Transactions on Machine Intelligence2821-16937120240101An Electromyography Recording and On-Line Driving System for a Robotic Wrist11019160310.47176/TMI.2024.1ENM.R.Sayyed NooraniDepartment of Mechatronics Engineering, University of Tabriz, Tabriz, Iran0000-0002-7487-8233F.MortezazadehGraduated in Medical Engineering-Biomechanics, Department of Mechatronics Eng., University of Tabriz, Tabriz, IranA.SabbaghiGraduated in Medical Engineering-Biomechanics, Department of Mechatronics Eng., University of Tabriz, Tabriz, IranJournal Article20231106Rehabilitation for individuals suffering from motor disabilities caused by conditions such as spinal cord injuries, neuromuscular disorders, or stroke-related complications remains a significant challenge in both clinical practice and the daily lives of patients. As the global incidence of such conditions continues to rise, the need for effective and accessible rehabilitation solutions has become more urgent. In particular, the development of home-based assistive technologies, which enable patients to undergo therapy without frequent visits to medical centers, has become a key area of interest in biomedical engineering. Among these technologies, exoskeleton robots have emerged as promising tools for restoring lost motor functions. Recent advancements have shifted from predefined motion execution toward intelligent systems capable of recognizing the user's movement intentions. This study presents the design and implementation of a wrist exoskeleton prototype controlled by electromyographic (EMG) signals. The system uses an Arduino microcontroller integrated with EMG modules to detect muscle activity in the forearm and drive a servo motor, enabling wrist movements such as flexion-extension and abduction-adduction. EMG signals were recorded in a controlled laboratory environment following standard motor task protocols. Signal preprocessing and movement classification were carried out using MATLAB, utilizing its serial communication toolbox to interface with the Arduino board. The developed algorithm generates three-state control commands to drive the motor, allowing smooth, real-time imitation of voluntary wrist movements. The results demonstrate the feasibility of this approach for future application in wearable, intelligent rehabilitation systems tailored to individual users.https://www.tmachineintelligence.ir/article_191603_a78f1fe35f716113c592e691f8038328.pdfTransactions on Machine Intelligence2821-16937120240101Thermal Management of Insulated-Gate Bipolar Transistor Modules by Air and Liquid Cooling: A Numerical Study112219151010.47176/TMI.2024.11ENA.SoleimaniSchool of Mechanical Engineering, College of Engineering, University of Tehran, IranA.MoghoufeSchool of Mechanical Engineering, College of Engineering, University of Tehran, IranM.SaffaripourSchool of Mechanical Engineering, College of Engineering, University of Tehran, IranJournal Article20231116Thermal management of three insulated-gate bipolar transistor (IGBT) modules with application in power inverters is studied numerically with two different cooling approaches, an air-cooled heat sink and a liquid-cooled cold plate. Four air-cooled heat sink configurations have been studied and compared: a natural-convection configuration, a forced parallel-flow configuration, and two forced impingement-flow configurations with different fans. For the cold plate, a suitable coolant is chosen based on the operating conditions of the current study. The maximum total heat dissipation rate of the three IGBT modules, which includes switching and conductive power dissipation rates, is calculated to be 1200 W at peak load. The air-cooled and liquid-cooled cases are studied under a variety of operating conditions, including different ambient temperatures and heat dissipation rates. The results show that natural convection can only be used for total heat dissipation rates below 300 W. Whereas, the forced impingement-flow air cooling configuration with large fans and the liquid cooling configuration can keep the junction temperature of the IGBTs below the maximum permissible value under all the operating conditions used in this study. The liquid-cooled cold plate has the lowest thermal resistance and because of retaining the largest safety margin for junction temperature, this method is suitable for power dissipation rates higher than 1200 W. The results indicate that the performance of the air-cooled heat sinks is not a strong function of air flow direction and mainly depends on air flow rate.https://www.tmachineintelligence.ir/article_191510_68b5d1dc71960af14b3159a151a7f761.pdfTransactions on Machine Intelligence2821-16937120240101Dynamic Modeling of Trains for Application in Railway Transportation Automation Systems233719151410.47176/TMI.2024.23ENF. Z.AmirhosseiniMaster's Graduate, Control Group, Faculty of Electrical Engineering, Iran University of Science and Technology, Tehran, IranB.Ghorbani VagheiAssistant Professor, Control and Railway Signal