency design
constraints.

High-power Silicon Carbide (SiC) converters are fundamental to modern
energy systems, such as HVDC transmission, but their operational reliability
remains a critical challenge, this research presents an integrated
computational framework designed to holistically address both the reliability
assessment and fault-tolerant operation of these systems, the methodology
bridges the gap between micro-level material degradation and macro-level
system dynamics through a sequential, two-phase approach implemented
entirely in Python, the first phase develops a high-fidelity predictive
reliability model based on Physics-of-Failure (PoF) principles, meticulously
modeling the dominant failure mechanisms of Time-Dependent Dielectric
Breakdown (TDDB) in the gate oxide and thermo-mechanical fatigue in
bond wires, this is achieved by linking an electro-thermal simulation, which
calculates device stresses from a given mission profile, with a Monte Carlo
analysis to generate probabilistic lifetime predictions, the second phase
focuses on designing a robust Fault-Tolerant Control (FTC) strategy for a
Modular Multilevel Converter (MMC), this includes non-invasive
algorithms for rapid fault diagnosis (open- and short-circuit), immediate
isolation of faulty submodules, and a dynamic compensatory control
algorithm that rebalances arm voltages and capacitor states to ensure
seamless post-fault operation, the primary contribution of this work is the
creation of a unified simulation tool that models the entire converter
lifecycle, from gradual physical degradation to adaptive system response,
providing crucial insights for designing more resilient and dependable power
electronic systems.

Autonomous vehicles (AVs) are a significant advancement in transportation
technology, offering greater safety, performance and comfort. Traditional
cyber-security solutions frequently fail to defend antivirus software from
sophisticated assaults that exploit software and communication network
flaws. This study proposes using Support Vector Machine Fused Defense
Generative Adversarial-Network (SVM-DGAN) to improve cyber-security
in autonomous vehicles. In cyber-security, SVMs can be employed for threat
intelligence by analyzing data to identify patterns and classify data into
different categories, such as normal behaviour or anomalous activities
indicating potential cyber threats. Defense GANs are employed for adaptive
defense mechanisms to enhance the resilience of machine learning models
against adversarial attacks in autonomous vehicle cyber-security. Initially we
gather dataset from controller area networks (CAN). Pre-process the
gathered data using minimum-maximum normalization. The following
elements were used to compare the traditional and proposed methods in
terms of accuracy (98.50%), precision (98%), recall (98.25%), and F1-score
(97%). It shows that our proposed method is effective in autonomous vehicle
cyber-security. The findings of our study affirm the viability and
effectiveness of the SVM-DGAN framework as a formidable defence
against cyber threats targeting autonomous vehicles.

Cybercrime and attacks due to brute force have increased in the last decades.
Old-fashioned detection methods have become useless and unreliable.
Researchers are now working on new ideas and ways to improve
information security and privacy. One prominent strategy involves machine
learning. The aim of this master’s research is to assess how machine learning
significantly impacts identifying severe cyber threats due to brute force
attack. Thus, analysis was performed on the CICIDS2017 datasets.
Furthermore, we used numerical analyses with Python programming to
verify that machine learning enables enhanced performance and reliability in
network threat detection. The computational simulations and theoretical
evaluations indicated that the voting algorithm exhibited superior
performance (in terms of accuracy, precision, recall, and F1-score) in
identifying benign brute force assaults. Moreover, the voting algorithm
demonstrated peak accuracy across three distinct categories of attacks. The
Gradient Boosting (GB) model demonstrated the highest F1-score in relation
to FTP and SSH brute force vulnerabilities. The peak accuracy levels,
precision, recall, and F1-score recorded in the three experimental conditions
were 99.1%, 75.4%, 59.6%, and 70.1% for the voting, second voting,
decision tree (DT), and gradient boosting (GB) methodologies, respectively.

