Sertaç Savaş
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Sertaç Savaş has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Şahin Yıldırım
Department of Mechatronics Engineering, Faculty of Engineering, Erciyes University, Kayseri, Turkey
3 shared publications - Ahmet Kırnap
Department of Mechatronics Engineering, Erciyes University, 38039 Kayseri, Turkey
2 shared publications - Banu Ulu
Department of Software Engineering, Faculty of Engineering, Architecture and Design, Kayseri University, Kayseri, Türkiye
1 shared publication - Burak Ulu
Department of Mechatronics Engineering, Erciyes University, 38039 Kayseri, Turkey
1 shared publication Ikbal Eski
1 shared publication- Kamber Külahcı
Mechatronics Engineering, Graduate School of Natural and Applied Sciences, Erciyes University, 38039 Kayseri, Türkiye
1 shared publication
Latest Publications
- 15 May 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (3) , 1008-1021
The RNN-Based Deep Learning Model Design to Predict ICU Occupancy Rate and Number of Intubated Patients for Effective Healthcare System Management
Epidemics have been one of the most significant health threats in human history. Today, as new epidemics such as COVID-19 and Monkeypox emerge, it is critical for healthcare systems to be prepared for such crises. Predicting the progression of an epidemic is essential for healthcare systems to respond effectively. In this study, an artificial intelligence model design is proposed to predict the number of intubated and intensive care unit patients during a pandemic. LSTM, BiLSTM and GRU models belonging to the RNN family of machine learning algorithms are used in the predictor design and the grid search method is applied for hyperparameter optimization. In the design of the proposed model, the number of patients intubated and treated in intensive care during the COVID-19 pandemic in Türkiye is used as the dataset. The results show that the GRU model achieves the best performance with RMSE values of 15.7277 and 6.6494 for intensive care and intubated patient numbers, respectively, using an 80/20% train/test ratio. Similarly, GRU provides the highest accuracy with RMSE values of 9.9085 and 7.0271 for the same datasets using a 90/10% train/test ratio. These findings reveal that the simple structure of the GRU model, with fewer parameters and reduced computational complexity, is compatible with the dataset and provides better generalization capability, demonstrating that the deep learning model we designed can be used to predict the number of intensive care and intubated patients in order to facilitate healthcare system management in epidemic processes.
- 15 May 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (3) , 1008-1021
The RNN-Based Deep Learning Model Design to Predict ICU Occupancy Rate and Number of Intubated Patients for Effective Healthcare System Management
Epidemics have been one of the most significant health threats in human history. Today, as new epidemics such as COVID-19 and Monkeypox emerge, it is critical for healthcare systems to be prepared for such crises. Predicting the progression of an epidemic is essential for healthcare systems to respond effectively. In this study, an artificial intelligence model design is proposed to predict the number of intubated and intensive care unit patients during a pandemic. LSTM, BiLSTM and GRU models belonging to the RNN family of machine learning algorithms are used in the predictor design and the grid search method is applied for hyperparameter optimization. In the design of the proposed model, the number of patients intubated and treated in intensive care during the COVID-19 pandemic in Türkiye is used as the dataset. The results show that the GRU model achieves the best performance with RMSE values of 15.7277 and 6.6494 for intensive care and intubated patient numbers, respectively, using an 80/20% train/test ratio. Similarly, GRU provides the highest accuracy with RMSE values of 9.9085 and 7.0271 for the same datasets using a 90/10% train/test ratio. These findings reveal that the simple structure of the GRU model, with fewer parameters and reduced computational complexity, is compatible with the dataset and provides better generalization capability, demonstrating that the deep learning model we designed can be used to predict the number of intensive care and intubated patients in order to facilitate healthcare system management in epidemic processes.
