Search

Ensuring the safety of workers at workplaces is a crucial task for every company. The usage of personal protective equipment represents the basic form of protection. Hard hats are very useful in protecting head from injuries. However, workers often neglect the importance of wearing safety helmets and do not wear them. Systems for monitoring and detecting unsafe behaviors can be very helpful for maintaining security. For that purpose, this research examines the success of the application of the latest YOLO algorithm for detecting the presence of safety helmets on workers that can be applied in those systems. Two models with different numbers of parameters are trained for this purpose – YOLOv9c and YOLOv9e. The results showed that YOLOv9c model achieved mean average precision of 97.2%, 93%, and 92.9% in training, validation, and testing, respectively, while YOLOv9e reached slightly higher mean average precisions of 97.5% in training, 93.4% in validation and 93.4% in testing.

Compared to methodology provided by standards concerning thermal comfort, by using models based on various approximation methods or artificial intelligence, it may be possible to ensure more time efficient and accurate calculation of thermal comfort indices. The aim of this study is to compare Predicted Mean Vote (PMV) computation model established by using multiple linear regression and trained artificial neural network, from the standpoint of accuracy. Both models are established on the basis of the same dataset which consists of 400 combinations of 4 thermal comfort parameters. These parameters are the air temperature, mean radiant temperature, relative humidity and clothing resistance, while activity level and air velocity are adopted as 1.1 met (office typing activity) and 0.05 m/s, respectively, and are considered constant values for selected type of indoor environment. Clothing resistance is adopted as 0.5 clo for summer period and 1.0 clo for winter period, while the air temperature, mean radiant temperature and relative humidity are values which are randomly generated within appropriately selected ranges. Taking into account that coefficients of determination which correspond to it are over 95%, resulting first degree polynomial relation obtained by using multiple linear regression can be considered a satisfactory approximation of PMV model as it is given in ASHRAE Standard 55-2020. Furthermore, there are certain input value combinations for which PMV values obtained by using this model coincide with the ones calculated by using algorithm which is provided by standard. However, results obtained by using trained neural network with one hidden layer coincide with PMV values calculated on the basis of ASHRAE Standard 55-2020 for each input value combination. Therefore, from the standpoint of accuracy, it is concluded that neural network provides significantly better approximation of PMV model.

This paper provides a nonlinear technique that uses a fuzzy inference system and neural networks for the identification purposes of heat flow transfer in the chamber. Firstly, linear models are obtained by transfer functions with delay using Matlab identification tools for heat exchange. Three different transfer functions are provided (for three sensors in different positions along the chamber), and after it has been concluded that the second model has the smallest error, it is tested using different input. In this case, the linear model failed to represent the behaviour of the system precisely, making the error more than 1.5 C in the steady state. This was expected because linear models are trustworthy only around certain operating ranges. In order to make the new model, which will be unique and valid in the whole state space, another identification method using an adaptive neuro-fuzzy inference system (ANFIS) was presented. Furthermore, for the best performance, the ANFIS architecture was found using one of the most famous population-based optimizations: the genetic evolutionary algorithm. With two inputs and 70 parameters found by optimization (40 premises and 30 consequent) ANFIS greatly outperforms standard identification technique in terms of the mean square error. This nonlinear model was also tested on the different input, which was not used in the training process, and it was concluded that the nonlinear model identifies the real object with a neglectable error, which is 45 times smaller than the linear one.

This paper covers the application of artificial intelligence in agriculture. Technology development has enabled measurement, collecting and processing high quality big data. These data can be successfully used to significantly improve numerous segments of agriculture sector. The accent in this paper is given on the models used for energy use prediction. The application of Artificial Neural Networks with different structure is presented, such as Feedforward Neural Network, Radial basis Function Network and Adaptive Neuro-Fuzzy Inference System. Support Vector Machine model is also shown. The improvements of individual models are elaborated, through the analysis of the ensemble and hybrid approach. All of the proposed models are capable of solving complex problem of prediction of energy use based on real, measured data. Ensemble and hybrid models are promising, as it has been shown that the prediction accuracy is improved by combining different single models in proposed manner.

