Keywords: Industrial Applications, Control Systems Design, Control Applications
Abstract: Actual industrial processes and production lines contain two or more control loops. One of the main challenges consists in ratio control in case one or more included parameters / elements fails or in disturbances presence. The paper presents solutions of real time balance for two or more parallel control structures and maintaining the ratio between controlled parameters. Applicability is proved on a real time hardware and software control structure of three ratios, a situation encountered to the modern transportation, chemical and other real processes.

Keywords: Industrial Applications, Control Systems Design, Adaptive and Robust Control
Abstract: The paper aims at bringing enhancements for the control of thermo-energetic processes by proposing anticipative action in order to reduce the perturbation effects in the process. Our contribution is organized in three connected sections. After a short introduction, in the first section, the dynamic model of the transfer energy between agent and product is estimated based on thermal balance equations, to explore the static and dynamic evolution of the process and to design the control system algorithms. In the second section, the nominal digital system is designed based on dynamic model and the imposed performances for the close loop system are validated through simulation. In the final section, the cascade and feedforward structures are designed respectively. The performances based on anticipative action to guarantee the invariance of the systems are shown through in simulation environment. The simulations results confirm the effectiveness of this research and the possibility of transferring these results towards the industrial heat processes. Finally, the conclusions and perspectives are given.

Keywords: Industrial Applications, Machine/Reinforcement Learning, Optimization and Optimal Control
Abstract: In large scale global supply chains, the inventory cost sensitivity due to supplier disruption can be high. Dual- sourcing, a inventory policy which leverages two suppliers to minimize the cost of supplier disruption, is often applied to minimize the inventory cost in the face of uncertain lead times and consumer demand. However, computing optimal policies for dual-sourcing faces challenges in modern supply chains due to the large scale of the system. We introduce Dyanmic Programming with Monte Carlo Value Approximation (DPMC), an approximate dynamic programming algorithm with polynomial time complex- ity which applies Monte Carlo simulation to estimate the optimal value function to address the large scale dual-sourcing problem. We show that DPMC is theoretically guaranteed to converge to the optimal policy by improving the value function approximator and/or increasing the number of Monte Carlo iterations. Via empirical simulation, we demonstrate that DPMC is competitive and often exceeds the cost-minimizing performance of other state- of-the-art dual-sourcing policies, specifically in scenarios where suppliers subject to disruption and/or fixed ordering costs.

Keywords: Machine/Reinforcement Learning, Distributed Parameter Systems, Adaptive and Robust Control
Abstract: This paper delves into optimal adaptive control for a plug flow reactor (PFR) partial differential equations (PDEs) model using the integral reinforcement learning (IRL) technique. Initially, it introduces a policy iterative algorithm designed to learn the solution of the corresponding matrix Riccati differential equation in real time. Notably, this method operates independently of explicit insight into the internal dynamics of the PFR process. Moreover, this paper establishes the convergence of the algorithm, contingent upon the initial action being stabilizing. Furthermore, an alternative algorithm is introduced to enhance the practical implementation of the IRL approach. The paper substantiates its findings through numerical simulations, demonstrating the efficacy of the developed algorithm.

Keywords: Machine/Reinforcement Learning, Control Systems Design, Nonlinear Systems
Abstract: This paper proposes a reinforcement learning (RL) framework for controlling and stabilizing the Twin Rotor Aerodynamic System (TRAS) at specific pitch and azimuth angles and tracking a given trajectory. The complex dynamics and non-linear characteristics of the TRAS make it challenging to control using traditional control algorithms. However, recent developments in RL have attracted interest due to their potential applications in the control of multirotors. The Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm was used in this paper to train the RL agent. This algorithm is used for environments with continuous state and action spaces, similar to the TRAS, as it does not require a model of the system. The simulation results illustrated the effectiveness of the RL control method. Next, external disturbances in the form of wind disturbances were used to test the controller’s effectiveness compared to conventional PID controllers. Lastly, experiments on a laboratory setup were carried out to confirm the controller’s effectiveness in real-world applications.

Keywords: Nonlinear Systems
Abstract: Prior to the current work, the various composition products of non--commutative formal power series corresponding to the cascading of Chen--Fliess series were defined using the shuffle product of words which is the symmetrization of the non--commutative and non--associative half--shuffle product (Zinbiel algebra). The composition products of generating series were interpreted as a homomorphism of shuffle algebra of non--commutative formal power series in its left argument. The goal of this paper is to explain that composition product is a homomorphism of right--associative bracketing Zinbiel algebras in its left argument fundamentally.

Keywords: Nonlinear Systems
Abstract: The decision of the controllability property of a nonlinear system is still not settled completely. For input-linear systems the well-known Lie algebra rank condition can be used. In this contribution we make use of this rank condition to compute the set of not locally controllable points of polynomial dynamical systems by algebraic methods.