- 29 March 2026
- Published by Firat Universitesi in Fırat Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 38 (1) , 425-434
Comparative Analysis of Statistical and Deep Learning Models for Daily Climate Forecasting: A Case Study on the Delhi Dataset
As the impacts of climate change continue to intensify, accurate forecasting of meteorological parameters is of critical importance for many applications ranging from agricultural planning to disaster management. This study presents a systematic comparison of seven forecasting models – two baseline statistical (ARIMA, SARIMA) and five deep learning (LSTM, CNN, MLP, CNN-BiLSTM-Attention, and Wavelet-CNN-LSTM) – for daily prediction of mean temperature, humidity, wind speed, and mean pressure using Delhi climate data covering 2013–2017. Hyperparameters for all models are optimized via Random Search, and 10 independent runs with different random seeds are conducted to ensure statistical reliability. Model performance is evaluated using RMSE, MAE, MAPE, MASE, and R² metrics, while the Friedman test with Nemenyi post-hoc analysis is employed to assess statistical significance of performance differences. Results indicate that the Wavelet-CNN-LSTM model achieves the best overall ranking, followed by LSTM, MLP, and CNN, all of which performed significantly better than the ARIMA and SARIMA models. High prediction accuracy is obtained for temperature (R² = 0.922) and pressure (R² = 0.9), whereas wind speed remains the most challenging variable for all models (R² < 0.2) due to its inherently stochastic nature. Predictability analysis confirms that the chaotic behavior of wind speed, characterized by low autocorrelation and high variability, imposes fundamental limits on deterministic forecasting at daily resolution.
- 29 March 2026
- Published by Firat Universitesi in Fırat Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 38 (1) , 425-434
Comparative Analysis of Statistical and Deep Learning Models for Daily Climate Forecasting: A Case Study on the Delhi Dataset
As the impacts of climate change continue to intensify, accurate forecasting of meteorological parameters is of critical importance for many applications ranging from agricultural planning to disaster management. This study presents a systematic comparison of seven forecasting models – two baseline statistical (ARIMA, SARIMA) and five deep learning (LSTM, CNN, MLP, CNN-BiLSTM-Attention, and Wavelet-CNN-LSTM) – for daily prediction of mean temperature, humidity, wind speed, and mean pressure using Delhi climate data covering 2013–2017. Hyperparameters for all models are optimized via Random Search, and 10 independent runs with different random seeds are conducted to ensure statistical reliability. Model performance is evaluated using RMSE, MAE, MAPE, MASE, and R² metrics, while the Friedman test with Nemenyi post-hoc analysis is employed to assess statistical significance of performance differences. Results indicate that the Wavelet-CNN-LSTM model achieves the best overall ranking, followed by LSTM, MLP, and CNN, all of which performed significantly better than the ARIMA and SARIMA models. High prediction accuracy is obtained for temperature (R² = 0.922) and pressure (R² = 0.9), whereas wind speed remains the most challenging variable for all models (R² < 0.2) due to its inherently stochastic nature. Predictability analysis confirms that the chaotic behavior of wind speed, characterized by low autocorrelation and high variability, imposes fundamental limits on deterministic forecasting at daily resolution.