Chamber dryers are widely used in various industries in order to remove the moisture from solid materials efficiently. Optimizing the design and operational parameters of chamber dryers plays a crucial role in enhancing their performance and energy efficiency. In order to maintain the temperature at the desired level, it is necessary to implement a good control system. To be able to facilitate the process of finding and setting parameters of the controller, for many control algorithms it is essential to make the reliable model of the object. The aim is to develop both reliable and accurate predictive model that can assist in optimizing the design, structures, and inspection processes of chamber dryers, which will lead to enhanced energy efficiency, harvesting and improved drying performance. In this paper, the authors propose a novel approach for modeling heat flow transfer in chamber dryers using an Adaptive Neuro-Fuzzy Inference System (ANFIS). The Quanser chamber was selected as the object of the research because of how closely its geometry, material choice, and air flow resemble the structural properties of a dryer. To obtain the most realistic model possible, parameters of ANFIS were found using Particle Swarm Optimization algorithm. By incorporating historical operational data of experimental measurements, the ANFIS model can learn and adapt to the dynamic behavior of the dryer system.

The use of artificial intelligence in modern agriculture is on the rise, due to the fact that it provides a possibility for more efficient production, better decision making and reduction of the costs. This research takes into consideration the use of the convolutional neural networks for diagnosing plant illnesses based on the leaf image. Detection of plant diseases in the early phase can improve the quality of the food products and minimize the loses. Convolutional neural networks are a type of deep learning method that is one of the most used models for solving image recognition, classification and detection tasks. Therefore, it is justified to anticipate that they can be very effectively applied in the agriculture sector. This paper covers plant species that are the most significant for Serbian production. Various models have been presented and analyzed, while highlighting their advantages and disadvantages when applied for solving this task.

Konvolucione neuronske mreže su vrsta dubokih neuronskih mreža koje se koriste u zadacima klasifikacije, identifikacije, predikcije i detekcije objekata, a pogodne su za rad sa ulaznim podacima različitih dimenzija, kao što su signali, slike, video zapisi. O njihovom značaju svedoči činjenica da su u upotrebi više od bilo koje druge vrste dubokih mreža. Upravo zbog toga se konstantno radi na razvoju novih algoritama koji usavršavaju postojeće modele ili kreiranju novih modela koji ubrzavaju ili poboljšavaju proces učenja. Primenu ostvaruju u najrazličitijim oblastima nauke i industrije zbog mogućnosti postizanja visoke tačnosti i jednostavnosti implementacije. U ovom radu se predstavlja struktura konvolucionih mreža, a naročito se razmatraju novosti u sferi istraživanja konvolucionog sloja, gde se tumače različiti tipovi konvolucija. Takođe, posebno se obraća pažnja na upotrebu ovih mreža u sistemima automatskog upravljanja poslednjih godina, kao rezultata pojave Industrije 4.0. Prilikom analiziranja naučnih radova, primena konvolucionih mreža je razgraničena prema dimenzionalnosti ulaznih podataka, odnosno prema dimenzionalnosti mreža i zadacima koje je moguće rešiti pomoću njih.

A company’s development performance and growth may be impacted by a wide range of different factors, which unquestionably affect number of the employees in the loop. Taking into account all influencing factors, companies would benefit if have possibility to predict the degree of change in the number of employees in future period in order to adjust their internal strategy or to make appropriate decisions that enable the survival and progress of the company in the market. The aim of this research is to predict the change in number of employees based on current state of contingency and quality management factors, using information obtained from a survey of 67 different companies from Serbia. In the first part of the research, a correlation analysis is used with the aim to identify the specific contingency and quality management factors that are most closely associated to the subject of interest, which is, in this case, degree of change in the number of employes. The second part of the research involves feedforward neural network training for prediction of the degree of change in number of employees based on feature extraction of main factors. The training accuracy that proposed network achieved is 77.36%, while testing accuracy amounts 71.43%.

In this research, control of the Direct Current motor is accomplished using a neuro controller in the Internal Model Control scheme. Two Feed Forward Neural Networks are trained using historical input-output data. The first neural network is trained to identify the object's dynamic behavior, and that model is used as an internal model in the control scheme. The second neural network is trained to obtain an inverse model of the object, which is applied as a neuro controller. Experiment is conducted on the real direct current motor in laboratory conditions. Obtained results are compared to those achieved by implementing the Direct Inverse Control method with the same neuro controller. It was demonstrated that the proposed control method is simple to implement and the system robustness is achieved, which is a great benefit, aside from the fact that no mathematical model of the system is necessary to synthesize the controller of the real object.