weakly singular kernels in Volterra-type integral equations appear in
viscoelasticity applications and anomalous diffusion, and in fractional-order
control. When the kernel has weak singularities, particularly near the initial
time, numerical solvers usually become less accurate and less stable. The
paper presents a unified framework known as Weak Singularity Modulation,
which enhances classical methods such as convolution quadrature, product
integration, and collocation. The system design uses steady-state weights,
compact start-up correctors, time mesh that grows in size to get convergence
stable. Through thorough investigation, it is shown that WSM recovers the
optimal accuracy predicted by local asymptotic while being robust to both
linear and nonlinear problems. Examples with solutions that are known show
how fast the method improves the solution. The presented methodology
makes little to no impact on computational effort. Additionally, it easily
integrates with developed or existing solvers. Thus, large-scale simulations
are possible in a fractional modeling of practical applications.

The Kigali Amendment is leading to a rapid global phase-down of high
GWP refrigerants, which will require the thoughtful redesign of vapor
compression systems to achieve the same performance while mitigating
environmental impact. In this work, we developed a hybrid mathematical
and computational framework that includes thermodynamic modeling,
surrogate-based forward simulation, and multi-objective Pareto optimization
to study and compare refrigerants R-410A, R-32, R-1234ze(E), and R-717.
We used synthetic yet physically-consistent datasets to predict both
coefficient of performance (COP) and total equivalent warming impact
(TEWI) under reasonable operating conditions and real-world scenarios. We
found that while low-GWP fluids provide a path to low-impact thermal
mechanical systems, there are fundamental trade-offs associated with
efficiency and climate impact that need to be considered when transitioning
from high-GWP fluids. Based on optimization, we recommend R-1234ze(E)
and R-717 as the two best low-GWP fluids with the potential of continued
use over the long-term to support regulatory compliance, while R-32
continues to be efficient switching fluid during transitions. The framework
gives a scalable and reproducible guidance tool to assist in refrigerant
selection in line with the urgency of meeting Kigali phase-down targets, and
in the context of wider objectives for both sustainable energy and carbon
mitigation.

A key component of the workable Machine learning (ML) is the solution for
physical object systems. which can be characterized as the ability of an
intelligent smart device to alter its state and behavior. Both human and
machine-generated data can contain useful information that machine
learning (ML) can comprehend. Machine learning algorithms can be applied
to a variety of tasks, including regression and classification. Additionally, in
an Internet of Things network, security services are provided by Machine
Learning (ML). The application of machine learning (ML) in cyberattack
detection challenges is becoming more and more popular. ML may be
applied to a variety of cyber-security prosecutions. Assuming that the
summary's various study types were combined with Machine Learning (ML)
techniques to determine the most accurate ways to detect offensive,
individual poor groundwork exists on effective detection techniques
advisable for practical physical object system environments

Digital technologies, particularly big data analytics and artificial intelligence
(AI), are transforming the operation of analytical chemistry. The literature
review explores the effect in terms of such novel tools on the current
chemical analysis from enhancing the accuracy of analysis, data
interpretation, to decision-making in real time. The systematic review
examines the peer-reviewed literature published in the period from 2015
2025, and the focus of the application is on AI as well as machine learning
technologies in chromatographic techniques, spectroscopic analysis,
environmental monitoring as well as process control. The emphasis is on
predictive modelling, pattern recognition and systems for automatic
processing to cope with increasing data volume and complexity in analytical
data. The review focuses on modern issues, such as issues of model
interpretability and data standardization and the need for interdisciplinary
know-how. The research work provides a sign of potential of digital
technology in enhancing analytical chemistry's speed and reliability, as well
as the breadth of the analytical chemistry discipline and research area in
terms of future trends and developments. Lessons learned is important for
scholars, researchers, and institutions interested in using next generation
analytic methods in both academia and business.

