Potential For Real-time Applications Research Articles

드론을 활용한 실시간 원격 흘수 정밀 계측 프레임워크

Measuring a ship's draft is a critical task that enhances the efficiency of bulk cargo transportation and prevents accidents such as vessel damage due to exceeding maximum draft depth. Traditional draft measurement methods require workers to approach the exterior of the ship for measurement, posing risks of accidents caused by marine environmental factors like waves and wind. To address these issues, this paper proposes an automatic draft measurement framework utilizing drones. The proposed framework estimates the drone's position to transform the images into undistorted frontal views of the draft marks and accurately detects the waterline without the need for training by employing a segmentation foundation model. Subsequently, a YOLO model is used to detect the draft marks, and the final draft depth is measured based on the detected waterline and draft marks. Experimental results demonstrate that the proposed framework achieves a rapid execution time of 0.853 seconds, including both drone pose estimation and draft measurement stages. This is significantly shorter than the time required for manual measurement by workers, indicating its potential for real-time application. Moreover, an average draft measurement error of less than 20mm showcases a high level of accuracy suitable for practical field implementation.

Development of an Integrated AI Model Based on CNN-SVM for Electricity Theft Detection

This research presents the development and implementation of an integrated artificial intelligence model for electricity theft detection, combining Convolutional Neural Networks (CNN) and Support Vector Machines (SVM). The primary objective was to create a more accurate, efficient, and scalable method for identifying fraudulent electricity consumption patterns. Our CNN-SVM hybrid model leverages CNNs for automatic feature extraction from complex consumption data and SVMs for effective classification. This synergy allows for superior performance in detecting subtle anomalies indicative of electricity theft. The methodology involved pre-processing a large dataset of electricity consumption records, training the CNN to extract relevant features, and optimising the SVM classifier to distinguish between normal and fraudulent patterns. We evaluated the model's performance using metrics including accuracy, precision, recall, F1-score, and ROC AUC. Results demonstrated that our integrated CNN-SVM model significantly outperformed conventional machine learning techniques and standalone models in electricity theft detection. The model achieved an accuracy of 96.6%, with a precision of 97.2% and a recall of 96.1%. Comparative analysis against other state-of-the-art algorithms revealed consistently superior performance across all evaluation metrics. To enhance practical applicability, we developed and deployed a web application that implements the model, allowing for user-friendly interaction and real-time theft detection. This addition bridges the gap between research and real-world implementation, providing utility companies with an accessible tool for fraud detection. The study also explored the model's potential for real-time application and scalability to large-scale utility operations. Our findings suggest that the computational efficiency of the CNN-SVM model, coupled with the web application, offers utility companies a powerful and accessible tool for rapid response to potential fraud. This research contributes to the field of electricity theft detection by introducing a novel, high-performance AI model with a practical web-based implementation. The proposed approach not only improves detection accuracy but also offers potential for immediate real-world application, paving the way for more effective fraud prevention in the utility sector.

SHM System for Composite Material Based on Lamb Waves and Using Machine Learning on Hardware.

There is extensive use of nondestructive test (NDT) inspections on aircraft, and many techniques nowadays exist to inspect failures and cracks in their structures. Moreover, NDT inspections are part of a more general structural health monitoring (SHM) system, where cutting-edge technologies are needed as powerful resources to achieve high performance. The high-performance aspects of SHM systems are response time, power consumption, and usability, which are difficult to achieve because of the system's complexity. Then, it is even more challenging to develop a real-time low-power SHM system. Today, the ideal process is for structural health information extraction to be completed on the flight; however, the defects and damage are quantitatively made offline and on the ground, and sometimes, the respective procedure test is applied later on the ground, after the flight. For this reason, the present paper introduces an FPGA-based intelligent SHM system that processes Lamb wave signals using piezoelectric sensors to detect, classify, and locate damage in composite structures. The system employs machine learning (ML), specifically support vector machines (SVM), to classify damage while addressing outlier challenges with the Mahalanobis distance during the classification phase. To process the complex Lamb wave signals, the system incorporates well-known signal processing (DSP) techniques, including power spectrum density (PSD), wavelet transform, and Principal Component Analysis (PCA), for noise reduction, feature extraction, and data compression. These techniques enable the system to handle material anisotropy and mitigate the effects of edge reflections and mode conversions. Damage is quantitatively evaluated with classification accuracies of 96.25% for internal defects and 97.5% for external defects, with localization achieved by associating receiver positions with damage occurrence. This robust system is validated through experiments and demonstrates its potential for real-time applications in aerospace composite structures, addressing challenges related to material complexity, outliers, and scalable hardware implementation for larger sensor networks.