- 15 March 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (2) , 952-961
PID Controller Tuning for Heat Exchanger Systems: A Comparative Study of Classical, Model-Based, Optimization, and Reinforcement Learning Approaches
Heat exchangers are among the fundamental components of industrial processes, and effective temperature control is critical for process efficiency and product quality. This study presents a comparative analysis of PID controller tuning methods for a heat exchanger system from four different paradigms. As the classical approach, Ziegler–Nichols (ZN); as the model-based approach, Internal Model Control (IMC); as the metaheuristic optimization approach, Particle Swarm Optimization (PSO); and as the reinforcement learning approach, Soft Actor-Critic (SAC) are investigated. For the ZN and IMC methods, a single run is executed using fixed hyperparameters, whereas a two-stage methodology is followed for the PSO and SAC methods. Hyperparameter selection is performed via random search, evaluating 20 configurations and selecting the parameters that yield the lowest ITAE. Using the chosen hyperparameters, 20 independent runs are conducted, and statistical analysis is performed. For all tuning methods, the controller's tracking performance for step, sinusoidal, triangular, and square-wave reference signals is computed using RMSE, IAE, ISE, and ITAE metrics. The results show that the PSO-PID method achieves the lowest error metrics for all reference signals. In the step response, PSO provides 90.8% improvement in ITAE and 25.2% improvement in RMSE compared to ZN. The Wilcoxon rank-sum test indicates that the differences between PSO and SAC are statistically significant for most metrics (P
- 15 March 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (2) , 952-961
PID Controller Tuning for Heat Exchanger Systems: A Comparative Study of Classical, Model-Based, Optimization, and Reinforcement Learning Approaches
Heat exchangers are among the fundamental components of industrial processes, and effective temperature control is critical for process efficiency and product quality. This study presents a comparative analysis of PID controller tuning methods for a heat exchanger system from four different paradigms. As the classical approach, Ziegler–Nichols (ZN); as the model-based approach, Internal Model Control (IMC); as the metaheuristic optimization approach, Particle Swarm Optimization (PSO); and as the reinforcement learning approach, Soft Actor-Critic (SAC) are investigated. For the ZN and IMC methods, a single run is executed using fixed hyperparameters, whereas a two-stage methodology is followed for the PSO and SAC methods. Hyperparameter selection is performed via random search, evaluating 20 configurations and selecting the parameters that yield the lowest ITAE. Using the chosen hyperparameters, 20 independent runs are conducted, and statistical analysis is performed. For all tuning methods, the controller's tracking performance for step, sinusoidal, triangular, and square-wave reference signals is computed using RMSE, IAE, ISE, and ITAE metrics. The results show that the PSO-PID method achieves the lowest error metrics for all reference signals. In the step response, PSO provides 90.8% improvement in ITAE and 25.2% improvement in RMSE compared to ZN. The Wilcoxon rank-sum test indicates that the differences between PSO and SAC are statistically significant for most metrics (P
- Open Access
Statistical Feature Engineering for Robot Failure Detection: A Comparative Study of Machine Learning and Deep Learning Classifiers
Industrial robots are widely used in critical tasks such as assembly, welding, and material handling as core components of modern manufacturing systems. For the reliable operation of these systems, early and accurate detection of execution failures is crucial. In this study, a comprehensive comparison of machine learning and deep learning methods is conducted for the classification of robot execution failures using data acquired from force–torque sensors. Three different feature engineering approaches are proposed. The first is a Baseline approach that includes 90 raw time-series features. The second is the Domain-6 approach, which consists of 6 basic statistical features per sensor (36 in total). The third is the Domain-12 approach, which comprises 12 comprehensive statistical features per sensor (72 in total). The domain features include the mean, standard deviation, minimum, maximum, range, slope, median, skewness, kurtosis, RMS, energy, and IQR. In total, ten classification algorithms are evaluated, including eight machine learning methods and two deep learning models: Support Vector Machines (SVM), Random Forest (RF), k-Nearest Neighbors (KNN), Artificial Neural Network (ANN), Naive Bayes (NB), Decision Trees (DT), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM-LGBM), as well as a One-Dimensional Convolutional Neural Network (CNN-1D) and Long Short-Term Memory (LSTM). For traditional machine learning algorithms, 5 × 5 nested cross-validation is used, whereas for deep learning models, 5-fold cross-validation with a 20% validation split is employed. To ensure statistical reliability, all experiments are repeated over 30 independent runs. The experimental results demonstrate that feature engineering has a decisive impact on classification performance. In addition, regardless of the feature set, the highest accuracy (93.85% ± 0.90) is achieved by the Naive Bayes classifier using the Baseline features. The Domain-12 feature set provides consistent improvements across many algorithms, with substantial performance gains. The results are reported using accuracy, precision, recall, and F1-score metrics and are supported by confusion matrices. Finally, permutation feature importance analysis indicates that the skewness features of the Fx and Fy sensors are the most critical variables for failure detection. Overall, these findings show that time-domain statistical features offer an effective approach for robot failure classification.