The problem is the design of an adaptive, distributed anomaly detection
system which protects privacy, and has the ability to evolve in real-time to
identify and mitigate new threats while minimizing maximum detection
delay and damage, in very large distributed environments. In this work, we
propose an A-FCL (Adaptive Federated Collaborative Learning) architecture
to enhance the separability of unknown anomaly patterns, adopt meta
learning approach (Meta-UAD) to decrease the false positive rate (FPR) and
enhance adaptability, and develop a fully automated response architecture
based on SOAR/SDN together with RL-AR (Reinforcement Learning with
Adaptive Regulation) frameworks to dynamically and securely adapt low
latency response strategies. The relevance of this work is that it is the first to
confront most challenging cybersecurity problems of the digital age and
deliver an extremely high throughput autonomous security capability in
large-scale environments slashing response times from minutes to
milliseconds. The research methodology employed analytical, operational,
and simulation methods, in which an advanced computer simulation
environment was developed to evaluate the A-FCL framework —comprising
contrastive learning as well as trust & reputation schemes to mitigate
poisoning attacks—on extensive, heterogeneous (Non-IID) network datasets.
The result was that the A-FCL model outperformed all the other models with
an accuracy of 98.7% and the FPR was greatly reduced to 0.5% (a reduction
of about 81.25%) by applying adaptive learning schemes. The system also
exhibited excellent scalability, achieving a processing throughput of 22.1
GB/s for 50 edge nodes and an automated response time to critical attacks
(such as DDoS) of no more than 45 milliseconds, representing a 96.6%
reduction compared to manual response, while ensuring operational safety
by reducing the safety violation rate by RL-AR to less than 0.4%.

This research work does a comprehensive nonlinear finite element
simulation of thin elastic shells when subject to a localized loading and
explains the role of geometric nonlinearity in large deformation. The
researchers analyzed spherical and cylindrical shell models to investigate
differences in stiffness, stress concentration, and post-buckling response.
The model used the second Piola-Kirchhoff stress and Green-Lagrange
strain tensor to account for large strain effects in an updated Lagrangian
context. The results showed that the spherical shells exhibit sharp snap
through instability, while the cylindrical shells behave with gradual
reduction in stiffness with increasing indentation. Load-displacement and
strain energy analyses clearly revealed the switch from bending-dominated
deformation to membrane-dominated deformation. According to the study,
linear theories will not be able to adequately predict the responses of shells
to concentrated loads. Thus, nonlinear finite element analysis is necessary.
This is essential for assessing the stability and performance of advanced
structural applications.

This study investigates three types of porous media to improve heat
exchanger performance: copper foam, ceramic foam, and a third medium
that was a composite medium. The results showed that ceramic foam
provided low pressure loss and a slight improvement in heat transfer, while
copper foam achieved a greater increase in heat transfer, about 50%. The
composite medium showed balanced performance with a PEC value of about
1.5, making it the optimal choice for low-power applications. Numerical
simulation of the experimental results confirmed that porous media improve
heat exchanger efficiency without causing a significant increase in energy
consumption.

The increasing complexity of cyber threats has underscored the constraints
of conventional Intrusion Detection Systems (IDS) and the need to have
more flexible and intelligent security systems. The growth of artificial
intelligence (AI) has made automated threat response, real-time analysis, and
learning in the field of intrusion detection a focus. The review paper related
to AI-driven IDS published between 2023 and 2025 have been examined,
and efforts have focused on the hybrid metaheuristic model, machine
learning, and deep learning. Well-known datasets, including CICIDS2017,
CICIDS2018, UNSW-NB15, and NSL-KDD, have been tested on their
appropriateness in measuring the performance of a system. It is reported that
hybrid deep learning metaheuristic structures have a higher detection
efficiency and flexibility than fifteen modern models, but these are reported
to be highly computational. The future research direction is presented,
including the explanation of AI to enhance transparency, the creation of
lightweight IDS that can be implemented in the conditions of the IoT, and
the enhancement of adversarial attack resistance. Overall, AI-based IDS may
be deemed as a serious breakthrough in intelligent, scalable, and resilient
network security.

