DIYALA JOURNAL OF ENGINEERING SCIENCES

Air-cooled condensers are an important part of modern refrigeration systems because they affect how well the system works and how much energy it uses. This study develops an integrated numerical model to simulate the behaviour of an air-cooled condenser using Fortran and RefProp 8.0 to evaluate the thermophysical properties of R22 and R32 refrigerants. The model solves five equations at the same time. These include the Navier-Stokes equations for the main refrigerant\\'s mass, momentum, and energy conservation, the energy balance equation for the condenser wall, and the enthalpy equation for the secondary fluid (air). Due to the nonlinear nature of the system, the full implicit Preismann scheme was applied in conjunction with the Newton-Raphson and bisection methods to improve numerical stability and accelerate convergence. The results demonstrate that the developed model precisely forecasts the steady-state performance of the condenser, offering thorough analyses of the pressure drop and heat transfer coefficients in both the single-phase and two-phase flow regions. The effective condenser length was found to be 20 meters for R22 refrigerant and 48 meters for R32, with the two-phase zone dominating by 96%, indicating that condensation occurs primarily within this section. Furthermore, the refrigerant behaves as an incompressible fluid in the partially cooled liquid zone. Comparison with EVAP-COND 4.0 and previous studies revealed a deviation of less than ±1% in the performance parameters. These results demonstrate the long-term promise of the developed model as an effective tool for designing and optimizing air-cooled condensers for industrial refrigeration and air conditioning applications.

Email phishing and spam pose considerable cybersecurity risks. They require trustworthy, effective, and feasible detection methods. This research work proposes a model-based methodology for e-mail spam detection and phishing is based on artificial intelligence (AI). It works with a binary classification system with two phases. At the first stage, the system classifies email contents into malicious and non-malicious. In the next stage, it scans embedded URLs, which may or may not be phishing hooks. This modular design reduces the complexity of the feature space and enables separate optimizations for the email and the URL analysis. The system is trained with 18650 email samples and 549346 url samples from publicly accessible datasets, with 70% for training and 30% for testing. The preprocessing step consisted in eliminating duplicates and null values, text normalizing, balancing classes, stemming and feature extraction using TF-IDF for email and CountVectorizer for url. Four lightweight ML algorithms were evaluated: Naive Bayes, Decision Tree, Random Forest and K-Nearest Neighbors. The result indicated that the Naive Bayes achieved the highest baseline accuracy of 96% in email classification and 97% in URL classification. Random Forest, on the other hand, was more resilient to adversarial attacks and demonstrated better generalization. The selected model was deployed with Gmail for real time inbox detection with an accuracy of 85% in real world applications. The results demonstrate that by integrating lightweight machine learning, modular design, and relatively clean pre-processing, a new generation of effective, scalable detectors for both phishing and spam e-mail can be constructed.

In engineering components with geometric discontinuities, the fatigue crack growth of structures subjected to cyclic loading is of utmost importance. This study numerically investigates the modified Single Edge Notched Bend (SENB) specimen with a hole adjacent to the crack tip under three-point bending (3SENB) and four-point bending (4SENB) conditions. The Finite Element Method (FEM) was implemented using ANSYS Workbench 19.2. Specialized meshing was performed to determine the stress field and predict the fatigue crack propagation using the Paris’ Law. This study quantitatively compares the effects of these two loading conditions on the fracture mechanics parameters of this non-standard geometry for the first time. The findings indicate that the specimen response is significantly different; the 3SENB configuration, with twice the maximum bending moment of the 4SENB, has 1.931 times the maximum von Mises stress, 1.823 times the maximum von Mises strain, and 1.965 times the stress intensity factor. The main contribution of this analysis is the development of a strong and quantifiable correlation between the maximum bending moment, and the fracture severity as defined by this work. Consequently, the predicted fatigue life of the 3SENB specimen is substantially reduced due to higher stress intensity factors. It has been found that the selection of bending configuration has a decisive and quantifiable influence on fatigue life of the modified SENB specimen with a geometric discontinuity under the specified loading condition.

