Search Results (639)

The increasing demand for poultry meat, driven by its favorable nutritional profile, including low cholesterol and high protein content, has resulted in intensified production volumes and, consequently, elevated ammonia (NH₃) emissions. Artificial intelligence-based predictive approaches offer an effective alternative to conventional treatment-oriented methods by enabling faster and more accurate estimation of NH₃ removal performance. This study aimed to predict the ammonia removal efficiency of broiler litter generated during a production cycle under controlled laboratory-scale conditions using artificial neural networks (ANNs) trained with different learning algorithms. Four ANN models were developed based on the Levenberg–Marquardt (LM), Fletcher–Reeves (FR), Scaled Conjugate Gradient (SCG), and Bayesian Regularization (BR) algorithms. The results showed that the LM-based model with 12 hidden neurons achieved the highest predictive performance (R² = 0.9777; MSE = 0.0033; RMSE = 0.0574; MAPE = 0.0833), while the BR-based model with 10 neurons showed comparable accuracy. In comparison with the FR and SCG models, the LM algorithm demonstrated superior predictive accuracy and generalization capability. Overall, the findings suggest that ANN-based modeling is a reliable, data-informed approach for estimating NH₃ removal efficiency, providing a potential decision-support framework for ammonia mitigation strategies in poultry production systems. Full article

Fourier transform infrared (FTIR) spectroscopy-based gas remote sensing has been widely applied for long-range atmospheric composition analysis. However, when deployed for longwave infrared methane detection, spectral features of methane are significantly interfered by water vapor variations at the edge of atmospheric window, which compromises detection performance. To address the spectral fitting degradation caused by relative changes between methane and water vapor signals, this study incorporates temperature, relative humidity, and sensing distance into the cost function, establishing a continuous optimization space with concentration path lengths (CLs) as variables, which are the product of the concentration and path length. A hybrid differential evolution and Levenberg–Marquardt (D-LM) algorithm is developed to enhance parameter estimation accuracy. Combined with a three-layer atmospheric model for real-time reference spectrum generation, the algorithm identifies the optimal spectral combination that provides the best match to the measured data. Algorithm performance is validated through two experimental configurations: Firstly, adaptive detection using synthetic spectra covering various humidity–methane concentration combinations is conducted; simulation results demonstrate that the proposed method significantly reduces the mean squared error (MSE) of fitting residuals by 95.8% compared to the traditional LASSO method, effectively enhancing methane spectral feature extraction under high-water-vapor conditions. Then, a continuous monitoring of controlled methane releases over a 500 m open path under high-outdoor-humidity conditions is carried out to validate outdoor performance of the proposed algorithm; field measurement analysis further confirms the method’s robustness, achieving a reduction in fitting residuals of approximately 57% and improving spectral structure fitting. The proposed approach provides a reliable technical pathway for adaptive gas cloud detection under complex atmospheric conditions. Full article

To address the localization and mapping challenges for quadruped robots autonomously scouting under corn canopies, this paper proposes CK-SLAM, a SLAM algorithm integrating robot motion constraints and crop row features. The algorithm is implemented on the Jueying Mini quadruped robot, fusing data from 3D LiDAR, IMU, and joint sensors. First, an Invariant Extended Kalman Filter (InEKF) fuses multi-source motion information, dynamically adjusting observation noise via a foot contact probability model (derived from joint torque data) to achieve initial motion state estimation and reliable pose references for point cloud deskewing. Second, three feature extraction schemes are designed, inheriting line/plane features from LeGO-LOAM and adding an innovative crop plane feature extraction module, which uses grid filtering, differential evolution for crop row detection, and RANSAC plane fitting to capture corn plant structural features. Finally, a two-step Levenberg–Marquardt iteration realizes feature matching and pose optimization, with factor graph optimization fusing motion constraints and laser odometry for global trajectory and map refinement. CK-SLAM effectively adapts to gait-induced measurement noise and enhances feature matching stability under canopies. Experimental validation across four corn growth stages shows it achieves an average Absolute Pose Error (APE) RMSE of 2.0939 m (15.7%/56.4%/72.2% lower than A-LOAM/LeGO-LOAM/Point-LIO) and an average Relative Pose Error (RPE) RMSE of 0.0946 m, providing high-precision navigation support for automated field monitoring. Full article