Adaptivni neuro fazi sistemi zaključivanja (eng. Adaptive Neural Fuzzy Inference Systems) ANFIS imaju sve veću tendenciju upotrebe u naučnim istraživanjima i praktičnim primenama. Digitalizacija proizvodnje i pojava Industrije 4.0 omogućila je razvoj ovog trenda, pre svega, zbog sposobnosti prilagođavanja zadatku integrisanjem veštačkih neuronskih mreža i fazi logike, čime se potencijalno mogu iskoristiti prednosti obe tehnike u jedinstvenim okvirima. Ovaj pristup olakšao je procese modelovanja, analize podataka, klasifikacije i upravljanja. Pogodnost ANFIS sistema, u odnosu na konvencionalne metode, se ogleda u mogućnosti predviđanja izlaza na osnovu skupa ulaza i baze pravila. Takođe, ovi sistemi su pogodni za korišćenje u upravljanju, jer pružaju mogućnost za podešavanje parametara upravljačkog sistema. U ovom radu je predstavljena struktura ANFIS sistema i dat je detaljan prikaz dosadašnjih dostignuća, kroz komparativnu analizu, pri čemu su istaknute neke moguće sfere interdisciplinarne primene. Razmatrane su mogućnosti za varijacije, poboljšanja i inovacije algoritma, kao i smanjenja složenosti same arhitekture mreže. Prikazani su predlozi za neke nove, još neiskorišćene kombinacije sa metaheurističkim metodama optimizacije. Konačno, date su bitne smernice o tome kada i gde je korisno primeniti ANFIS sisteme.

The study proposes a novel method for synthesis of a discrete-time parallel distributed compensation (PDC) controller for the nonlinear discrete-time Takagi–Sugeno (TS) fuzzy plant model. For each of the fuzzy plant model linear subsystems, a local linear first order proportional-sum (PS) controller is designed. The algebraic technique is used in two-dimensional parameter space, utilizing the characteristic polynomial of the row nondegenerate full transfer function matrix. Each system’s relative stability is accomplished in relation to the selected damping coefficient. The supplementary criterion is the minimal value of the performance index in the form of the sum of squared errors (SSE). However, unlike the traditional technique, output error is influenced by all simultaneous actions on the system: nonzero inputs and nonzero initial conditions. The full transfer function matrix of the system allows for the treatment of simultaneous actions of the input vector and unknown unpredictable initial conditions. In order to show the improvement caused by the application of a new optimization method that considers nonzero initial conditions, a comparison of PDC controllers designed under zero and nonzero initial conditions is given, where the system in both cases starts from the same nonzero initial conditions, which is often the case in practice. The simulation and experimental results on a DC servo motor are shown to demonstrate the suggested method efficiency.

This paper presents the proportional–derivative fuzzy controller for trajectory tracking of the gripping mechanism with two degrees of freedom. Aiming to achieve movement of the gripping mechanism without sudden starting and stopping, a polynomial velocity profile is utilized. The African vultures optimization, as one of the latest metaheuristic algorithms, is used to obtain the optimal input/output scaling gains of the proposed fuzzy controller according to the selected fitness function. The results obtained by this algorithm are compared with the other three new and popular metaheuristic algorithms: the whale optimization, the ant lion optimization and the sine cosine algorithm. Moreover, a simulation study was done for the defined initial position and for the scenario where there is a certain deviation because the gripping mechanism is not at its original initial position. Finally, the robustness of the controller is tested for the case when the masses of the segments increase three times. The results revealed that the suggested controller was capable of dealing with nonlinearities of the gripping mechanism, initial position and parameter changes. The movement of the gripping mechanism is smooth and follows the defined trajectory.

This paper deals with the problem of classifying images of real and fake faces as it is impossible to distinguish them with the bare eye. Two different convolutional neural networks architecture models are applied. The first one is pre-trained VGG16 model, where transfer learning method is applied on our dataset. The architecture of the second model is based on VGG16 and represents its smaller and lighter version. Techniques such as learning rate decay, dropout and batch normalization was applied in training process. Comparison of obtained results of both models is made.

