AUTHOR(S)
Georgi Panchev

ABSTRACT
This study presents a holistic analysis of radial shaft seals by means of 3D finite element analysis (FEA). Special emphasis is given to the sealing edge, examining its complex behavior under operating conditions, and including the surface roughness profile on the microscale for detailed 3D model simulation. In addition, the underlying equations of hyperelastic material models, which are important for appropriately simulating the elastomeric parts of the seal, are discussed to give a sound mathematical foundation to the simulations. This encompasses thorough discussions on boundary conditions and mathematical models necessary for replicating the complicated defor- mation and contact mechanics of the sealing interface. The tangential distortion of the sealing edge, an important factor determining sealing performance, is examined using multiscale models linking macroscopic seal geometry to microscopic surface roughness.

DOI

DOWNLOAD
https://unitech-selectedpapers.tugab.bg/images/2025/8-Advanced%20Fluid%20and%20Heat%20Engineering/p86_s8_u124_id100-SP.pdf

How to cite this article:
Georgi Panchev, FINITE ELEMENT ANALYSIS OF A RADIAL SHAFT SEAL WITH A NOVEL MESH REFINEMENT APROACH, UNITECH – SELECTED PAPERS - 2025

AUTHOR(S)
R. La Regina¹, A.Porcaro¹, D. Gentiluomo², M. C. De Simone², C. M. Pappalardo², and D. Guida^*2

ABSTRACT
Accurate segmentation of medical images is essential for diagnosis and treatment planning. Traditional manual methods in platforms such as 3D Slicer are precise but time-consuming and dependent on the operator’s expertise. With the development of deep learning, neural networks (particularly U-Net) have enabled automatic segmentation with improved efficiency and precision. This paper compares manual and automatic segmentation on MRI spleen scans. Manual methods included thresholding, painting, and tracing techniques, while automatic segmentation was performed using the MONAI Label framework integrated with U-Net. Evaluation with the Dice coefficient showed high overlap between methods, with values above 0.9 in most cases. Results confirm that deep learning - based segmentation provides faster and reliable outcomes, supporting its application in clinical practice.

DOI

DOWNLOAD
https://unitech-selectedpapers.tugab.bg/images/2025/6-Mechanical%20engineering%20Machines%20and%20Instruments%20design/p282_s13_u4_id456-SP.pdf

How to cite this article:
R. La Regina¹, A.Porcaro¹, D. Gentiluomo², M. C. De Simone², C. M. Pappalardo², and D. Guida^*2, CERERE PROJECT: A PRACTICAL AND METHODOLOGICAL FRAMEWORK FOR DESIGNING AN OPTIMAL UGV FOR PRECISION AGRICULTURE IN WATER-LIMITED ENVIRONMEN UNITECH – SELECTED PAPERS - 2025

AUTHOR(S)
Georgi Panchev

ABSTRACT
Accurate segmentation of medical images is essential for diagnosis and treatment planning. Traditional manual methods in platforms such as 3D Slicer are precise but time-consuming and dependent on the operator’s expertise. With the development of deep learning, neural networks (particularly U-Net) have enabled automatic segmentation with improved efficiency and precision. This paper compares manual and automatic segmentation on MRI spleen scans. Manual methods included thresholding, painting, and tracing techniques, while automatic segmentation was performed using the MONAI Label framework integrated with U-Net. Evaluation with the Dice coefficient showed high overlap between methods, with values above 0.9 in most cases. Results confirm that deep learning - based segmentation provides faster and reliable outcomes, supporting its application in clinical practice.

DOI

DOWNLOAD

https://unitech-selectedpapers.tugab.bg/images/2025/8-Advanced%20Fluid%20and%20Heat%20Engineering/p85_s8_u124_id99-SP.pdf

How to cite this article:
Georgi Panchev, REVIEW OF FUNCTIONAL THEORIES AND NUMERICAL APPROACHES IN ELASTOHYDRODYNAMIC LUBRICATION OF RADIAL SHAFT SEALS, UNITECH – SELECTED PAPERS - 2025

AUTHOR(S)
Mustafa Kemal TEZCAN*, Umit KURSUN

ABSTRACT
This study proposes a new enhanced hybrid deep learning model, combining LSTM and Autoencoder architectures with a non-parametric Isolation Forest (IF) filter, to address the high false alarm issues of traditional systems. The model's anomaly threshold was determined using the statistically 99th percentile (0.0649). The system's robustness was validated using a noisy synthetic time-series dataset. Experimental results establish the foundational superiority of the Hybrid Model: while the single Autoencoder and LSTM benchmarks detected 187 and 189 anomalies, respectively, the standalone Hybrid Model successfully filtered over 280% of that noise, detecting only 49 anomalies. The subsequent integration of the Isolation Forest filter, trained on a multi- dimensional error vector, provided a final, critical safety layer, further reducing the count to 47. This confirms the reliability of the hybrid approach and introduces a robust, multi-layered validation system, paving the way for future implementation on local IoT hardware for real-world validation.

DOI

DOWNLOAD
https://unitech-selectedpapers.tugab.bg/images/2025/5-Automation%20and%20Robotics/p182_s5_u189_id454-SP.pdf

How to cite this article:
Mustafa Kemal TEZCAN*, Umit KURSUN, A HYBRID LSTM-AUTOENCODER-ISOLATION FOREST DEEP LEARNING MODEL FOR SMART GAS LEAKAGE ALERT SYSTEM, UNITECH – SELECTED PAPERS - 2025