Group, Faculty of Railway Engineering, Iran University of Science and Technology, Tehran, Iran0000-0002-1698-7773H.BolandiProfessor, Control Group, Faculty of Electrical Engineering, Iran University of Science and Technology, Tehran, Iran0000-0002-3501-6120Journal Article20231211The current railway transportation industry utilizes automation across various domains such as safety, stability, motion control, and train scheduling to optimize the use of railway resources and facilities. A prerequisite for these automation systems is the ability to predict and simulate train motion, which requires accurate train motion modeling. Factors such as air friction, the complexity of rail routes, the interaction forces between wagons, dynamics of force generation in actuators, and mechanical part friction result in nonlinear equations for train motion dynamics. Experimental values for some of these factors are uncertain due to wear and structural changes in components, while others are unmeasurable, complicating control conditions. This paper describes multi-particle and single-particle dynamic models of train motion. For practical operation, time-varying or unknown parameters in these equations are identified using a recursive least squares algorithm, and the estimated values are applied in a sliding mode control signal to compensate for baseline resistance and route disturbances. The designed sliding mode controller at the core of this system mitigates the effects of uncertainties and accurately tracks the desired speed-location profile. Simulation results presented in this paper demonstrate precise parameter estimation along with favorable tracking outcomes for the speed-location characteristic.https://www.tmachineintelligence.ir/article_191514_c44154946efbeaf8cbf063050c158667.pdfTransactions on Machine Intelligence2821-16937120240101Detection of Phishing Website Attacks in Electronic Banking Using a Principal Component Analysis Algorithm and Multi-Layer Perceptron Neural Network Algorithm385018439610.47176/TMI.2024.38ENM.GhasemiDepartment of Computer Science, Islamic Azad University, Yasuj Branch, Yasuj, Iran0009-0000-8241-4933Journal Article20231120Phishing, commonly known as the unauthorized acquisition of personal information from users of online platforms and clients of digital stores and financial institutions, has witnessed a notable surge in recent years. This surge has fueled the growth of a thriving criminal enterprise, particularly targeting financial service providers. Given the magnitude of this threat, we adopted a dual approach involving the application of a Principal Component Analysis (PCA) algorithm and a multi-layer perceptron neural network algorithm to identify and combat phishing attacks within the realm of electronic banking. Initially, we employed the PCA algorithm to streamline the identification process, reducing the number of features from an initial 30 to a more manageable 14. Following this feature reduction step, we fine-tuned the accuracy of detecting phishing website attacks using the multi-layer perceptron neural network algorithm. This algorithm, functioning as a binary classification technique, adeptly determines whether an input vector belongs to a specific class. Acting as a linear classifier, it relies on the weighted linear combination of input factors to make predictions. To further fortify our defenses, we implemented the Waka tool, an online algorithm capable of meticulously examining individual inputs. Through the strategic integration of the PCA and multi-layer perceptron neural network algorithms, we achieved a substantial enhancement in the accuracy of detecting phishing website attacks in the electronic banking domain, reaching an impressive 91.64%.https://www.tmachineintelligence.ir/article_184396_0f5370c0bcdc56110da549a6c76dd751.pdfTransactions on Machine Intelligence2821-16937120240101Performance Improvement for Multi-Criteria Decision Making Using Collaborative Filtering-Based Recommender Systems516019151310.47176/TMI.2024.51ENM.FazlAazadMaster student, Computer Department, Faculty of Engineering, Yasouj Branch, Islamic Azad University, Yasouj, Iran0009-0001-5034-5563Journal Article20230102In the past, making decisions or recommendations, as well as processing data, did not pose significant challenges due to the limited data related to a small number of users. However, with the continuous growth of the population and the exponential increase in data and user profiles across global databases, the task of generating improved decisions or recommendations in terms of time, location, cost, and other characteristics has become more complex. Recommender systems, data mining techniques, and algorithms play a crucial role in addressing these challenges. The escalating attention of both researchers and practitioners towards recommender and data mining systems reflects the increasing difficulty in handling vast datasets efficiently. This article aims to analyze a relatively extensive dataset