Magnetic localization technology plays a significant role in medical device navigation and human–computer interaction. However, existing localization methods based on local optimization suffer from poor initial solutions and slow convergence. To address the aforementioned challenges, this paper presents a hybrid localization approach, referred to as the Improved Artificial Lemming Algorithm (IALA) Integrated with Levenberg–Marquardt (LM) Optimization. Building upon the Artificial Lemming Algorithm (ALA), the proposed method incorporates an adaptive Gaussian–Lévy hybrid mutation strategy designed to enhance search performance through improved exploration–exploitation dynamics, as quantitatively demonstrated by the diversity-based analysis where IALA maintains higher exploration percentages on multimodal functions while achieving superior optimization results on high-dimensional problems. By introducing a competitive foraging mechanism inspired by the aggressive behavior of the Tasmanian Devil Optimization (TDO) algorithm, it enhances population diversity and search initiative. Furthermore, a time-varying tracking and escape strategy is adopted to improve dynamic optimization performance in complex solution spaces. The proposed method leverages IALA to generate high-quality initial solutions, significantly accelerating the convergence speed and stability of the LM algorithm, thereby improving the overall performance of the permanent magnet localization system. The experimental results show that, using a horizontal test platform of 60 mm × 60 mm with 41 uniformly distributed test points, and acquiring data at vertical heights ranging from 15 mm to 65 mm in 5 mm increments for two distinct orientations of the permanent magnet, the IALA-LM algorithm achieves an average localization success rate of 96.9% over 902 trials, with a mean position error of 1.1 mm and a mean orientation error of 0.17°. Compared with the standard LM algorithm, the proposed IALA-LM algorithm reduces the position error by approximately 66.7% (from 3.3 mm to 1.1 mm) and the orientation error by approximately 94.3% (from 3.0° to 0.17°). Consequently, the proposed method enables high-precision, high-stability, and high-efficiency localization of permanent magnets. It can provide reliable spatial pose estimation support for demanding applications such as miniature implantable or ingestible medical devices (e.g., capsule endoscopy, intramedullary nail fixation, and tumor localization), human–computer interaction, and industrial inspection. Full article

This paper evaluates neural-network approaches for lithium-ion battery state-of-charge (SoC) estimation under a unified pipeline, fixed data partitions, and identical preprocessing. We study FNNs trained with Levenberg–Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) across three hidden sizes (10, 20, 30) and three topologies: Fitting, Nonlinear Input–Output (Nonlinear I/O), and time-series NAR/NARX. Models are assessed using test MSE and RMSE, correlation (R), generalization gap, convergence indicators (final gradient, damping factor), wall time per epoch, and a relative compute-cost index. On the Fitting task, BR-Fitting-FNN with 20 neurons provides the best accuracy-efficiency balance, while LM-Fitting-FNN with 30 neurons reaches slightly lower error at a higher cost. For Nonlinear I/O, BR-Nonlinear I/O-FNN with 30 neurons achieves the lowest test MSE with clear evidence of effective weight shrinkage; LM-Nonlinear I/O-FNN with 20 neurons is a close alternative. In time-series settings, LM-NAR-FNN with 10 neurons attains the lowest test error and fastest epochs but shows a very negative gap that suggests test-split favorability; BR-NAR-FNN with 30 neurons is more costly yet consistently strong. For NARX, LM-NARX-FNN with 20 neurons yields the best test accuracy and robust convergence. Overall, BR delivers the most reliable accuracy–robustness trade-off as networks widen, LM often achieves the best raw accuracy with careful split validation, and SCG offers the lowest training cost when resources are limited. These results provide practical guidance for selecting SoC estimators to match accuracy targets, computing budgets, and deployment constraints in battery management systems. Full article