Keywords: Nonlinear Systems, Linear Systems
Abstract: The paper develops a comprehensive framework for analyzing the invariance of polyhedral sets with respect to the trajectories of nonlinear continuous-time dynamic systems. The fundamental contribution consists in stating and proving a general theorem, which takes into account nonlinear time-varying systems and, respectively, polyhedral sets with arbitrary time-dependence. To the intrinsic mathematical value of this result, the work adds the significance of some problems with specific contexts, which can be derived as particular cases of our general setting. At the same time, it is shown that some of these particular cases are correlated with specific invariance properties, which were studied individually, through separate researches. The specificity of the properties refers both to characteristics of dynamics and to subclasses of polyhedral sets.

Keywords: Nonlinear Systems, System Identification and Modeling, Control Applications
Abstract: This paper presents a model-based control strategy for usage in multilevel pressure boosting systems with centrifugal pumps operating in parallel. The control is based on a nonlinear system model using exact input-output linearization. With the help of an optimization algorithm, the total consumption flow rate is distributed optimally in such a way that the total hydraulic efficiency of all running pumps is maximized in steady state while complying with all constraints. Trajectories for every single pump are planned on model-based relations in order to reduce overshoots in the pipeline pressure, in particular during start-up processes. The whole control strategy is validated on a test bench.

Keywords: System Identification and Modeling, Nonlinear Systems, Data - Driven Control
Abstract: This paper investigates the data-driven discovery of equations of motion for twin-tailed fighter aircraft, with a specific focus on addressing challenges posed by buffeting-induced vibrations in the F-15 model. Employing the Sparse Identification of Nonlinear Dynamics (SINDy) algorithm, our research delves into the complexities of F-15 twin-tail dynamics. The utilized nonlinear model takes into account critical factors such as aerodynamic damping, cubic nonlinearities and inter-tail coupling. While the SINDy algorithm demonstrates promise in capturing the original dynamics, its sensitivity to wind disturbances prompts the application of the SINDYc algorithm. This enhanced approach accurately predicts system dynamics and provides valuable insights into effective control inputs. In this study we follow a step-by-step process, initially excluding wind disturbances to focus on intrinsic dynamics, followed by their introduction to assess adaptability. Coefficients derived from the system identification algorithm closely align with the original synthetic data we generated using an existing model and the observed negligible error underscores the success of our data-driven methodology. This research contributes to advancing data-driven dynamics in aerospace engineering, providing an insight of how data-driven approaches could be utilized in the field.

Keywords: System Identification and Modeling, Optimization and Optimal Control, Uncertain Systems
Abstract: The magnitude frequency-based response data fitting mechanism is available for the case of proper or strictly proper linear and time-invariant (LTI) systems. This paper presents an extension to this mechanism for improper single-input and single-output (SISO) systems described by linear differential-algebraic equations (DAEs). Such descriptor models are computed as the solution to an optimization problem formulated as a Log-Chebyshev approximation, with additionally-imposed upper boundness, stability, and minimum phase constraints, useful in the context of robust synthesis. Moreover, a direct implication of such descriptor systems identification in the robust feedback linearization (RFL) control problem is underlined. A numeric example validates the proposed method of descriptor system identification applied to solve an RFL problem.

Keywords: Engineering Education, Machine Learning, Cloud Computing
Abstract: Digitalization and the use of artificial intelligence (AI) in education could help both learners and teachers. In fact, young people can be considered digital natives, having, and using digital technologies from an early age to play, communicate and access information. Teachers can use these digital technologies to create attractive teaching activities, to streamline and optimize their teaching work, but also to help learners cope with the digital transformation (known also as digit(all)ization – a portmanteau for two trending terms, digitization and digitalization) inherent in digital literacy. The barriers of resistance to change and digital transformation can be overcome through digital literacy, as well as the correct understanding and application of digitalization and artificial intelligence, but also providing confidence to users of digital technologies. A study was conducted on the perception of digitization, digital transformation, and the use of artificial intelligence in two Romanian universities.

Keywords: Engineering Education, Control Systems Design, Control Applications
Abstract: This paper presents the latest results of a methodology applied in 2022 and 2023 to a group of Processes and Services Automation students. This course is an elective in the Bachelor's Degree in Electrical Engineering. The methodology is oriented towards professional skills and was developed in collaboration with industry. It is worked by stages that include activities such debate forums, field trips and technical classes. These were complemented with commercial and industrial scale workshops and projects that exposes students to real processes. The results showed that the methodology helped to integrate theory with technical abilities and provided opportunities for students to improve their professional profiles.

