IRJEAS

Volume 13 Issue 4                        October-December 2025

Research Article

LICENCED UNDER

FIND ON

CROSSREF

SCILIT

 

Preliminary CFD Evaluation of Selected Hypotheses for a Wind Energy Concentrator

Country- INDIA

Tarun Kumar Ahirwar, Dr. Paulami Sahu, Dr. Prashant V. Baredar

PAPER ID: IRJEAS04V13I4001

Published: Oct 2025

Journal: IRJEAS, Volume 13, Issue 4

Pages: 01-11

Cite this Article

Tarun Kumar Ahirwar, Paulami Sahu, Prashant V. Baredar. Preliminary CFD Anomaly Identification in Cyber Logs Through Machine Learning Techniques. International Research Journal of Engineering & Applied Sciences (IRJEAS). 12(4), pp. 01-11, 2025. https://doi.org/10.55083/irjeas.2025.v13i04001   

Abstract:

This paper presents a preliminary computational fluid dynamics (CFD) assessment of four conical frustum windconcentrator configurations (Models A–D) intended for augmentation of small wind turbines in low-wind-speed regions. The designs share a fixed 3:1 inlet-to-outlet area ratio and differ only in frustum length and wall angle, providing a controlled comparison of geometric effects. Steady Reynolds-averaged Navier–Stokes (RANS) simulations with the Realizable k-ε model in ANSYS Fluent quantify outlet velocity amplification (V_Amp = V_out / V_in), radial uniformity and turbulence kinetic energy (TKE). Model B (short and steep taper) achieves the highest V_Amp, whereas longer designs (Models C and D) yield more uniform outflow at the expense of reduced gain. Model A strikes an intermediate balance. The results establish clear geometry–performance trade-offs that are practically relevant: overly high tapers risk non-uniform, turbulent outflow, while long ducts add material and drag but offer diminishing returns. This paper includes only the preliminary interpretation derived from test-run CFD observations. These CFD trends form the basis for turbine integrated modeling and controlled experimentation in subsequent work.

Research Article

LICENCED UNDER

FIND ON

CROSSREF

SCILIT

 

Anomaly Identification in Cyber Logs Through Machine Learning Techniques

Country- INDIA

Ram Dwivedi, Nitesh Gupta, Anurag Shrivastava

PAPER ID: IRJEAS04V13I4002

Published: Oct 2025

Journal: IRJEAS, Volume 13, Issue 4

Pages: 12-20

Cite this Article

Ram Dwivedi, Nitesh Gupta, Anurag Shrivastava. Anomaly Identification in Cyber Logs Through Machine Learning Techniques. International Research Journal of Engineering & Applied Sciences (IRJEAS). 12(4), pp. 12-20, 2025. https://doi.org/10.55083/irjeas.2025.v13i04002   

Abstract:

With the rising volume and sophistication of cyber attacks, traditional rule-based security systems have become insufficient to detect emerging and unknown threats. Machine Learning (ML) provides a dynamic and scalable solution for anomaly detection in cyber security logs, enabling systems to identify unusual patterns with minimal human intervention. This study offers a thorough framework that combines supervised, unsupervised, and semi-supervised models for anomaly identification using machine learning.

Using the HDFS log dataset, we implement and evaluate Isolation Forest, Auto encoder, and Random Forest algorithms. The results demonstrate that ML techniques significantly enhance detection accuracy and reduce false positives, with Random Forest achieving the highest performance across all evaluation metrics.

Research Article

LICENCED UNDER

FIND ON

CROSSREF

SCILIT

 

Overcoming Operational Bottlenecks: A Quantitative Analysis of the Benefits of Digital Weighbridge Integration with PCS 1X and Sagar Setu

Country- INDIA

Venkata Ramana Akkaraju

PAPER ID: IRJEAS04V13I4003

Published: Oct 2025

Journal: IRJEAS, Volume 13, Issue 4

Pages: 21-37

Cite this Article

Venkata Ramana Akkaraju. Overcoming Operational Bottlenecks: A Quantitative Analysis of the Benefits of Digital Weighbridge Integration with PCS 1X and Sagar Setu. International Research Journal of Engineering & Applied Sciences (IRJEAS). 12(4), pp. 21-37, 2025. https://doi.org/10.55083/irjeas.2025.v13i04003     

Abstract:

Indian ports process more than 90% of the country’s trade volume, which makes efficient operations a key to economic competitiveness. Conventional manual weighbridge procedures have been recognized as a cause of bottlenecks for years, leading to long truck turnaround times, inaccurate data, and decreased throughput. In an effort to solve these issues, this research assesses the effect of implementing digital weighbridge systems in combination with the Port Community System (PCS 1x) and the SAGAR SETU mobile application on total operating performance. A six-month observation study was carried out in two of India’s largest seaports, comparing pre- and post-integration performance measures. The most important indicators were average truck turnaround time, weighment error rates, and levels of throughput per day. Statistical methods like paired t-tests were used to compare differences in performance, whereas regression analysis established integration-throughput improvement relationship. The findings show a 28% decrease in mean truck turnaround time, a 62% decrease in weight-related mistakes, and a 15% improvement in daily truck processing rate, even during peak cargo load periods. The results emphasize the capability of digital synchronization and real-time data sharing in making logistics processes more efficient, minimizing manual interventions, and enhancing decision-making effectiveness. This study offers policy formulation insights under the Sagarmala program with a focus on the urgency for accelerated digital integration in Indian ports. This research not only establishes the operational advantages of automation but also provides a model for scalability, furthering India’s vision of becoming a world-class port infrastructure and globally competitive supply chain nation. These results—showing a 28% reduction in truck turnaround time, a 62% decline in weighbridge errors, and a 15% rise in daily throughput—demonstrate that digital weighbridge integration not only enhances operational efficiency but also yields measurable financial returns, making it a scalable solution aligned with the Sagarmala program and India’s Maritime Vision 2030.