This study deals with the challenge of developing highly dependable self
developing software (SDS) operating in full autonomy in complex and
dynamic work environments with a strict adherence to ethical and legal
responsibility considerations. The significance of this is that it addresses a
fundamental problem in both the software engineering and artificial
intelligence domains and brings understanding of social and technical issues
to a quantitative level. The research hypotheses were formed to validate RL
models as the more adaptable and powerful decision making mechanism in
SE and that use of AI results in higher cycloramic complexity and security
verification efforts, and that presentation of XAI boosts trust and decreases
instances of contestation of automated outputs, and that AI-based self
healing solutions reduce unplanned downtime by at least 25%. The study is
rooted in a meta-methodology of critical interpretivist and quantitative
research and employs advanced statistical models including multiple
regression, z-testing, and Markov Decision Processes (MDPs) for data
analysis. The results verified that the model of software self-development
was mainly led by artificial intelligence, and the automated testing and
verification tools had substantial operational improvements, the reliability of
test was improved by 85% and the coverage of test was increased by 95%.
Predictive self-healing systems also reduced unintended downtime by
30.0% and increased mean time between failures (MTBF) by 35.6%,
resulting in a 18.6% saving in annual maintenance costs. However, the
research identified deep socio-technical problems: automatically the
churned code has 40% more cycloramic complexity and 37.1% more
security vulnerabilities than human-written code. It also revealed a
“productivity paradox”: developers decelerate by 16% on certain tasks even
though they believe their productivity is increasing by 20%. Lastly, the
findings demonstrated that XAI increased trust and reduced security
breaches, and that rejection of self-modifying decisions, predicated on
opaqueness, is a significant impediment to full automation with a propensity
score of 0.79.

Background: According to American Cancer Society, brain tumor represents
a highly feared disease in medical science, and it is among the most
prevalent malignant tumors to be detected in people of different ages.
Treatment of brain tumors requires early and accurate detection of the
disease. Methodology: With using machine learning and deep
features classifiers, this paper describes the classification of brain tumors.
Using a set of pre-trained deep convolutional neural networks (DCNNs),
deep features were extracted from a collection of brain magnetic resonance
imaging (MR) images by incorporating the concept of translational learning
(TL) into our proposed framework. After that, a number of machine learning
(ML) classifiers assess the deep features that were recovered. Deep features
are used for defining and sequencing several ML feature classifiers. For the
prediction of final output, these are after that input into a number of
ML classifiers. In this study, Three publicly available brain MRI datasets
were used to evaluate the effectiveness of different types of pre-trained
models, including machine learning classifiers and deep feature extractors,
in classifying brain tumors.Results: The experimental results show that deep
feature extractors significantly improved performance levels, and in most
cases, a convolutional neural network (CNN) was used. This study achieved
an accuracy of 0.98% and an accuracy of 0.99% for the valves.

Makes use of representative numerical computation to proposal a
comprehensive performance examination perovskite solar cells (PSC) of L
cost produced at different temperatures for the study. The effect of
dispensation heat on device specifications was investigation Through the
creation and check of multiple sample. Optical characteristics were
described using UV–Vis spectroscopy, morphological traits were described
using scanning electron microscopy (SEM), and electrical performance was
described using current–voltage (I–V) measurements. The results
demonstrate that temperature affects the optical bandgap, grain structure,
and power conversion efficiency (PCE). The measured bandgap varied
around 1.55 eV, and the average grain diameter was approximately 210 nm.
The PCE increased to 18.3% under typical AM1.5G light and optimal
temperature conditions. Comparative benchmarking with earlier PSC
research directs that low-cost fabrication techniques can create high
performance devices while lowering material and production costs. [1].

The growing use of Unmanned Aerial Vehicles (UAVs) for things like
surveillance and disaster response makes it harder to meet strict real-time
performance standards. To solve this problem, the proposed research
combines Mobile Edge Computing (MEC), Network Slicing (NS), and a
lightweight Latency-Aware Deep Deterministic Policy Gradient (LLDDPG)
algorithm to make task offloading more efficient and lower system latency.
We divided UAV image data into three groups based on file size and sent
them to different MEC nodes. The first tests with static and heuristic load
balancing strategies showed some improvement. With LLDDPG training,
the time it took to process improved from (569, 119, 99,222) seconds to
about (425, 320, 3,050) seconds.