Currently, meeting grid standards for grid-connected photovoltaic (PV) solar power systems is a major challenge, particularly in terms of energy quality under conditions of non-linear loads, fluctuations in solar radiation, and parameter uncertainties. Conventional strategies based on proportional–integral–derivative (PID) regulators often suffer from limited robustness, higher total harmonic distortion (THD) of current, and noticeable fluctuations in voltage and power. To address these limitations, this work proposes a high-performance control strategy based on a fuzzy fractional-order PID (FFOPID) controller for a grid-connected PV system with efficient energy filtering capability. The proposed approach combines the robustness of fuzzy logic with the flexibility and memory characteristics of fractional-order control to regulate the DC-link voltage, ensure a unity power factor, reduce THD, and enhance dynamic performance. A systematic design methodology is developed to determine and optimize the FFOPID parameters. The studied system includes a PV array, a two-level inverter using space vector modulation, an inductive filter, a nonlinear load, and the utility grid. The proposed method is validated through MATLAB simulations and compared with the PI approach. Results demonstrate that the FFOPID significantly improves the overall system performance. In particular, the THD of the grid current is reduced from 4.05% with the PI to 0.63% with the proposed method. In addition, DC-link voltage fluctuations are minimized, and power oscillations are effectively suppressed. Stable operation is also maintained under sudden variations in solar irradiance and load conditions, confirming the effectiveness of the proposed approach for advanced grid-connected PV systems requiring high power quality.

The COVID-19 pandemic has highlighted the urgent need for fast, accurate, and trustworthy diagnostic tools to complement routine testing procedures. While Chest X-ray (CXR) imaging is a valuable alternative, it requires specialized interpretation that is prone to subjectivity, creating a bottleneck in clinical workflows. This study introduces COVIXNet, a robust and explainable deep transfer learning framework designed to automate the detection of COVID-19 from CXR images with high accuracy and provide clinically interpretable justifications for its predictions. COVIXNet is built upon a DenseNet-121 backbone, enhanced with a spatial attention mechanism to focus on diagnostically significant lung regions. Explainability is achieved using Gradient-weighted Class Activation Mapping (Grad-CAM). The model was trained and evaluated on COVIDXSet, a curated multi-source dataset of 8,591 CXR images. A two-phase transfer learning strategy was employed for effective feature adaptation. COVIXNet demonstrated state-of-the-art performance, achieving an accuracy of 96.8% (95% CI: 95.9% - 97.7%), a precision of 97.2%, and an AUC-ROC of 0.993 (95% CI: 0.989 - 0.997). The explainability module generated clinically meaningful heatmaps, with 92% rated as relevant by expert radiologists and achieving a high Pointing Game Accuracy of 81.5%. The model also showed consistent performance across demographic subgroups and varying image quality. COVIXNet offers a powerful combination of high diagnostic accuracy, robustness, and validated explainability. With a 48 ms inference time and 33.2 MB model size, COVIXNet is a promising and efficient tool for deployment in clinical settings to assist healthcare professionals in the rapid triage and diagnosis of COVID-19.

This work is based on classical and modern research on reinforcement learning, utilizing the CartPole-v0 environment for Q-learning, Hill Climbing, the REINFORCE algorithm, various versions of the Deep-Q-Network algorithm, and policy gradient approaches such as PPO and A2C. All the above models have been implemented and compared in terms of convergence speed, stability, efficiency, and robustness in noisy environments, as well as their generality in modified environments. In the experiments, the optimization of the hyperparameters and other algorithmic modifications, such as the Double DQN, Dueling DQN, and Reward Shaping, have also been attempted. When compared with Classical Control approaches like the Linear Quadratic Regulator, the flexibility, stability, and efficiency of Deep RL approaches prove to be far better. The use of reward curves, policy heatmaps, and trajectory plots has proved beneficial as a way of ensuring the algorithm\'s interpretability. The results demonstrated that the reinforcement learning algorithms PPO and Dueling DQN are the most efficient methods with good convergence rates and stability, while providing insights on computational efficiency. This research created a hybrid framework that combines the use of Hill Climbing and reinforcement learning as a way of improving stability while ensuring better efficiency through the introduction of stability reward shaping techniques, which proved to be better than the use of classical reinforcement learning and control approaches.