Review Article

LICENCED UNDER

FIND ON

CROSSREF

SCILIT

 

The Evolution of Phishing Detection: From Static Signatures to Autonomous Generative Agents

Country- INDIA

Ashwarya Singh, Kalpana Mishra

PAPER ID: IRJEAS04V13I4004

Published: Oct 2025

Journal: IRJEAS, Volume 13, Issue 4

Pages: 38-47

Cite this Article

Ashwarya Singh, Kalpana Mishra. The Evolution of Phishing Detection: From Static Signatures to Autonomous Generative Agents. International Research Journal of Engineering & Applied Sciences (IRJEAS). 12(4), pp. 38-47, 2025. https://doi.org/10.55083/irjeas.2025.v13i04004.

Abstract:

Phishing is still one of the most common and harmful threats in cybersecurity. It is the main way that most data breaches happen. As a result, it is very important to create strong and quick detection methods. This paper offers an extensive literature analysis on the progression of phishing detection systems, outlining their development from initial static methods to the contemporary state-of-the-art. We start by looking at classic solutions, such blacklist-based and heuristic-based systems, and pointing out how they don’t work well against new, zero-day threats. Next, we look at the big change that machine learning brought about, which made it possible to create more flexible solutions by using feature engineering from URLs and visual similarity analysis of webpages. A lot of attention is being paid to the rise of reference-based detection systems, which check the validity of web pages by comparing them to a database of real brands. We critically examine advanced dynamic systems such as DynaPhish, which try to automate knowledge base expansion, revealing their intrinsic fragility and reliance on inflexible logic. Lastly, we look at the cutting edge of phishing detection, which is defined by the use of generative AI and autonomous agents. We contend that Large Language Model (LLM)-powered agents, endowed with human-like reasoning, multi-modal analysis, and dynamic tool utilisation, constitute a possible remedy to the shortcomings of previous approaches. This study brings together the most important progress, points out ongoing problems, and suggests that the future of phishing defence rests in making smart, self-driving systems that can think and change in real time to deal with the changing nature of modern phishing threats.

Research Article

LICENCED UNDER

FIND ON

CROSSREF

SCILIT

 

ADEPT: An Autonomous, LLM-Powered Agent for Dynamic, Reference-Based Phishing Detection

Country- INDIA

Ashwarya Singh, Kalpana Mishra

PAPER ID: IRJEAS04V13I4009

Published:

Journal: IRJEAS, Volume 13, Issue 4

Pages: 81-96

Cite this Article

Ashwarya Singh, Kalpana Mishra. ADEPT: An Autonomous, LLM-Powered Agent for Dynamic, Reference-Based Phishing Detection. International Research Journal of Engineering & Applied Sciences (IRJEAS). 12(4), pp. 81-96,
2025. https://doi.org/10.55083/irjeas.2025.v13i04009.

Abstract:

Conventional phishing detection systems, which predominantly rely on static blacklists and rigid, signature-based heuristics, are increasingly ineffective against the dynamic and polymorphic nature of modern phishing attacks. Reference-based detection, which verifies a webpage’s authenticity against a known brand’s identity, offers a more robust approach but is critically hampered by the need to maintain a comprehensive and constantly updated knowledge base. This paper addresses this fundamental challenge by introducing ADEPT (Autonomous Dynamic Agent for Phishing Threat), a novel framework architected around a sophisticated Large Language Model (LLM) acting as an autonomous agent. We first present a granular error analysis of a state-of-the-art dynamic reference-based system, DynaPhish, revealing its brittleness and critical failures stemming from rigid logic and dependency on external APIs. To overcome these deficiencies, we designed and implemented the ADEPT framework, which equips an LLM agent with a multi-modal perception pipeline and a toolkit for real-time information retrieval from the web. The agent mimics human cognitive processes to dynamically investigate suspicious webpages, analyze visual and textual content, and reason about a brand’s identity without relying on a pre-existing static knowledge base. Through a series of rigorous experiments on a balanced dataset of 400 phishing and benign samples, the ADEPT framework, utilizing the GPT-4 model, achieved a phishing detection accuracy of 0.945. This represents a significant improvement over both the DynaPhish baseline (0.499 accuracy) and simpler LLM-based methods, empirically validating that an autonomous, agent-based approach provides a more resilient and effective solution to the pervasive threat of phishing.