In this paper a novel metaheuristic method, artificial gorilla troops optimizer, is used in order to optimize classical proportional-integral controller for liquid level system, that has wide application in many industries. In optimization process nonlinear model of the system is used. Obtained results are provided. It is shown that optimized controller represents superior solution compared to classical controller.

In this paper artificial neural network is proposed as a method to classify defective and non-defective casting products in order to improve quality inspection process. Three different models of convolutional neural networks are trained and tested on dataset of submersible pump impeller images which has uneven number of image samples in each class. In order to inspect if slightly imbalanced classes have impact on result, two experiments are done. All of the models are ImageNet pre-trained networks, InceptionV3, Xception and MobileNetV2, where transfer learning method is applied with fine-tuning. Stochastic gradient descent algorithm is implemented for optimization. Obtained results of all models are presented and comparison is made.

Several studies dealing with position control of the DC motor have reported issues concerning friction force. This article demonstrates a nonlinear control and optimization strategy for position control of a series servo motor. Once it is empirically verified that the linear model does not adequately reflect the system, the model is upgraded from linear to nonlinear. In the course of the research, the nonlinear feedback linearizing the controller's behavior is examined. A grey wolf metaheuristic optimization algorithm is used to find the coefficients of the controller's gains. In this way, modern methods are applied to take a fresh look at the existing problem. Furthermore, performance for various targeted output signals is compared to show the approach proposed in the study. Also, a compara- tive analysis with whale optimization algorithm is performed. The experimental results acquired on the stated system are shown, and they validate the usage of the nonlinear control, demonstrating the effectiveness of using optimum feedback linearization in electrical machines.

Even though, it is mostly used by process control engineers, the temperature control remains an important task for researchers. This paper addressed two separate issues concerning model optimization and control. Firstly, the linear models for the three different operating points of the heat flow system were found. From these identified models a Takagi-Sugeno model is obtained using fixed membership functions in the premises of the rules. According to the chosen objective function, parameters in the premise part of Takagi-Sugeno fuzzy model were optimized using the grey wolf algorithm. Furthermore, by using the parallel distributed compensation a fuzzy controller is developed via the fuzzy blending of three proportional + sum controllers designed for each of the operating points. In order to evaluate performance, a comparison is made between the fuzzy controller and local linear controllers. Moreover, the fuzzy controllers from the optimized and initial Takagi-Sugeno plant models are compared. The experimental results on a heat flow platform are presented to validate efficiency of the proposed method.

The paper presents a new approach to the design of the classical PD controller for the plant in the closed loop control system. Proposed controller has two unknown adjustable parameters which are designed by the algebraic method in the two dimensional parameter space, by using the newly discovered characteristic polynomial of the row nondegenerate full transfer function matrix. The system relative stability, in regard to the chosen damping coefficient is achieved. The optimality criterion is the minimal value of the performance index and sum of squared error is taken as an objective function. The output error used in the performance index is influenced by all actions on the system at the same time: by nonzero initial conditions and by nonzero inputs. In the classical approach the output error is influenced only by nonzero inputs. Experimental results confirm that by taking into account nonzero initial conditions, an optimal solution is obtained, which is not the case with the classical method where optimization is performed at zero initial conditions.

The aim of this study is to investigate nonlinear DC motor behavior and to control velocity as output variable. The linear model is designed, but as it is experimentally verified that it does not describe the system well enough it is replaced by the nonlinear one. System's model has been obtained taking into account Coulomb and viscous friction in the firmly nonlinear environment. In the frame of the paper the dynamical analysis of the nonlinear feedback linearizing control is carried out. Furthermore, a metaheuristic optimization algorithm is set up for finding the coefficient of the proportional-integral feedback linearizing controller. For this purpose Gray wolf optimization technique is used. Moreover, after the introduction of the control law, analysis of the pole placement and stability of the system is establish. Optimized nonlinear control signal has been applied to the real object with simulated white noise and step signal as disturbances. Finally, for several desired output signals, responses with and without disruption are compared to illustrate the approach proposed in the paper. Experimental results obtained on the given system are provided and they verify nonlinear control robustness.