Keywords: Engineering Education, Web services and applications, Software Methods and Tools
Abstract: Gamification can be an important driver for motivating students to attend peer assessment activities. As existing literature shows, the research in this area is quite limited and in the few available papers the gamified actions are usually submitting assignment solutions and evaluating peers’ work. In this paper, we propose an extensive gamification model designed to encourage students to perform all types of peer assessment related activities. The model makes use of well-known gamification elements, such as experience points, levels, badges, and leaderboard, but also employs additional elements, more rarely available in other models, such as progress bars, timers, increasing difficulty, virtual identity, or notifications. The proposed model follows a circular workflow made of four main stages. Initially, the student performs peer assessment activities. Next, upon completing specific activities, the learner gains various badges. Subsequently, each badge awards the learner with a certain amount of experience points, allowing different levels of expertise to be reached: novice, intermediate, advanced, and expert. Finally, the experience points also enable the student to acquire a higher rank in the leaderboard compared to other learners who engage less in peer assessment activities. Noteworthy, aiming at a higher rank promotes competition which can be a driver for students to be more actively involved with the peer assessment activities, leading to a virtuous circle process.

Keywords: Internet of Things, Embedded Systems, Cloud Computing
Abstract: This paper presents the development and implementation of an IoT-based system for monitoring and controlling temperature in a bioreactor. The system comprises an ESP8266 microcontroller, an OLED display, a solid-state relay (SSR), a temperature sensor, and a peltier module. The entire system operates an air or water pump which delivers warm or cool into the bioreactor by using a bypass. This work emphasizes the electronic design aspects and the implementation of IoT capabilities. By utilizing the microcontroller's PWM signal to modulate the SSR output, the system achieves precise temperature regulation. The IoT functionality enables real-time data acquisition and transmission to a cloud-based platform, facilitating remote monitoring and control of the bioreactor operation.

Keywords: Internet of Things, Machine Learning, Smart Cities/Houses/Grids
Abstract: Smart homes use a variety of technology that automate tasks and increase convenience in our daily lives. Functions performed by the devices can be improving security access, controlling lighting and adjusting temperature. Given the importance of convenience and cost efficiency in such settings, as well as the large number of devices involved, it is necessary to monitor power usage in smart homes. Furthermore, rising energy consumption increases the carbon footprint, exacerbates climate threats and puts a strain on energy supplies. As a result, monitoring energy use becomes critical for guaranteeing sustainability and efficiency. In this paper, we analyzed the energy consumption and usage in a home with multiple smart devices that are connected to a smart meter and we tried to offer optimization. This research focuses on optimizing energy usage while balancing user convenience, incorporating factors such as air pressure, dew point, and wind speed. To evaluate the optimization process, a hybrid approach using Linear Regression, Random Forest Regression and Gradient Boosting was developed. The study aimed to optimize energy consumption by predicting net energy usage. By utilizing predictions and optimized control strategies, smart home appliances can be managed proactively and efficiently.

Keywords: Machine Learning
Abstract: This work intends/aims to analyze the process of detection and binary classification of emotional states (neutral, happy and sad), associated with thermal facial images from a local database. An important procedure when it comes to the detection/classification of people and their associated emotional states is the extraction and analysis of features. In this work, Run-Length Features were used, which are a set of statistical measures that quantify the distribution of consecutive homogeneous regions within an image. These statistical measures/features are then used as input to a TPOT AutoML, which is a Python-based open-source machine learning library that utilizes low-code techniques to automate the process of creating and managing machine learning workflows. Afterwards, the classifier is selected and the results are validated with bagging method. There are three categories of images used in the detection and classification process, i.e., the face area, the mouth area and the eyes area. The input thermal images are organized in paired classes, i.e., normal with sad and normal with happy. The preliminary results show particularly encouraging classification accuracy (ACC).

Keywords: Machine Learning, Intelligent Systems
Abstract: Recent advancements in medical imaging technologies with a particular focus on the integration of artificial intelligence in image analysis, show potential in tackling clinical challenges related to the detection of breast cancer, evaluating treatment responses, and monitoring the progression of the disease. In this paper, a new algorithm based on bands that cross the lesion (three verticals, and three horizontals), with widths of 2, 4, and 8 pixels to classify breast lesions is proposed. A new octave band analysis is proposed to optimize the model’s feature extraction. Thus, each selected band is split into twelve sub-bands, and seventy-two features are obtained. The vector features dimensionality is reduced based on the features’ meaningfulness. To overcome the various machine learning models that fail to accurately classify these images due to their complex and diverse nature, several Automated Machine Learning libraries (AutoML) with PyCaret are used. They tune the hyperparameters for each selected classifier. For each selected band only eight features feed an AutoML PyCaret algorithm. In the case of the band having a width of 4 pixels, the binary classification results provide an accuracy of 0.812, an Area Under the Curve (AUC) of 0.844, and an F1-score of 0.867. The results are validated by the bagging method. It provided for the test dataset an accuracy of 0.813.