Prediction of clinical outcome in patients with COVID-19 is important for timely intervention, appropriate ICU triage, and best utilization of hospital resources. We propose a deep learning framework to predict the patient outcomes by utilizing longitudinal clinical data from a tertiary care hospital in India. After quality control (QC) preprocessing and filtering, 126,716 diagnostic records with 13 key laboratory and vital sign variables were retained for analysis. We propose and test several neural architectures for the classification of ICU vs ward patients, namely GRU, LSTM, BiLSTM, CNN-LSTM, and Transformer models. A two-stage experimental setup was employed. We began by doing an exploratory train-test analysis for hyperparameter tuning. Next, we performed a rigorous comparative analysis based on the five-fold cross-validation results. We evaluated the performance according to such criteria as accuracy, sensitivity, specificity, recall, precision, F1-score, ROC curve and AUC. Of the models we tested, the Transformer performs best overall, with 91.9% accuracy, 85.4% sensitivity, 75.6% specificity, and 83.1% recall. It clearly demonstrates a good specificity in the view of high-risk patients, but with a minimum false negative rate. The CNN-LSTM had a comparably good performance and the GRU was an attractive low-cost alternative with reasonable predictive power and computational burden. The generalizability of our models was validated externally, where the Transformer also yielded the best performance on unseen clinical data. These results highlight the promise of deep learning, in particular Transformer-based models, as robust instruments for COVID-19 prognosis and ICU admission management.

The growing adoption of electric vehicles (EVs) has led to a significant rise in electricity demand, creating challenges for the reliability of distribution networks. Strategic placement of EV charging stations (EVCSs) is crucial for ensuring grid stability and minimizing disruptions. To mitigate the impact of EVCSs on radial distribution networks (RDNs), this study explores the integration of distributed generators (DGs) within a reconfigured network, where optimal switch positions and power flow adjustments enhance overall system efficiency. However, poor placement and sizing of DGs and EVCSs can significantly affect system performance. This study analyses the impact of EVCSs on RDNs using Particle Swarm Optimization (PSO). The planning process consists of two stages: first, using PSO to determine the best location for EVCSs, and second, reconfiguring the network while simultaneously determining the optimal locations and sizes of DGs using Second-Order Cone Programming (SOCP) to ensure greater system stability. The methodology, validated on the IEEE 33-bus network, aims to reduce power losses and enhance voltage stability. The outcomes demonstrate how well the suggested strategy works to maximize DG and EVCS installation without sacrificing grid stability. This planning strategy not only optimizes network configuration by reducing power losses and voltage deviations but also contributes to a more resilient and efficient power distribution system in the era of widespread EV adoption.

Metal matrix composites (MMCs) have emerged as promising materials for aerospace and automotive applications due to their enhanced mechanical performance. Among various reinforcements, fly ash has gained increasing attention for aluminium-based MMCs because of its low cost, low density, and favourable mechanical characteristics. However, the influence of fly ash content on the microstructural evolution and mechanical behaviour of aluminium MMCs fabricated via cooling slope casting remains insufficiently explored. In this study, aluminium MMCs reinforced with 4, 8, and 12 wt.% fly ash were produced using cooling slope casting under different pouring temperatures and slope lengths, followed by T6 heat treatment. Mechanical properties were evaluated through tensile and hardness testing, while microstructural characterization was performed using optical microscopy, FESEM with EDX, and XRD analysis. Quantitative image analysis was employed to determine the globule size and shape factor of the primary α-Al phase. The results demonstrate that increasing fly ash content promotes grain refinement, reduces globule size, and enhances globularity, indicating improved microstructural stability after heat treatment. These microstructural improvements, together with precipitation hardening and more effective load transfer, lead to significant mechanical strengthening, with yield strength, ultimate tensile strength, and hardness increasing by 43%, 25%, and 29%, , while elongation is maintained with a modest improvement of approximately 4%. Furthermore, the lower density of fly ash contributes to overall weight reduction, making these composites attractive for lightweight structural applications. The findings highlight the potential of fly ash-reinforced aluminium MMCs as cost-effective, high-performance materials for engineering applications.