Open AccessStatistical Feature Engineering for Robot Failure Detection: A Comparative Study of Machine Learning and Deep Learning Classifiers
Industrial robots are widely used in critical tasks such as assembly, welding, and material handling as core components of modern manufacturing systems. For the reliable operation of these systems, early and accurate detection of execution failures is crucial. In this study, a comprehensive comparison of machine learning and deep learning methods is conducted for the classification of robot execution failures using data acquired from force–torque sensors. Three different feature engineering approaches are proposed. The first is a Baseline approach that includes 90 raw time-series features. The second is the Domain-6 approach, which consists of 6 basic statistical features per sensor (36 in total). The third is the Domain-12 approach, which comprises 12 comprehensive statistical features per sensor (72 in total). The domain features include the mean, standard deviation, minimum, maximum, range, slope, median, skewness, kurtosis, RMS, energy, and IQR. In total, ten classification algorithms are evaluated, including eight machine learning methods and two deep learning models: Support Vector Machines (SVM), Random Forest (RF), k-Nearest Neighbors (KNN), Artificial Neural Network (ANN), Naive Bayes (NB), Decision Trees (DT), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM-LGBM), as well as a One-Dimensional Convolutional Neural Network (CNN-1D) and Long Short-Term Memory (LSTM). For traditional machine learning algorithms, 5 × 5 nested cross-validation is used, whereas for deep learning models, 5-fold cross-validation with a 20% validation split is employed. To ensure statistical reliability, all experiments are repeated over 30 independent runs. The experimental results demonstrate that feature engineering has a decisive impact on classification performance. In addition, regardless of the feature set, the highest accuracy (93.85% ± 0.90) is achieved by the Naive Bayes classifier using the Baseline features. The Domain-12 feature set provides consistent improvements across many algorithms, with substantial performance gains. The results are reported using accuracy, precision, recall, and F1-score metrics and are supported by confusion matrices. Finally, permutation feature importance analysis indicates that the skewness features of the Fx and Fy sensors are the most critical variables for failure detection. Overall, these findings show that time-domain statistical features offer an effective approach for robot failure classification.
- Open Access
- 31 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (3) , 1453
Reinforcement Learning-Based PID Gain Optimization for Delta Parallel Robot Trajectory Tracking
In this study, a PID gain tuning approach using Deep Deterministic Policy Gradient (DDPG), a reinforcement learning (RL) algorithm, is proposed for trajectory tracking of delta parallel robots. Owing to their 3-degree-of-freedom (3-DOF) parallel kinematic structure, delta robots offer higher stiffness, precision, and speed capabilities than serial manipulators; they are therefore widely used in high-speed pick-and-place applications due to their low moving mass and the stiffness provided by the closed-chain mechanism. In this study, the proposed DDPG-PID approach is comparatively investigated against the conventional Ziegler–Nichols (ZN) and Cohen–Coon (CC) tuning methods; DDPG is designed to optimize the PID gains () within predefined bounds in a continuous action space. In simulations conducted on four different trajectories—circle, lemniscate, diamond, and star—RMSE, IAE, ISE, ITAE, and maximum error metrics are used for evaluation. According to the results, DDPG-PID achieves the lowest error on all trajectories, reducing RMSE by approximately 35–58% compared to ZN-PID and by approximately 79–82% compared to CC-PID; similarly, improvements are observed in IAE/ISE/ITAE and maximum error values. These findings indicate that DDPG-PID provides more stable and accurate tracking, particularly on complex trajectories involving sharp direction changes, and demonstrate that the proposed method offers a superior automatic PID tuning alternative to classical tuning rules for industrial parallel robot control applications.- 31 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (3) , 1453
Open AccessReinforcement Learning-Based PID Gain Optimization for Delta Parallel Robot Trajectory Tracking