Dual Control Strategy for Non-Minimum Phase Behavior Mitigation in DC-DC Boost Converters Using Finite Control Set Model Predictive Control and Proportional–Integral Controllers

Model Predictive Control (MPC) has emerged as a promising alternative for controlling power converters, offering benefits such as flexibility, simplicity, and rapid control response, particularly when short-horizon algorithms are employed. This paper introduces a system using a short-horizon Finite Control Set MPC (FCS-MPC) strategy to specifically address the challenge of non-minimum phase behavior in boost converters. The non-minimum phase issue, which complicates the control process by introducing an initial inverse response, is effectively mitigated by the proposed method. A Proportional–Integral (PI) controller is integrated to dynamically adjust the reference current based on the output voltage error, thereby enhancing overall system stability and performance. Unlike conventional PI-MPC methods, where the PI controller has an influence on the system dynamics, the PI controller in this approach is solely used for tuning the reference current needed for the FCS-MPC controller. The PI controller addresses small deviations in output voltage, primarily due to model prediction inaccuracies, ensuring steady-state accuracy, while the FCS-MPC handles fast dynamic responses to adapt the controller’s behavior based on load conditions. This dual control strategy effectively balances the need for precise voltage regulation and rapid adaptation to varying load conditions. The proposed method’s effectiveness is validated through a multi-stage simulation test, demonstrating significant improvements in response time and stability compared to traditional control methods. Hardware-in-the-loop testing further confirms the system’s robustness and potential for real-time applications in power electronics.

Ensemble learning model for Protein-Protein interaction prediction with multiple Machine learning techniques

Protein-Protein Interactions (PPIs) encompass the physical interactions or chemical combinations between proteins, which cells employ to carry out diverse biological functions. PPIs Prediction is of great significance for understanding and studying molecular mechanisms and disease mechanisms in organisms. Traditional machine learning algorithms for predicting PPIs face challenges related to imbalanced sample data, loss of long sequence features, and variations in datasets across different species. This paper takes advantage of the attention mechanism in processing sequence tasks and proposes the fusion of Bi-directional Long Short- Term Memory, an integrated learning model of convolutional neural network and multi-head attention mechanism, referred to as CNN BiLSTM Multi-head Att. The model first performs one-hot encoding on the protein sequence and obtains a low-dimensional word vector representation, and then calculates the encoding matrix through the embedding matrix; it uses the convolutional neural network (CNN) and the bidirectional long short-term memory network (Bi-LSTM) to extract the amino acids in the protein sequence. The feature matrix is obtained by collecting information on the time, site, and physical and chemical characteristics; the multi-head attention mechanism is used in weighted calculation and merger of the subspace features of each sequence, and classification is performed after global average pooling and the PPI prediction is output. Cross-validation experiments on different proportions of positive and negative protein data sets and proteins of different species show that CNN BiLSTM Multi-head Att can better predict imbalanced data sets and protein interactions of different species. Compared to the state-of-the-art model, our model improves accuracy by 0.025, precision by 0.046, F1-score by 0.052, and recall by 0.014. Exploring the scalability and adaptability of the proposed model for large-scale biological datasets, along with investigating its potential in real-time applications, could lead to significant advancements in the field of protein–protein interaction prediction.

COMPREHENSIVE EVALUATION OF REAL-TIME OBJECT DETECTION ALGORITHM BASED ON EXTENDED CRITERIA

In this study, we implemented YOLO (You Only Look Once) for real-time object detection and evaluated its performance based on key metrics such as processing speed, frame rate, and object detection accuracy. Our approach emphasizes both the precision and efficiency of YOLO, focusing on its ability to detect objects in real-world scenarios while maintaining a low computational cost. To identify and count objects, the YOLO algorithm was applied to analyze three images. It divided each image into a grid, and each cell predicted bounding boxes and confidence scores for potential objects. Following the processing of these predictions using non-max suppression to remove duplicates, each image contained an accurate count of the items that were detected. The model achieved a processing time of 17.68 seconds, with an average of 0.25 seconds per frame, demonstrating the system's capability for rapid object detection in near real-time applications. On average, 1.32 objects were detected per frame, with a maximum of 1.67 objects in a single frame and a minimum of 1 object per frame, indicating consistent detection across the dataset. The standard deviation of objects per frame (0.113) shows a low variability in object detection rates, reflecting the robustness of the model in handling diverse input frames. The achieved frame rate of 4.2 FPS demonstrates the model's potential for real-time applications, particularly in environments where processing speed is critical. The scientific novelty of this work lies in demonstrating YOLO’s adaptability for efficient object detection while maintaining high detection rates and consistent performance across varying scenarios. This study contributes to the field by showcasing YOLO's applicability in real-time systems, where object detection speed and accuracy are paramount. Our findings provide a foundation for further optimization in high-performance, low-latency object detection tasks, as well as its scalability for more complex detection systems. The results underscore YOLO's potential in both academic and industrial settings.