with diverse characteristics, seeking to achieve optimal clustering or categorization and regression in the shortest possible time, considering economic efficiency and other key features inherent to the dataset. The dataset under consideration exhibits data oscillation in four features. Initially, clustering and regression are performed using the implicit method. Additionally, employing the collaborative filtering approach based on recommender systems, specifically the collaborative filtering method using the ranking matrix (user-item collaboration), is employed. This method yields highly effective recommendations for new users based on a variety of essential criteria.https://www.tmachineintelligence.ir/article_191513_2ceba3a64849a6fb5af624d1753f701e.pdfTransactions on Machine Intelligence2821-16937120240101Remaining Useful Life Estimation of Bearings Using Vibration Signal Processing Based on Continuous Wavelet Transform and LSTM Deep Learning Network616919151110.47176/TMI.2024.61ENP.AmjadianAssistant Professor, Department of Mechanical Engineering, Sahneh Branch, Islamic Azad University, Sahneh, Iran0000-0001-6225-8443Journal Article20230107This study introduces a predictive model for estimating the remaining useful life (RUL) of bearings, leveraging a deep learning approach based on Long Short-Term Memory (LSTM) networks and Continuous Wavelet Transform (CWT). The vibration signal data, which is essential for condition monitoring, were sourced from a well-established dataset. To enhance the model’s ability to capture time-frequency features, each vibration signal was processed using CWT, resulting in scalogram representations. These scalograms were then fed into the LSTM network to create an accurate RUL prediction model. The performance of the proposed LSTM-based deep learning model was thoroughly assessed by comparing it with three conventional artificial neural network (ANN) models, each trained using a different algorithm: Trainbr, Trainlm, and Trainscg. The results demonstrated that the LSTM model significantly outperformed the traditional ANN models, yielding a Root Mean Square Error (RMSE) of 0.18 and a Mean Absolute Percentage Error (MAPE) of 0.0103. In contrast, the three ANN models resulted in much higher average RMSE and MAPE values of 12.4377 and 1.5557, respectively. These findings confirm the superiority of the LSTM-based model for RUL estimation in bearing health monitoring and its potential for real-world industrial applications.https://www.tmachineintelligence.ir/article_191511_e45a4529c8557124d9a8bf4674f75811.pdfTransactions on Machine Intelligence2821-16937120240101Demographic and Efficiency Analysis of Street Transportation Network in Districts of Tehran Metropolitan708122401610.47176/TMI.2024.70ENA.GhahremanlouPhD Candidate, Department of Civil Engineering, Science and Research Branch, Islamic Azad University,
Tehran, Iran0000-0002-7543-4781M.SaffarzadehProfessor, Department of Civil and Environmental Engineering, Tarbiat Modares University, Tehran, Iran0000-0002-9713-2776A.NaderanAssistant Professor, Department of Civil Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran0000-0001-5842-0188H.JavanshirPhD, Assistant Professor, Department of Industrial Engineering, South Tehran Branch, Islamic Azad University, Tehran, Iran0000-0001-6884-3352Journal Article20231205The primary goal of transportation accessibility is to improve travel efficiency by minimizing the distance between trip origins and destinations, thereby enhancing user experience. While conventional performance assessments focus on speed and distance, this study integrates demographic and spatial analytics to evaluate Tehran’s mobility infrastructure. Using district-level population and census data from Tehran Municipality and the Statistics Center, four analytical steps were undertaken: mapping population density via ArcGIS 10.8, measuring real and straight-line distances with Google Earth Pro, and calculating accessibility using the route factor. The 2021 forecast reveals that approximately 67% of Tehran’s population resides in suburban districts, with District 4 alone accounting for 10.57%. Despite this concentration, peripheral districts remain sparsely populated compared to central ones. Accessibility assessment classified internal district connections into five categories: excellent (38%), moderate (32%), and poor (30%). Tehran’s average route factor of 1.52 indicates a moderate level of accessibility. These findings underscore a significant imbalance in population distribution and transport network efficiency. To address this, targeted investments in underperforming areas should be prioritized. This study demonstrates the value of combining demographic insights with spatial network analysis, delivering actionable intelligence for urban planners to optimize resource allocation and strengthen connectivity in Tehran’s evolving metropolitan landscape.https://www.tmachineintelligence.ir/article_224016_e59634a000234e0a9d640b08417cf8a8.pdf