In this study, a PID gain tuning approach using Deep Deterministic Policy Gradient (DDPG), a reinforcement learning (RL) algorithm, is proposed for trajectory tracking of delta parallel robots. Owing to their 3-degree-of-freedom (3-DOF) parallel kinematic structure, delta robots offer higher stiffness, precision, and speed capabilities than serial manipulators; they are therefore widely used in high-speed pick-and-place applications due to their low moving mass and the stiffness provided by the closed-chain mechanism. In this study, the proposed DDPG-PID approach is comparatively investigated against the conventional Ziegler–Nichols (ZN) and Cohen–Coon (CC) tuning methods; DDPG is designed to optimize the PID gains () within predefined bounds in a continuous action space. In simulations conducted on four different trajectories—circle, lemniscate, diamond, and star—RMSE, IAE, ISE, ITAE, and maximum error metrics are used for evaluation. According to the results, DDPG-PID achieves the lowest error on all trajectories, reducing RMSE by approximately 35–58% compared to ZN-PID and by approximately 79–82% compared to CC-PID; similarly, improvements are observed in IAE/ISE/ITAE and maximum error values. These findings indicate that DDPG-PID provides more stable and accurate tracking, particularly on complex trajectories involving sharp direction changes, and demonstrate that the proposed method offers a superior automatic PID tuning alternative to classical tuning rules for industrial parallel robot control applications. - Open Access
- 29 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (3) , 1369
Machine Learning-Based Energy Consumption and Carbon Footprint Forecasting in Urban Rail Transit Systems
In the fight against global climate change, the transportation sector is of critical importance because it is one of the major causes of total greenhouse gas emissions worldwide. Although urban rail transit systems offer a lower carbon footprint compared to road transportation, accurately forecasting the energy consumption of these systems is vital for sustainable urban planning, energy supply management, and the development of carbon balancing strategies. In this study, forecasting models are designed using five different machine learning (ML) algorithms, and their performances in predicting the energy consumption and carbon footprint of urban rail transit systems are comprehensively compared. For five distribution-center substations, 10 years of monthly energy consumption data and the total carbon footprint data of these substations are used. Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Nonlinear Autoregressive Neural Network (NAR-NN) models are developed to forecast these data. Model hyperparameters are optimized using a 20-iteration Random Search algorithm, and the stochastic models are run 10 times with the optimized parameters. Results reveal that the SVR model consistently exhibits the highest forecasting performance across all datasets. For carbon footprint forecasting, the SVR model yields the best results, with an of 0.942 and a MAPE of 3.51%. The ensemble method XGBoost also demonstrates the second-best performance (). Accordingly, while deterministic traditional ML models exhibit superior performance, the neural network-based stochastic models, such as LSTM, ANFIS, and NAR-NN, show insufficient generalization capability under limited data conditions. These findings indicate that, in small- and medium-scale time-series forecasting problems, traditional machine learning methods are more effective than neural network-based methods that require large datasets.- 29 January 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (3) , 1369
Open AccessMachine Learning-Based Energy Consumption and Carbon Footprint Forecasting in Urban Rail Transit Systems
In the fight against global climate change, the transportation sector is of critical importance because it is one of the major causes of total greenhouse gas emissions worldwide. Although urban rail transit systems offer a lower carbon footprint compared to road transportation, accurately forecasting the energy consumption of these systems is vital for sustainable urban planning, energy supply management, and the development of carbon balancing strategies. In this study, forecasting models are designed using five different machine learning (ML) algorithms, and their performances in predicting the energy consumption and carbon footprint of urban rail transit systems are comprehensively compared. For five distribution-center substations, 10 years of monthly energy consumption data and the total carbon footprint data of these substations are used. Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Nonlinear Autoregressive Neural Network (NAR-NN) models are developed to forecast these data. Model hyperparameters are optimized using a 20-iteration Random Search algorithm, and the stochastic models are run 10 times with the optimized parameters. Results reveal that the SVR model consistently exhibits the highest forecasting performance across all datasets. For carbon footprint forecasting, the SVR model yields the best results, with an of 0.942 and a MAPE of 3.51%. The ensemble method XGBoost also demonstrates the second-best performance (). Accordingly, while deterministic traditional ML models exhibit superior performance, the neural network-based stochastic models, such as LSTM, ANFIS, and NAR-NN, show insufficient generalization capability under limited data conditions. These findings indicate that, in small- and medium-scale time-series forecasting problems, traditional machine learning methods are more effective than neural network-based methods that require large datasets. - 15 January 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (1) , 433-443