Keywords: Machine Learning, Software Methods and Tools, Other Topics
Abstract: As urban traffic volume rises, traffic incidents emerge as a pressing concern for cities. This study delves into the utilization of explainable artificial intelligence (XAI) methodologies alongside visual aids to forecast injury severity in traffic incidents. It specifically hones in on pivotal variables such as collision type, surface condition, weather, traffic control, and vehicle body type, all of which may exert influence on injury severity. Leveraging machine learning (ML) models, such as Decision Tree and Gradient Boosting, known for their capacity to unveil intricate relationships while maintaining interpretability, to advance an in-depth comprehension of the purposefulness of applying Visual Explainable AI (vXAI) methods in the thoroughly evaluation of these critical factors, thereby furnishing invaluable insights into enhancing road safety measures.

Keywords: Machine Learning, Automotive Control Systems, Control Systems Design
Abstract: The car suspension system has a major impact over the drivers’ safety, since the public road might suffer various elevation changes and damage. This work focuses on finding optimal parameters of suspensions models, which are responsible for both safety and comfort of passengers while driving. We considered 4 different suspension architectures: 2-DOF, 3-DOF, 4-DOF quarter car and 5-DOF half car. These optimal configurations have been obtained using relatively recent proposed bio-inspired algorithms such as RDA, IRDA and MOGWO. All the simulations have been performed on different road profiles and different types of cars. Additionally, we propose 6 meta-optimization schemes called super-position (SP1 – SP6), that is to combine algorithms (RDA+MOGWO) in order to enhance the quality of the obtained configurations. The results show that the proposed super-position methods of RDA+MOGWO outperform previous similar methods, such as NSGA-II+SPEA2, by an increase of 3.44 % in quality.

Keywords: Machine Learning, Manufacturing Systems, Computer Vision
Abstract: In today's industrial landscape, factory workers often face challenges in interpreting complex defect detection data, particularly juniors with limited experience. This article explores the innovative application of language models (LLMs) in simplifying the interpretation of defect detection results in industrial settings. By leveraging ChatGPT's natural language processing (NLP) capabilities, the system translates defect data into plain language. This transformation allows junior worker to quickly understand, respond to, and learn from various defect scenarios, enhancing decision-making and operational efficiency on the factory floor. Moreover, this approach contributes to generating and collecting human readable and structured data from factory floors. The article outlines the technical process of integrating ChatGPT with defect detection systems, highlighting its real-time assistance and cost basis. It also addresses the accuracy and reliability of ChatGPT's interpretations, alongside ethical and privacy considerations in handling sensitive factory data. The conclusion highlights the pivotal role and potential of LLMs in democratizing access to complex technical data, positioning them as indispensable tools for cultivating a more informed and efficient workforce in industrial settings.

Keywords: Machine Learning, Statistical Learning, Smart Cities/Houses/Grids
Abstract: Technologies supporting smart grid vision and renewable energy management are reshaping traditional energy system distribution, transforming modern buildings into active energy agents capable of generating, storing and trading energy. Furthermore, integrating electric vehicles (EVs) into the energy ecosystem presents both challenges and opportunities. Buildings can support EV charging, and optimized energy management systems can ensure energy efficiency, with an important compo- nent of forecasting both generation and vehicle charging needs. This paper aims to develop a Lithium-Ion Battery State of Charge estimation method based on a neural network model, capable of dealing with the main difficulties encountered by the usual methods. After a brief introduction in section I, section II outlines our contribution to the state of the art in electric vehicle state of charge estimation with neural networks. Section III presents the conceptualized methods for the stated problem and exemplification of the implemented methods. The main findings are highlighted in Section IV along with parameter estimation. Section V concludes the paper with perspectives on future work.

Keywords: Embedded Systems
Abstract: Virtually all digital IC platforms today are based on flexible programmable processor cores, with a trend towards multi/manycore architectures comprising 100+ cores. This trend is imposed by high performance and energy efficiency demands. Specifically in competitive embedded application domains like smartphones, mobile radio infrastructure, IoT and automotive, there are tight efficiency constraints on power, energy, timing, design cost, and security of the underlying HW/SW platforms. The need for flexibility and efficiency leads to heterogeneous platform architectures, composed of off-the-shelf (yet partially customizable) IP cores and custom application-specific processors, such as DSPs or AI accelerators. This presentation covers various advanced system-level design methodologies and tools for managing the skyrocketing HW/SW platform design complexity, while simultaneously optimizing systems and components for performance, power, security and cost. Special focus will be on the implications of current megatrends like RISC-V and neuromorphic computing. We will also exemplify how to maximize research impact via academia-to-industry technology transfer in the above domains.