Guided by principles of symmetry to achieve a proper balance among model consistency, accuracy, and complexity, this paper proposes a new approach for identifying the unknown parameters of nonlinear one-degree-of-freedom mechanical systems using nonlinear regression methods. To this end, the steps followed in this study can be summarized as follows. Firstly, given a proper set of input time histories and a virtual model with all parameters known, the dynamic response of the mechanical system of interest, used as output data, is evaluated using a numerical integration scheme, such as the classical explicit fixed-step fourth-order Runge–Kutta method. Secondly, the numerical values of the unknown parameters are estimated using the Levenberg–Marquardt nonlinear regression algorithm based on these inputs and outputs. To demonstrate the effectiveness of the proposed approach through numerical experiments, two benchmark problems are considered, namely a mass-spring-damper system and a simple pendulum-damper system. In both mechanical systems, viscous damping is included at the kinematic joints, whereas dry friction between the bodies and the ground is accounted for and modeled using the Coulomb friction force model. While the source of nonlinearity is the frictional interaction alone in the first benchmark problem, the finite rotation of the pendulum introduces geometric nonlinearity, in addition to the frictional interaction, in the second benchmark problem. To ensure symmetry in explaining model behavior and the interpretability of numerical results, the analysis presented in this paper utilizes five different input functions to validate the proposed method, representing the initial phase of ongoing research aimed at applying this identification procedure to more complex mechanical systems, such as multibody and robotic systems. The numerical results from this research demonstrate that the proposed approach effectively identifies the unknown parameters in both benchmark problems, even in the presence of nonlinear, time-varying external input actions. Full article

Although the use of adaptive pulleys enhances the motion characteristics of cable-driven parallel robots (CDPRs), it significantly increases the complexity of the kinematics model. Conventional methods often fail to account for the influence of adaptive pulley motion on cable length variation, making it difficult to establish a precise kinematics model. To deal with the problem, this study presents a kinematic modeling and solution method for CDPRs, which incorporates adaptive pulley kinematics. First, the structural design of the CDPR driven by eight cables is analyzed. Then, the generalized kinematics model and the improved kinematics model with adaptive pulley considerations are established. Furthermore, a hybrid Levenberg–Marquardt and Genetic algorithm is proposed to achieve the efficient and high-precision solution of kinematics equations by combining the rapid global search and precise local optimization. Finally, the proposed method is validated through straight path simulation and elliptical path simulation. The simulation results indicate that the tracking accuracy of the end-effector is better than the 1 × 10⁻⁷ level for the proposed method. Full article

Accurate potential energy surfaces (PESs) are the prerequisite for precise studies of molecular dynamics and spectroscopy. The permutationally invariant polynomial neural network (PIP-NN) method has proven highly successful in constructing full-dimensional PESs for gas-phase molecular systems. Building upon over a decade of development, we present CQPES v1.0 (ChongQing Potential Energy Surface), an open-source software package designed to automate and accelerate PES construction. CQPES integrates data preparation, PIP basis generation, and model training into a modernized Python-based workflow, while retaining high-efficiency Fortran kernels for processing dynamics interfaces. Key features include GPU-accelerated training via TensorFlow, the robust Levenberg–Marquardt optimizer for high-precision fitting, real time monitoring via Jupyter and Tensorboard, and an active learning module that is built on top of these. We demonstrate the capabilities of CQPES through four representative case studies: CH₄ to benchmark high-symmetry handling, CH₃CN for a typical unimolecular isomerization reaction, OH + CH₃OH to test GPU training acceleration on a large system, and Ar + H₂O to validate the active learning module. Furthermore, CQPES provides direct interfaces with established dynamics software such as Gaussian 16, Polyrate 2017-C, VENUS96C, RPMDRate v2.0, and Caracal v1.1, enabling immediate application in chemical kinetics and dynamics simulations. Full article