Simultaneous Feature Selection and Hyperparameter Tuning of K-Nearest Neighbors for Glass Classification: A Comparative Study of Artificial Bee Colony, Simulated Annealing, and Hill Climbing
Glass identification is essential in forensic science and industrial uses. However, the effectiveness of classification algorithms heavily relies on selecting the correct hyperparameters and relevant features. This study investigates the efficacy of three optimization algorithms—Artificial Bee Colony (ABC), Simulated Annealing (SA), and Hill Climbing (HC)—for optimizing the k value, which is the neighbor value of the k-Nearest Neighbors (k-NN) classifier, the distance measure expressing the proximity between two samples, and the feature subset of the UCI Glass Identification dataset. By reframing the classification problem as a multi-dimensional optimization task, the algorithms are assessed based on accuracy, precision, recall, and F1 score. The results show that the population-based ABC algorithm, as a meta-heuristic approach, outperforms local search methods by avoiding local optima and effectively reducing the dimensionality of the data. RI (refractive index) and Mg (magnesium oxide content) emerge as particularly discriminative features. The study demonstrates that combining automatic parameter tuning with feature selection can substantially enhance the predictive performance of k-NN in complex classification tasks.
- 15 January 2026
- Published by Black Sea Journal of Engineering and Science in Black Sea Journal of Engineering and Science
- Vol. 9 (1) , 433-443
Simultaneous Feature Selection and Hyperparameter Tuning of K-Nearest Neighbors for Glass Classification: A Comparative Study of Artificial Bee Colony, Simulated Annealing, and Hill Climbing
Glass identification is essential in forensic science and industrial uses. However, the effectiveness of classification algorithms heavily relies on selecting the correct hyperparameters and relevant features. This study investigates the efficacy of three optimization algorithms—Artificial Bee Colony (ABC), Simulated Annealing (SA), and Hill Climbing (HC)—for optimizing the k value, which is the neighbor value of the k-Nearest Neighbors (k-NN) classifier, the distance measure expressing the proximity between two samples, and the feature subset of the UCI Glass Identification dataset. By reframing the classification problem as a multi-dimensional optimization task, the algorithms are assessed based on accuracy, precision, recall, and F1 score. The results show that the population-based ABC algorithm, as a meta-heuristic approach, outperforms local search methods by avoiding local optima and effectively reducing the dimensionality of the data. RI (refractive index) and Mg (magnesium oxide content) emerge as particularly discriminative features. The study demonstrates that combining automatic parameter tuning with feature selection can substantially enhance the predictive performance of k-NN in complex classification tasks.
- Open Access
Tuning the Proportional–Integral–Derivative Control Parameters of Unmanned Aerial Vehicles Using Artificial Neural Networks for Point-to-Point Trajectory Approach
Nowadays, trajectory control is a significant issue for unmanned micro aerial vehicles (MAVs) due to large disturbances such as wind and storms. Trajectory control is typically implemented using a proportional–integral–derivative (PID) controller. In order to achieve high accuracy in trajectory tracking, it is essential to set the PID gain parameters to optimum values. For this reason, separate gain values are set for roll, pitch and yaw movements before autonomous flight in quadrotor systems. Traditionally, this adjustment is performed manually or automatically in autotune mode. Given the constraints of narrow orchard corridors, the use of manual or autotune mode is neither practical nor effective, as the quadrotor system has to fly in narrow apple orchard corridors covered with hail nets. These reasons require the development of an innovative solution specific to quadrotor vehicles designed for constrained areas such as apple orchards. This paper recognizes the need for effective trajectory control in quadrotors and proposes a novel neural network-based approach to tuning the optimal PID control parameters. This new approach not only improves trajectory control efficiency but also addresses the unique challenges posed by environments with constrained locational characteristics. Flight simulations using the proposed neural network models have demonstrated successful trajectory tracking performance and highlighted the superiority of the feed-forward back propagation network (FFBPN), especially in latitude tracking within 7.52745 × 10−5 RMSE trajectory error. Simulation results support the high performance of the proposed approach for the development of automatic flight capabilities in challenging environments.