Dry reformation of methane (DRM), which uses two greenhouse gases, CO₂
and CH₄, at the same time, is a sustainable method of producing syngas.
DRM is hampered by catalyst deactivation brought on by coking, sintering,
and structural instability under extreme operating conditions, despite its
potential for the environment and industry. Recent developments demonstrate
how well cesium (Cs) works as a promoter in nanostructured catalysts, where
its strong basicity, large ionic radius, and electron-donating capacity improve
CO₂ activation, suppress coke formation, and stabilize active metal sites.
With a focus on their electrical and structural impacts, catalyst–reactant
interactions, thermodynamic and kinetic functions, and the impact of various
supports, this review synthesizes the state of knowledge about Cs-promoted
nanostructured catalysts for DRM. Stability trends, structure-performance
correlations, and Cs loading optimization for increased activity and durability
are all given special consideration. Complemented by a section on generic,
overarching concepts such as combined catalytic- and electrochemical
processes, reactor design driven by simulation and integrated reactor
concepts including hybrid applications, the authors discuss new strategies to
address long-term durability, impurity management, feedstock tolerance and
upscaling of power-to-X products. Overall, this investigation provides key
new insights into the special promotion role of cesium in preparing
nanostructured DRM catalysts for efficient and durable syngas production.

The paper discusses the potential of solar energy as a renewable form of
power in hydronic environments with sunlight. A techno-economic study
was conducted on whether solar thermal collectors could sustainably and
effectively
operate
small-scale
experimental models of thermal
desalinization conceived by the very researchers who developed them.
Besides temperature, irradiance and heat flow; thermal efficiency as
monitored from the temperature of the panel while it dissipates heat, type of
fluid used and angle at which the panel is oriented relative to the sun’s path
comprises. Results indicated that under adequate sunlight and favorable
conditions, solar thermal collectors are highly efficient in converting sun’s
rays into usable heat. Technoanalysis combination with economic analysis
helps to develop bigger models applicable for different locations and
enhances understanding about the feasibility of these systems .This research
possesses a significant scientific merits which will attract future solutions
based on solar energy and water having sun as their primary energy source.

Over the last two years, advancements in the security of physical
government-issued identity documents against forgery have been made
through three key research areas: the use of deep learning to detect forgeries,
the development of cryptographic verification systems, and improvements in
the security of QR codes. While advances have been made in these three
areas, the vast majority of these developments happen independently of each
other and create a significant vulnerability to hybrid attacks that use the
disconnect between paper-based and electronic security features. This paper
highlights a missing component to establish a comprehensive solution that
would combine tamper-evident physical security features (such as Guilloche
patterns) with powerful electronic mechanisms (like elliptic curve
cryptography and desirable secure data storage methods, such as QR Codes).
The authors suggest that research should focus on developing unified
multilayered security systems that support the principles of scalability,
interoperability, and usability so that we can effectively combat the growing
threat of sophisticated and frequent forgeries and, in turn, restore trust in
both electronic and physical forms of identification.

A large-scale Multiple-Input Multiple-Output (MIMO) device is a primary
5G and beyond method, which guarantees annotatable advantages in spectral
and power performance. However, their full capacity is an issue to the
challenge of obtaining the exact channel status information (CSI) for a huge
wide variety of antennas. Traditional methods together with the least squares
(LS) and minimum mean square errors (MMSE) are regularly
computationally intensive or require earlier statistical computations and
struggle with the high-dimensional, real-time of today’s Wi-Fi channels.
Over the beyond three years, Machine Learning (ML) and Deep Learning
(DL) have emerged as a powerful paradigm to move those boundaries.
Recent research has proven notable overall performance; the DL model has
carried out a 30% discount in normalized mean square errors (NMSE) as
compared to traditional linear estimates within the surroundings with low
signal-to-noise ratio (SNR). In addition, the time has shown an average
improvement of 15-25% in bit-error-rate (BER), time monitoring for the
hybrid DL structure, recurrent neural networks (RNNs) or conventional
neural networks (CNN). These advances are improved using the ML
architecture, which is intended to investigate complex channel structures and
reduce problems that cause pilot pollution, leading to a more scalable,
efficient, and powerful system.