Currently, in the building sector there is an increase in energy use due to the increased demand for indoor thermal comfort. Proper energy planning based on a real measurement data is a necessity. In this study we developed and evaluated hybrid artificial intelligence models for the prediction of the daily heating energy use. Building energy use is defined by significant number of influencing factors, while many of them are difficult to adequately quantify. For heating energy use modelling, the complex relationship between the input and output variables is hard to define. The main idea of this paper was to divide the heat demand prediction problem into the linear and the non-linear part (residuals) by using (Afferent statistical methods for the prediction. The expectations were that the joint hybrid model, could outperform the individual predictors. Multiple linear regression was selected for the linear modelling, while the non-linear part was predicted using feedforward and radial basis neural networks. The hybrid model prediction consisted of the sum of the outputs of the linear and the non-linear model. The results showed that both hybrid models achieved better results than each of the individual feedforward and radial basis neural networks and multiple linear regression on the same dataset. It was shown that this hybrid approach improved the accuracy of artificial intelligence models.

У овој докторској дисертацији предмет истраживања је пројектовање савремених система управљања применом развојних програмабилних система и применом савремене теорије рачуна нецелог реда. Родригов приступ је примењен у циљу добијања одговарајућег математичког модела роботског система који је моделован као један повезан систем крутих тела који формирају кинематички ланац без гранања. За решавање проблема везаног кретања роботског система у случaјевима када је остварен контакт хватаљке са радном површи или кретање по задатој линији, применом Лагранжевих једначина друге врсте у коваријантном облику добијене су диференцијалне једначине везаног кретања роботског система са хватаљком. Други приступ за добијање диференцијалних једначина везаног кретања роботског система је остварен применом Лагранж-Даламберовог принципа у генералисаним координатама. При томе, роботски системи који остварују контакт са радном површи могу се моделирати као сингуларни системи целог и у општем случају нецелог реда. У оквиру савремене теорије управљања итеративно управљање путем учења (ИУУ- iterative learning control) представља једну моћну интелигентну методологију управљања који на итеративни начин побољшава понашање репетитивних процеса и динамичких система; а овде је од посебног интереса примена истог на роботске системе. Овде се предлаже један нови НРИУУ (ИУУ нецелог реда) закон PD2D типа у отвореној спрези за класу линеаризованих роботских система. Спроведена је feedback линеаризација објекта управљања где се у затвореној спрези добија линеарна зависност улаза и излаза. Даље се разматра проблематика управљања у отворено-затвореној спрези ИУУ нецелог реда сингуларним динамичким системом нецелог реда. Посебна пажња је посвећена и примени закона ИУУ нецелог реда типа PD PD  NeuroArm роботским системом. Након тога разматран је избор и имплементација хардвера потребног за управљање NeuroArm роботске руке и то тако што за развојну плочу је изабран Beaglebone Black. Пројектује се и израђује пло, In this doctoral dissertation the subject of research is development of modern robot control systems using development programmable systems based on non-integer calculus theory. Rodriguez's approach was applied in order to obtain an appropriate mathematical model of a robotic system that was modeled as a single connected system of rigid bodies forming a kinematic chain without branching. To solve the problem of bound motion of the robotic system in cases when the gripper is in contact with the work surface or motion along a given line, by applying Lagrangian equations of second order in covariant form, differential equations of bound motion of the robotic system with the gripper are obtained. Another approach for obtaining differential equations of bound motion of a robotic system was realized by applying the Lagrange-Dalambert principle in generalized coordinates. In doing so, robotic systems that make contact with the work surface can be modeled as singular systems of the integer order and in the general case of the non-integer order. Within modern control theory, iterative learning control (ILC) is a powerful intelligent control methodology that iteratively improves the behavior of repetitive processes and dynamic systems; and here is of special interest to apply it to robotic systems. Here, a new FOILC (fractional order ILC) law of the PD2D type in open loop is proposed for a class of linear robotic systems. A feedback linearization of the control object was performed, where a linear dependence of inputs and outputs was obtained in a closed loop. The problem of control in open-closed loop of ILC of non-integer order using singular dynamic system using non-integer order is further considered. Special attention is given to the application of the ILC law of the non-integer order of PD PD  type for NeuroArm robotic system. After that, the selection and implementation of the hardware needed to control the NeuroArm robotic arm was discussed by selecting Beaglebone B

This paper deals with liquid level control as one of the frequent problems in industry. Several classical methods for tuning a PID like controller were applied. Furthermore, parameters for the controller were optimized using grey wolf optimizer. In addition to the classical controller, fuzzy PID like controller has also been designed and optimized using the same optimization algorithm. Experimental results obtained on the tank system are provided.