Optimizing network connectivity characterizes a major challenge in different applications, e.g. IoT, where the efficiency of connectivity under constraints like distance and adaptability is essential. The key objective of the Minimum Spanning Tree (MST) methods is to construct a subset of a graph in which every vertex is connected with the least amount of edge weight and without any cycles. As a robust framework for constructing distance-constrained network connectivity, this research proposes a novel algorithm named Recursive Node Connectivity Algorithm (RNCA). RNCA forms a tree-like networking structure motivated by minimum spanning tree (MST) principles, while explicitly considering distance constraints to ensure feasible communications. The RNCA utilizes repeated pruning and reweighting procedures to create cost-effective and distance-compliant network structures that can ensure flexible and scalable connectivity in dynamic contexts. The performance of RNCA is evaluated through conducting simulations with different network scales. The results show that RNCA achieves high adaptability, scalability, and reduced redundancy. RNCA outperforms Kruskal’s and Prim’s MST algorithms by reducing recomputed links during node updates by up to 80%. It also ensures minimum change in network connections when nodes are added or removed from the original network topology. Therefore, it can be considered as a transformative solution for many modern network applications, such as transportation systems, WSNs, and IoT.

The rapid growth of user density and heterogeneous service requirements in 5G and emerging 6G networks has intensified multi-user interference, particularly in power-domain non-orthogonal multiple access (PD-NOMA) systems. Conventional successive interference cancellation (SIC) techniques suffer from error propagation and performance degradation under imperfect channel conditions, limiting their reliability in practical deployments. This paper proposes a deep learning–based interference cancellation framework that combines Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks for end-to-end signal detection in downlink NOMA systems. The proposed model exploits CNN layers to extract spatial interference features from complex-valued received signals, while BiLSTM layers capture bidirectional temporal dependencies caused by channel fading and symbol correlation. Unlike traditional SIC-based receivers, DL-NOMA-CancelNet directly reconstructs user signals without explicit channel estimation or manual interference subtraction. Simulation results obtained using MATLAB under Rayleigh fading and AWGN conditions demonstrate that our framework significantly outperforms conventional Least Squares (LS), Minimum Mean Square Error (MMSE), and single-network deep learning detectors. Specifically, this method achieves up to an 8–10× reduction in bit error rate (BER) across a wide signal-to-noise ratio (SNR) range, while maintaining low inference latency and moderate model complexity. These results confirm that our work provides a robust and scalable solution for interference mitigation in future 5G/6G downlink NOMA systems, particularly for ultra-reliable low-latency communications (URLLC) and massive machine-type communications (mMTC).

Cloud storage is widely used for store a personalized and Sensitive information. Cloud usually stores the files in encrypted mode. To search a user specific file among the encrypted files is difficult. Traditional keyword search may be fails when users phrase differently. Typos and vocabulary differences further reduce accuracy. Existing encrypted search solutions typically trade security for search quality. To address this gap, this work introduces a dual-index encrypted search system. This system combines exact keyword matching with semantic understanding. A keyword index, protected with LFEHI model. It supports fuzzy matching for small typing error. A semantic index built with BERT captures meaning and organizes embeddings in a kd-tree for fast retrieval. During a search, both indexes run in parallel. The encrypted keywords match through the hash index, and a semantic embedding check for related concepts. A 70/30 keyword-to-semantic weight is used to integrate their results. To ensuring that exact matches continue to be given priority. Tests on 1,000 documents and 300 queries show clear improvements. The system reaches 90.2% recall—22% higher than keyword-only search and it handles single-character typos with 94.6% accuracy. Average query time stays below 64 ms, supporting real-time use, and the storage cost is only 18.5%. There is no evidence s of content leakage or query leakage following a security evaluation. Due to the dual-index design, encrypted search can deliver meaningful understanding without compromising speed and security.