In indoor environments, fusion localization methods that combine Wi-Fi fingerprinting and Pedestrian Dead Reckoning (PDR) are constrained by the high sensitivity of traditional filters, such as the Extended Kalman Filter (EKF), to initial states and by their susceptibility to nonlinear drift. This study presents a Wi-Fi/PDR fusion localization approach based on global geometric alignment optimized via a Genetic Algorithm (GA). The proposed method models the PDR trajectory as an integrated geometric entity and performs a global search for the optimal two-dimensional similarity transformation that aligns it with discrete Wi-Fi observations, thereby eliminating dependence on precise initial conditions and mitigating multipath noise. Experiments conducted in a real office environment (14 × 9 m, eight dual-band APs) with a double-L trajectory demonstrate that the proposed GA fusion achieves the lowest mean error of 0.878 m (compared to 2.890 m, 1.277 m, and 1.193 m for Wi-Fi, PDR, and EKF fusion, respectively) and an RMSE of 0.978 m. It also attains the best trajectory fidelity (DTW = 0.390 m, improving by 71.0%, 14.7%, and 27.8%) and the smallest maximum deviation (Hausdorff = 1.904 m, 52.4% lower than Wi-Fi). The cumulative error distribution shows that 90% of GA fusion errors are within 1.5 m, outperforming EKF and PDR. Additional experiments that compare the proposed GA optimizer with Levenberg–Marquardt (LM), particle swarm optimization (PSO), and Procrustes alignment, as well as tests with 30% artificial Wi-Fi outliers, further confirm the robustness of the Huber-based cost and the effectiveness of the global optimization framework. These results indicate that the proposed GA-based fusion method achieves high robustness and accuracy in the tested office-scale scenario and demonstrate its potential as a practical multi-sensor fusion approach for indoor localization. Full article

In this manuscript, we investigate a fractional-order conformable three-dimensional chaotic financial model with interest rate, investment demand, and price index compartments. On the application of fixed-point theorems and nonlinear analysis, we establish theoretical results regarding the existence and uniqueness of a solution and also study Ulam–Hyers criteria for the stability of the solution of the considered system. Further, we use the fractional-order Runge–Kutta (RK-4) method to approximate the solution of our problem. Also, deep neural network (DNN) techniques are applied to investigate the model from artificial intelligence (AI) perspectives. Numerical simulation shows that it reproduces accurately the qualitative dynamics and confirms the theoretical stability results of the mentioned system. Subsequently, for the DNN analysis, we follow the Levenberg–Marquardt algorithm using Matlab 2023. Different quantities like the root-mean-square error (RMSE), mean squared error (MSE), and regression coefficient and a comparison with numerical data are presented graphically. Also, absolute errors between numerical values and those predicted by DNNs corresponding to different fractional orders are presented. Full article

In imitation learning between humans and robots, the embodiment gap is a key challenge. By focusing on a specific body part and compensating for the rest according to the robot’s size, the embodiment gap can be overcome. In this paper, we analyze dynamic attention to body parts in imitation learning between humans and robots based on a Transformer model. To adapt human imitation movements to a robot, we solved forward and inverse kinematics using the Levenberg–Marquardt method and performed feature extraction using the k-means method to make the data suitable for Transformer input. The imitation learning process is carried out using the Transformer. UMAP is employed to visualize the attention layer within the Transformer. As a result, this system enabled imitation of movements while focusing on multiple body parts between humans and robots with an embodiment gap, revealing the transitions of body parts receiving attention and their relationships in the robot’s acquired imitation movements. Full article

Cellulose nanocrystals (CNCs) have emerged as highly efficient adsorbents for heavy metal removal owing to their biodegradability, wide availability, and rich surface chemistry. Their abundant hydroxyl and other reactive functional groups provide a high density of active sites, significantly enhancing their affinity and adsorption capacity for toxic metal ions such as chromium (VI). The green adsorbent was characterised using FTIR to identify the functional groups. The optimum conditions were pH 6, concentration 140 mg/L, time 120 min, and adsorbent dosage 6 g/L, with a percentage removal of 95%. Deep machine learning was employed to predict the removal capacity of green and biodegradable adsorbents for chromium (VI) removal from wastewater. The findings show that adaptive neuro-fuzzy inference systems effectively model the prediction of Chromium (VI) adsorption. The Levenberg–Marquardt algorithm (LM) was used to train the network through feedforward propagation. In the training dataset, R² was 0.966, Mean Square Error (MSE) 0.042, Absolute average relative error (AARE) 0.053, Root means square error (RMSE) 0.077, and average relative error (ARE) 0.053 for the artificial neural network. The RMSE of 0.021, AARE of 0.015, ARE of 0.01, MSE of 0.017, and R² of 0.998 for the adaptive neuro-fuzzy inference system. These findings confirm the strong adsorption potential of CNCs and the suitability of advanced machine learning models for forecasting heavy metal removal efficiency. Full article