A Human Activity Recognition model based on CNN and Transformer

Abstract This study aims to utilize data from built-in sensors in smartphones for human activity recognition. By analyzing the three-dimensional accelerometer and gyroscope data in user behavior, accurate classification of eight common activity states is achieved, including walking, standing, sitting, squatting, going up stairs, going down stairs, climbing ladders, and descending ladders. To enhance the model’s generalization capability, a method combining Transformer neural networks with one-dimensional Convolutional Neural Networks (CNNs) is employed, along with data sample augmentation. Experimental results demonstrate a significant improvement in recognition accuracy compared to traditional models, indicating the potential for real-time application on smartphones and other devices. This approach provides essential technical support for predictive human-computer interaction on smart devices and holds extensive application prospects.

ADM-SLAM: Accurate and Fast Dynamic Visual SLAM with Adaptive Feature Point Extraction, Deeplabv3pro, and Multi-View Geometry.

Visual Simultaneous Localization and Mapping (V-SLAM) plays a crucial role in the development of intelligent robotics and autonomous navigation systems. However, it still faces significant challenges in handling highly dynamic environments. The prevalent method currently used for dynamic object recognition in the environment is deep learning. However, models such as Yolov5 and Mask R-CNN require significant computational resources, which limits their potential in real-time applications due to hardware and time constraints. To overcome this limitation, this paper proposes ADM-SLAM, a visual SLAM system designed for dynamic environments that builds upon the ORB-SLAM2. This system integrates efficient adaptive feature point homogenization extraction, lightweight deep learning semantic segmentation based on an improved DeepLabv3, and multi-view geometric segmentation. It optimizes keyframe extraction, segments potential dynamic objects using contextual information with the semantic segmentation network, and detects the motion states of dynamic objects using multi-view geometric methods, thereby eliminating dynamic interference points. The results indicate that ADM-SLAM outperforms ORB-SLAM2 in dynamic environments, especially in high-dynamic scenes, where it achieves up to a 97% reduction in Absolute Trajectory Error (ATE). In various highly dynamic test sequences, ADM-SLAM outperforms DS-SLAM and DynaSLAM in terms of real-time performance and accuracy, proving its excellent adaptability.

Online Calibration of Extrinsic Parameters for Solid-State LIDAR Systems.

This work addresses the challenge of calibrating multiple solid-state LIDAR systems. The study focuses on three different solid-state LIDAR sensors that implement different hardware designs, leading to distinct scanning patterns for each system. Consequently, detecting corresponding points between the point clouds generated by these LIDAR systems-as required for calibration-is a complex task. To overcome this challenge, this paper proposes a method that involves several steps. First, the measurement data are preprocessed to enhance its quality. Next, features are extracted from the acquired point clouds using the Fast Point Feature Histogram method, which categorizes important characteristics of the data. Finally, the extrinsic parameters are computed using the Fast Global Registration technique. The best set of parameters for the pipeline and the calibration success are evaluated using the normalized root mean square error. In a static real-world indoor scenario, a minimum root mean square error of 7 cm was achieved. Importantly, the paper demonstrates that the presented approach is suitable for online use, indicating its potential for real-time applications. By effectively calibrating the solid-state LIDAR systems and establishing point correspondences, this research contributes to the advancement of multi-LIDAR fusion and facilitates accurate perception and mapping in various fields such as autonomous driving, robotics, and environmental monitoring.

Advancing vector-borne disease prediction through functional classifier integration: A novel approach for enhanced modeling

This paper evaluates various machine learning models for predicting vector-borne diseases, focusing on performance metrics that reveal insights into their efficacy. The Multilayer Perceptron (MLP) model demonstrated the highest accuracy at 92%, surpassing the Simple Logistic (SL) and Support Vector Machine (SVM) models, which achieved 88% and 90.87% accuracy, respectively. Notably, the MLP model excelled in precision, recall, and F-Measure, indicating superior classification accuracy. Conversely, the SVM model exhibited noteworthy computational efficiency with the lowest processing time at 0.3 seconds, emphasizing its potential for real-time applications in public health interventions. In contrast, the Radial Basis Function Network (RBFN) lagged in accuracy and other metrics. The results underscore the trade-offs between accuracy and computational efficiency, emphasizing the need for a nuanced model selection. Considering the holistic evaluation, the SVM model emerged as a compelling choice, balancing high accuracy and efficient processing, making it promising for real-time public health applications. This study contributes valuable insights into machine learning model performance, emphasizing the importance of selecting models tailored to the specific needs of vector-borne disease prediction. As we confront emerging infectious diseases, the SVM model stands as an indispensable tool, supporting a proactive and data-driven approach to mitigate the global health impact of vector- borne diseases.