Radar systems are used in many applications such as navigation,
surveillance and weather observation. Nonetheless, noise used in radar
signals is also a giant drawback, which makes target detection less precise
and reliable. Improving detectability of signals in noisy environments is
hence an important radar issue in signal processing. Matched filtering and
cross-correlation detection are two well-known detection algorithms, which
this paper shall investigate and compare in a bid to evaluate their usefulness
in different signal-to-noise ratio (SNR) scenarios. A simulation framework
was produced in Python and designed to simulate the behavior of radar
signals when adopting an additive noise model, and subsequently perform a
systematic performance analysis of both techniques. As an established
technique that maximizes the signal-to-noise-ratio in known-waveform
scenarios, matched filtering was compared to cross-correlation detection,
which evidences computational efficiency and flexibility to any improper
signal correlations. Findings indicate that the Matched Filter is more
successful in detection performance with a 4.97×10⁵ to 5.29×10⁵ detections
and an average first detection time of 1-6 µs, whereas Cross-Correlation
Detection has 4.56×10⁵ to 4.85×10⁵ detections, but with less computing
requirements. The results show that periodic waves (sine and triangle) have
the best noise immunity, and the best clarity of envelope, and amplitude
modulated waveforms give the best trade-off between bandwidth utilization
and information capacity. The hybrid idea, which is amplitude-modulated
and triangular waveforms, has the advantage of providing better noise
resilience and scalability to adaptive radar designs. Theoretically, the
matched filter maximizes convolutional energy to reach optimum SNR and
this is why it performs better in poor detecting situations under noisy
conditions. This study explains the feasible performance limits of the
traditional detection algorithms and forms a pedagogical basis of subsequent
hybrid radar architecture that incorporates the efficiency and the accuracy.

The challenge of providing URLLC (Ultra Reliable Low Latency
Communication) services in 5G networks is that traditional Top-down
Network Management techniques do not allow for URLLC applications
requiring delay latencies below 1 millisecond and higher degrees of
reliability than what is achievable with current Top-down Network
Management techniques. In response, this study offers a novel means for
real-time segmentation of the network through Dynamic Segmentation of the
Network (DSN) utilizing Deep Deterministic Policy Gradient (DDPG)
algorithms utilized in conjunction with software-defined networks (SDN).
Utilization of DSN by means of DDPG allows the controller to learn how to
allocate bandwidth in response to traffic predictions. Results of simulation
experiments demonstrated this technique's ability to provide reliability levels
in excess of 99.999% while maintaining delays less than or equal to 1
millisecond for eMBB (Enhanced Mobile Broadband) style services thus
improving the spectrum efficiency of these services. The results highlight
how AI and the coordination of AI systems are able to meet complex Quality
of Service (QoS) needs of critical applications within 5G networks.

The temporal graph data offers an effective model of complex time
dependent systems across various systems such as traffic networks, brain
connectivity, weather forecasting, epidemiology and physical systems. But
there is still a real challenge of closing the divide between the discrete-time
observations of networks and continuous-time dynamical models. It is a
systematic review of the approaches to extracting effective differential
equations to the data of temporal graphs by means of inverse modeling,
focusing on rigorous mathematical backgrounds. We divide the existing
methods into such families as graph neural differential equations,
continuous-time representation learning, PDE discovery frameworks,
operator learning methods, and physics-informed universal differential
systems. The theoretical matters covered in the review are identifiability,
stability, scalability, and robustness, with practical issues being model
validation and cross-domain applicability. Inverse methods have been
applied to solve problems in traffic, neuroscience, weather, epidemiology,
and groundwater systems to demonstrate how the inventive methods can be
used to reconstruct interpretable and physically significant governing
equations. The synthesis gives a full conceptual and mathematical view of
the temporal graph based inverse modeling and sets the basis of future
studies in data discovery of complex dynamical systems.