A phase-modulated ellipsometer enables non-contact, high-precision determination of thin-film optical parameters and thickness through polarized light modulation analysis. However, conventional ellipsometers suffer from limited measurement accuracy due to systemic calibration drift and environmental interference. This research presents a novel metrological approach integrating backpropagation neural networks (BP) with a hybrid Particle Swarm Optimization–Levenberg–Marquardt (PSO-LM) algorithm for thin-film thickness quantification. The proposed framework simultaneously determines system parameters and ellipsometry coefficients (ψ, Δ) via multi-objective optimization, achieving calibration-free thickness characterization with sub-nanometer precision. Experimental validation was performed on SiO₂/Si samples with thicknesses ranging from 20 nm to 500 nm. Results demonstrate that the proposed method achieves a root mean square error (RMSE) of <0.006 across the entire thickness range, outperforming the traditional calibration-based method (RMSE ~ 0.008). In addition, the adaptability and stability of the algorithm to complex optical systems are also verified, providing a new method for industrial film thickness monitoring. Full article

This paper presents a physically constrained grey-box identification framework for electromechanical systems, illustrated through the dynamics of brushed DC motors. The method estimates all electromechanical parameters by minimizing a normalized residual that combines current, velocity, and steady-state algebraic constraints under a current-limit condition. Classical approaches such as least-squares and black-box identification often lack physical interpretability and do not explicitly enforce steady-state consistency, making their estimates susceptible to nonphysical parameter drift. The proposed formulation incorporates these physical constraints within a Levenberg–Marquardt scheme with signal normalization, enabling the joint minimization of current and velocity errors. Validation was performed using step-response data from two DC motors under both synthetic and experimental conditions. When applied to unfiltered measurements, the method maintained steady-state relative errors below 1% and achieved low trajectory discrepancies, with NRMSE in velocity between 2.6 and 3.2% and NRMSE in current between 0.9 and 1.2% across both motors. Embedding physical and steady-state constraints directly into the cost function improves robustness and ensures physically consistent parameter estimates, even under high measurement noise and without filtering. The approach provides a general strategy for dynamic system identification under physical consistency requirements and is suitable for rapid calibration, diagnostic monitoring, and controller tuning in robotic and mechatronic applications. Full article

The Leaf Area Index (LAI) is a key vegetation biophysical variable extensively studied using various remote sensing platforms and applications. Most studies focused on retrieving LAI using remote sensing data have primarily applied visible to shortwave infrared (0.3–2.5 µm) data. While we have previously retrieved LAI using thermal infrared (TIR 2.5–14 µm) hyperspectral data under controlled laboratory conditions, this study aims to evaluate the reliability of our earlier findings using in situ and airborne TIR hyperspectral data. In this study, 36 plots, each 30 × 30 m in size, were randomly selected in the Bavarian Forest National Park in southeastern Germany. The EUFAR-TIR flight campaign, conducted on 6 July 2017, aligned with field data collection using an AISA Owl TIR hyperspectral sensor at 3 m spatial resolution. Statistical univariate and multivariate approaches have been applied to predict LAI using emissivity data. The LAI was derived using six narrowband indices, computed from all possible combinations of wavebands between 8 µm and 12.3 µm, via partial least squares regression (PLSR) and artificial neural network (ANN) models, applying the Levenberg–Marquardt and Scaled Conjugate Gradient algorithms. The results indicated that compared to LAI estimation under controlled conditions, TIR narrowband indices demonstrated poor performance in estimating in situ LAI (R² = 0.28 and RMSE_CV = 0.02). Instead, it was observed that the PLSR model unexpectedly achieved higher prediction accuracy (R² = 0.86 and RMSE_CV = 0.36) in retrieving LAI compared to the ANN approach using the Levenberg–Marquardt algorithm (R² = 0.56, RMSE_CV = 0.71); however, it was outperformed by the Scaled Conjugate Gradient algorithm (R² = 0.83, RMSE_CV = 0.18). The results revealed that wavebands located at 8.1 µm, 9.1 µm, 9.85–9.95 µm, and 9.99–10.27 µm are equally effective in predicting LAI, regardless of sensor or measurement/environmental conditions. Our findings have important implications for upscaling LAI predictions, as the identified wavebands are effective across varying conditions and align with the capabilities of upcoming thermal satellite missions such as Landsat Next and Copernicus LSTM. Full article