University of Thi-Qar Journal for Engineering Sciences

This research investigates the effectiveness of geopolymer adhesive (GPA) as an alternative to epoxy in the near
surface-mounted (NSM) technique for the shear strengthening of reinforced concrete (RC) beams. Five
reinforced concrete beam specimens with dimensions of 1200 mm in length, 120 mm in width, and 160 mm in
height were tested under three-point loading conditions. The specimens included one reference beam specimen
and four strengthened beam specimens. The present paper investigated parameters regarding the types of paste
adhesive (epoxy or geopolymer) and the types of strengthening bar materials. Major performance indicators
include the load that leads to the first appearance of flexural cracks, ultimate load-carrying capacity, deflection,
and crack pattern. According to the results, the epoxy adhesive can increase load capacity to 60.1% compared
to the reference beam specimen. While Geopolymer adhesive can be increase load capacity to 47.7% compared
to the reference beam. Compared with unstrengthened RC beam specimens, the first crack load on strengthened
specimens increased by up to 15.3% and 12.12% when using geopolymer and epoxy adhesive. Using epoxy
adhesive shows a greater increase in deflection, reaching up to 96%, while using geopolymer leads to a
deflection increase of up to 50%. The steel bar with geopolymer exhibits the lowest increase in deflection at
46.43%, indicating enhanced stiffness and better crack control. The feasibility of geopolymer adhesive in
structural strengthening applications can be established due to its adequate bonding strength, excellent thermal
resistance, and environmental sustainability.

Further evidence suggests that smart appliances with several types of controllability are necessary, such as smart
homes. This practical approach will be greatly appreciated by those who cannot access control of devices that
require direct human engagement. This study proposes and constructs a smart home setup based on enhanced
electroencephalography (EEG), which could potentially assist individuals with or without movement problems
to manage gadgets with greater comfort and ease. In this study, motor/imaginary EEG (MI-EEG) datasets were
used. This dataset includes 25 subjects who completed motor and imagery tasks. EEG signals were captured
using a 64-channel (BCI2000) instrument. The features are extracted from the CNN's convolutional layers, which
are then used to train the classifier. Features are extracted from both the test and training datasets, and the
labels are flattened to match the expected input format for the classifier. In addition, the proposed Support
Vector Machine (SVM), Convolutional Neural Network (CNN), and Linear Discriminant Analysis (LDA) models can
remove the excessive noise and overlaps that lead to misclassification and classify the chosen features into
separate classes. Based on the number of classes reached by Arduino UNO, the identified classes are tasked with
controlling the smart home. The high overall accuracies using SVM, CNN, and LDA are (99%, 96%, 99%),
respectively. According to the current research, the suggested approach for classifying MI-EEG is successful. It
may be suitable for BCI applications, allowing for controlling a smart home via brain waves. Results from this
study support the idea that the proposed EEG-based smart home could be helpful for older people and others
with mobility issues in the future.

Unmanned Aerial Vehicles (UAVs), or commonly known as drones, have been utilized in numerous applications,
including civilian and military environments, due to their low cost and high level of usability. These devices have
triggered significant security and protection issues, demanding the development of reliable detection and
classification systems. However, UAVs have been used to achieve illegal operations, resulting in emerging threats to
personal and public security. The traditional detection method suffers from small targets, uncertain situations, and
stealthy cases. Consequently, due to their accuracy and efficacy in detecting drones in a diverse range of scenarios,
Artificial Intelligence (AI) has become a significant solution for drone detection and classification. In this paper, an
audio-based mechanism has been developed using a convolutional recurrent neural network (CRNN) algorithm to
exploit the exclusive audio fingerprint of drones for detecting and identifying them. The model achieved 94.85%
accuracy with a 94.41% F1-score from different types of drone sound files collected from the DADS site. It operates
with a three-seed and thresholds to guarantee the stability and dependability of the model. This paper aims to
validate the use of this algorithm for drone detection in an actual environment and train the model with different
seeds to measure the model's stability against random variation.

Understanding the factors that affect induction motor performance is key to improving energy efficiency and
reliability in industrial use. This study examines how switching frequency, modulation index, and load types (fan,
pump, and standard) influence motor performance. We applied sinusoidal pulse width modulation to control
the DC-AC converter in MATLAB/Simulink. Fan and pump loads follow torque that rises with the square of speed.
They bring special challenges unlike standard loads. These include higher aerodynamic, hydraulic, and friction
losses. Results show that switching frequencies above 1 kHz enhance voltage waveforms. They lead to peak
motor efficiency of 80% and cut total harmonic distortion to 21%. Modulation indices affect performance in both
linear and non-linear ways. Efficiency rises steadily to 77% as the index reaches 0.8. Voltage also grows
consistently in under and over modulation regions. Yet current, input power, speed, and harmonic distortion
vary non-linearly during overmodulation. The lowest total harmonic distortion occurs at index 1. Fan loads show
slightly higher losses, up 7.34%, and lower efficiency, down 1.57%, compared to standard loads. Pump loads,
however, feature much lower losses, down 37%, alongside reduced efficiency of 5.8%.

As the growth of the requirement towards Sixth Generation 6G desirable for advanced technologies to offer
extreme performance and facilities, this makes Visible Light Communication (VLC) one of the greatest nominees.
This paper estimates the performance of VLC in line-of-sight (LOS) and Non-line-of-sight (NLOS) environments.
Investigate the effect of distance and various transmission angle adjustments on the received power and signal
to-noise ratio (SNR) from the transmitter side. Furthermore, from the receiver side, the impact of the optical
concentrator gain with distance is studied. The contour plot for both the received power and SNR is studied to
present the maximum power according to the effect of the considered parameters. The results show that a high
amount of the received power can be achieved with high values of irradiance angle, especially in small distances.
For SNR, the maximum value is evaluated at low values for both distance and irradiance angle under the effect
of high Lambertian order in the LOS case. In NLOS, a general reduction is notable, especially at large distances.
From the receiver perspective, the results show that large values for the received power and SNR are evaluated
as the values of the concentrator gain are increased. Also, the optical filter influence on received power and SNR
has been studied.

This paper presents the SH_ArabicIraqiAccent corpus, a newly developed speech corpus that captures
spontaneous Iraqi Arabic from the northern suburbs of Baghdad. This corpus comprises spontaneous speech
from 47 speakers (28 females aged 9–59 and 19 males aged 20–63) discussing both personal and general
community topics relevant to the region. The recordings were made in a controlled environment at a sampling
rate of 16 kHz, with 16-bit resolution and a mono channel, to ensure clarity and fidelity. The total duration is
approximately 2.5 hours and is professionally recorded with minimal background noise. The corpus addresses
the need for dialect-specific resources, as Modern Standard Arabic corpora do not adequately represent regional
spoken varieties. The dataset was carefully designed to address key parameters, including speaker diversity (age
and gender), phoneme coverage, and recording quality, and supports applications in natural language processing
(NLP), automatic speech recognition (ASR), text-to-speech (TTS), and speaker identification. The corpus
complements existing resources by focusing on underrepresented Iraqi Arabic dialects and offers detailed
acoustic analyses, including fundamental frequency and intensity measurements. This work situates the
SH_ArabicIraqiAccent alongside other primary Arabic linguistic resources and highlights its potential to enhance
understanding of Arabic dialects and to support model development. The corpus and relevant statistical analyses
are publicly available and contribute to Arabic speech and language research, particularly for dialectal NLP tasks.