The increasing sophistication of cyber threats has exposed critical
vulnerabilities in conventional cryptographic systems, necessitating
innovative and adaptive security measures. In response, a multi-agent
reinforcement learning framework is proposed to optimize cryptographic
operations dynamically through intelligent, adaptive agents. The system
integrates a Defense Agent, which selects appropriate encryption algorithms
and key sizes, and a Key Management Agent, which governs adaptive key
rotation strategies. Additionally, an Attacker Agent is employed to simulate
realistic
adversarial tactics, facilitating a co-evolutionary learning
environment. The approach is grounded in an asynchronous Actor-Critic
architecture, which continuously adjusts policy parameters based on the
advantage function to reinforce effective defense strategies. Experimental
evaluations across escalating threat scenarios reveal a significant
enhancement in the system’s resilience, as demonstrated by prolonged time
to-compromise, reduced damage impact, and a high threat interception rate
while maintaining acceptable resource overhead and encryption latency.
These results underscore the potential of the proposed adaptive framework
to substantially mitigate risks and elevate the robustness of cryptographic
systems, thus providing a promising avenue for next-generation
cybersecurity solutions.

Diabetic retinopathy (DR) is the leading cause of preventable vision loss in
working-age adults, affecting about 103 million people, with prevalence
expected to rise. Current deep learning methods for DR screening face
challenges such as lesion localization uncertainty and limited generalization
across diverse datasets. This study presents GMM-CNN, a novel framework
combining Gaussian Mixture Model (GMM) mechanisms with
convolutional neural networks to overcome these issues. Three synergistic
architectural innovations are introduced: (1) a mixture-inspired spatial
attention module that generates probabilistic attention weights over K=8
component distributions for multi-focal lesion localization; (2) mask-aware
rescaled pooling that prevents feature dilution from sparse pathological
lesions; and (3) a GMM-style dense branch that models multi-modal feature
distributions across DR severity grades. Evaluated on APTOS 2019 and
Messidor-2 benchmark datasets, GMM-CNN achieves 89.27% accuracy
(κ=0.8421) and 88.94% accuracy (κ=0.8389) respectively, surpassing state
of-the-art methods by 1.09–3.44 percentage points while using 2.2× fewer
parameters than ViT-CapsNet. Cross-dataset evaluation confirms robust
generalization across diverse imaging protocols, demonstrating clinical
deployment readiness for automated DR screening.

The building sector often causes significant environmental damage due to the usage of conventional concrete. In addition to this, such a material requires a substantial quantity of natural resources, while simultaneously resulting in significant quantities of carbon dioxide emissions. The recognition of environmental difficulties in building construction projects has contributed to the development of eco-friendly concrete techniques, particularly in the area of recycling. This study primarily focuses on the advancements in sustainable concrete in recent years, specifically on the use of recycled aggregates and industrial leftovers such as fly ash and slag. Additionally, it explores the use of sustainable supplementary materials including recyclables such as plastics and glass. They contribute to reducing the need for new raw materials and divert trash from landfills, therefore promoting the circular economy. The study focuses on the environmental, economic, and technical benefits of using recycled materials into concrete mixtures. Additionally, it addresses concerns related to performance, uniformity, and appearance. The research aims to provide a fundamental understanding of the impact of recycled materials on the concrete industry, therefore promoting sustainability in response to the global need to reduce carbon dioxide emissions and conserve resources. The research is important because users in the concrete industry are looking for methods to reduce carbon dioxide emissions and resource consumption in order to adopt sustainable technology. This procedure will be accomplished by assessing various evaluation methodologies and generating relevant case studies and current developments