The manufacturing industry frequently deals with the problem of gripping mechanism and their movement optimization. This paper presents an optimization methodology based on the whale optimization algorithm to design an optimal fuzzy PD controller of a two - link gripping mechanism (robot arm) as a part of mobile robot working cycle. The dynamical analysis of gripping mechanism investigates a coupling relation between the joint torques applied by the actuators and the position and acceleration of the robot arm. The proposed fuzzy controller optimizes the trajectory of the robot’s end effector. Additionally, a simulation study was done for the specific initial case and the trapezoidal velocity profile was generated. Based on the predefined acceleration, movement of the robot arm is shown to be smooth and without an abrupt braking.

This paper presents a feedback linearization controller for trajectory tracking of two degrees of freedom (2DOF) gripping mechanism. To reach this goal, after deriving the dynamical equations of the gripping mechanism, the feedback linearization approach is utilized to change the nonlinear dynamics to a linear one. Classical proportional-derivative controller with feedback linearization is applied for positioning and tracking control. Furthermore, in order to achieve movement of the mechanism without the sudden stopping at the desired point, a trapezoidal velocity profile is used to obtain desired trajectory. Numerical simulations using Matlab/Simulink successfully demonstrate the effectiveness of the proposed method.

In this paper the nonlinear feedback control system is presented for the speed control in direct current - DC motor. Nonlinear functions of dead zone, Coulomb and viscous friction were investigated and used for obtaining the mathematical model. The effectiveness and the comparison between linear and nonlinear control signal have been confirmed using Matlab/Simulink software. From the conclusions, based on the experimental results, it is easy to see that nonlinear control system is more acceptable and has a better performance for speed control. The validity of using feedback linearization in DC motors has been proven.

U radu je prikazan mogući pristup inteligentnog upravljanja servo motora jednosmerne struje korišćenjem veštačkih neuronskih mreža. Pored primene jedne od najzastupljenijih tehnika veštačke inteligencije, u radu je predloženo i dato matematičko modelovanje ovog široko zastupljenog objekta automatskog upravljanja. Takođe, u cilju prevazilaženja nedostataka vezanih za konvencionalno upravljanje servo motora jednosmerne struje, u radu su iskorišćene sposobnosti veštačkih neuronskih mreža da mogu da generalizuju i aproksimiraju izlaze ovog objekta primenom mašinskog učenja kroz proces njihovog obučavanja. Predloženi pristup, prvo je analiziran putem simulacije, a potom je i eksperimentalno verifikovan na primeru dva od četiri modela koji su razmatrani., In this paper the possible approach of DC servo motor intelligent control is presented by using artificial neural networks. Besides the application of one of the most important techniques of artificial intelligence, this paper suggests and gives the mathematical modelling of widely used object of automatic control. Also, this paper uses abilities of artificial neural networks in order to generalize and approximate the outputs of this object by applying machine learning through the process of its training, aiming to overcome the faults connected to the conventional control of the DC servo motor. The suggested approach, was analyzed firstly by simulation, and then it was experimentally verified in two out of four models which were taken into consideration.

The process industries have continually combated the problem concerning liquid level control. Effective control of a system depends largely on the accuracy of the mathematical model that predicts its dynamic behavior. In this paper the Takagi-Sugeno fuzzy model for the coupledtank system was acquired based on empirical technique. Furthermore, a metaheuristic algorithm was used as an optimizer on the coupled-tank model. Then, a juxtaposition was made when comparing models which were identified and optimized, leading to satisfactory results. Experimental results obtained on the coupled-tank system are provided.

A solution to the problem of structurally variable control synthesis for time-invariant Lurie systems is of both theoretical and engineering significance. A theorem on absolute stability of a set applied to Lurie systems solves the problem. The theorem is basic for proving control algorithms that ensure asymptotic stability of a sliding subspace, or its stability with finite time reachability. The selection of the sliding subspace can be such that the system nonlinearity does not influence the sliding motion. The proposed stabilising control algorithms have been tested applying them to a mathematical system and a real system, DC servo motor.