Open AccessTuning the Proportional–Integral–Derivative Control Parameters of Unmanned Aerial Vehicles Using Artificial Neural Networks for Point-to-Point Trajectory Approach
Nowadays, trajectory control is a significant issue for unmanned micro aerial vehicles (MAVs) due to large disturbances such as wind and storms. Trajectory control is typically implemented using a proportional–integral–derivative (PID) controller. In order to achieve high accuracy in trajectory tracking, it is essential to set the PID gain parameters to optimum values. For this reason, separate gain values are set for roll, pitch and yaw movements before autonomous flight in quadrotor systems. Traditionally, this adjustment is performed manually or automatically in autotune mode. Given the constraints of narrow orchard corridors, the use of manual or autotune mode is neither practical nor effective, as the quadrotor system has to fly in narrow apple orchard corridors covered with hail nets. These reasons require the development of an innovative solution specific to quadrotor vehicles designed for constrained areas such as apple orchards. This paper recognizes the need for effective trajectory control in quadrotors and proposes a novel neural network-based approach to tuning the optimal PID control parameters. This new approach not only improves trajectory control efficiency but also addresses the unique challenges posed by environments with constrained locational characteristics. Flight simulations using the proposed neural network models have demonstrated successful trajectory tracking performance and highlighted the superiority of the feed-forward back propagation network (FFBPN), especially in latitude tracking within 7.52745 × 10−5 RMSE trajectory error. Simulation results support the high performance of the proposed approach for the development of automatic flight capabilities in challenging environments.
- 28 July 2021
- Published by Springer Nature
- p. 32-45
Design of a Mobile Robot to Work in Hospitals and Trajectory Planning Using Proposed Neural Networks Predictors
- 28 July 2021
- Published by Springer Nature
- p. 32-45
Design of a Mobile Robot to Work in Hospitals and Trajectory Planning Using Proposed Neural Networks Predictors
- 1 January 2020
- Published by IGI Global Scientific Publishing
- p. 361-376
A Proposed Trajectory Planning Algorithm for Mobile Robot Navigation Based on A* Algorithm
This chapter proposes a new trajectory planning approach by improving A* algorithm, which is a widely-used, path-planning algorithm. This algorithm is a heuristic method used in maps such as the occupancy grid map. As the resolution increases in these maps, obstacles can be defined more precisely. However, the cell/grid size must be larger than the size of the mobile robot to prevent the robot from crashing into the borders of the working environment or obstacles. The second constraint of the algorithm is that it does not provide continuous headings. In this study, an avoidance area is calculated on the map for the mobile robot to avoid collisions. Then curve-fitting methods, general polynomial and b-spline, are applied to the path calculated by traditional A* algorithm to obtain smooth rotations and continuous headings by staying faithful to the original path calculated. Performance of the proposed trajectory planning method is compared to others for different target points on the grid map by using a software developed in Labview Environment.
- 1 January 2020
- Published by IGI Global Scientific Publishing
- p. 361-376
A Proposed Trajectory Planning Algorithm for Mobile Robot Navigation Based on A* Algorithm
This chapter proposes a new trajectory planning approach by improving A* algorithm, which is a widely-used, path-planning algorithm. This algorithm is a heuristic method used in maps such as the occupancy grid map. As the resolution increases in these maps, obstacles can be defined more precisely. However, the cell/grid size must be larger than the size of the mobile robot to prevent the robot from crashing into the borders of the working environment or obstacles. The second constraint of the algorithm is that it does not provide continuous headings. In this study, an avoidance area is calculated on the map for the mobile robot to avoid collisions. Then curve-fitting methods, general polynomial and b-spline, are applied to the path calculated by traditional A* algorithm to obtain smooth rotations and continuous headings by staying faithful to the original path calculated. Performance of the proposed trajectory planning method is compared to others for different target points on the grid map by using a software developed in Labview Environment.