Machine Learning-assisted Prediction and Optimization of Exergy Efficiency and Destruction of Cumene Plant under Uncertainty

Machine Learning (ML)'s growing role in process industries during the digitalization era is notable. This study combines Artificial Neural Networks (ANNs) and Aspen Plus to predict exergy efficiency, exergy destruction, and potential improvements in a cumene plant under uncertain process conditions. An optimization framework, using Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), was developed to enhance exergy efficiency amid uncertainty. Initially, a steady-state Aspen model evaluates exergy efficiency, irreversibility, and potential improvements. The proposed model is transitioned to a dynamic mode, introducing artificial uncertainties into key variables. An ANN model predicts exergy efficiency and exergy destruction under uncertainty. The PSO and GA-based optimization methods improve exergy efficiency and reduce exergy destruction. This work demonstrates the potential real-time application of intelligent methods in plant analysis.

Extremely Randomized Trees Approach to Voltage Stability Assessment of Power Distribution System

In modern power grids, ensuring operational stability necessitates real-time assessment of static voltage stability. This paper presents an advanced ensemble machine learning approach to estimate voltage stability in power distribution systems using data from phasor measurement units (PMUs). The methodology integrates multiple machine learning techniques into a unified predictive model to minimize bias and variance. A critical feature of this approach is the inclusion of a feature pre-processing step to eliminate redundant and irrelevant features, enhancing computational efficiency through dimensionality reduction. Additionally, a minority oversampling method for regression is utilized to address data disparity, ensuring robust model performance. Various ensemble models were evaluated using mean square error metrics, with the Bagging model ExtraTrees outperforming other algorithms. The proposed method's efficacy is validated on a 30-bus IEEE system, highlighting its potential for real-time applications. This research offers a robust solution for maintaining grid reliability and preventing power outages by providing accurate, real-time voltage stability assessments.

Goethite (α-FeOOH) nanoparticles wrapped on reduced graphene oxide nanosheet for sensitive electrochemical detection of arsenic(III)

The goethite (α-FeOOH) nanoparticles were wrapped on the reduced graphene oxide (rGO) to synthesize the α-FeOOH/rGO nanocomposites. The nanocomposites (NCs) were initially examined for their optical, structural, and morphological properties. The XRD data obtained the crystallite size of the α-FeOOH, showed that the average crystal size for pristine α-FeOOH and α-FeOOH/rGO nanocomposites were about 85 and 90 nm, respectively. The transmission electron microscope confirmed the nanoparticles (NPs) were evenly distributed throughout the reduced graphene oxide sheets. The nanocomposites improved glassy carbon electrodes (GCE), making them efficient sensors for detecting the arsenic(III) (As+3) in a pH 5 phosphate buffer solution with an Ag/AgCl reference electrode. The detection limit for As+3 was 0.07 μgL−1 and the resulting sensitivity was 0.39 μA−1 μgL−1 in the linear dynamic range of 0.1–10 μgL−1. The α-FeOOH/rGO/GCE was more sensitive than its original and showed a synergistic effect due to the influence of α-FeOOH on the properties of rGO. The α-FeOOH/rGO NCs-modified GCE electrode performed as a promising sensor, by separating the common interfering ions. Moreover, the modified electrode exhibited remarkable stability, repeatability, and potential real-time application towards the detection of arsenic(III). Additionally, the proposed approach has been successfully applied to the detection of As+3 in the real water sample.

Deep learning-based framework for the observation of real-time melt pool and detection of anomaly in wire-arc additive manufacturing

ABSTRACT Object detection has become a popular tool of deep learning in the era of digital manufacturing. In this study, the most powerful and efficient object detection algorithm, i.e., You Only Look Once (YOLO) algorithm, was used to detect anomalies in deposited beads of wire-arc additive manufacturing (WAAM) using melt pool images. This study used the latest version of YOLO algorithm to train and validate the custom image dataset of the melt pool obtained by conducting experiments using a robotic-controlled WAAM. The mean average precision (mAP) for the “Regular bead” class and the “Irregular bead” class reached 99% at an Intersection over Union (IoU) threshold of 0.5, for both training and validation. When the model was tested for new or unseen datasets by conducting four new experimental trials, the mAP value for the “Regular bead” class reached 98.47% and for the “Irregular bead” class reached 96.68% at an average processing time of 0.014 s/frame. The object detection algorithm YOLO has shown an excellent processing time of 15 ms per frame, which shows its potential for real-time application in the manufacturing industry.

Real-Time Water Level Monitoring Based on GNSS Dual-Antenna Attitude Measurement

Real-time and high-precision water level monitoring is crucial for the fields of hydrology, hydraulic engineering, and disaster prevention and control. The most prevalent method for measuring water level is through the use of water level gauges, which can be costly and have limited coverage. In recent years, Global Navigation Satellite System Reflectometry (GNSS-R) technology has emerged as a promising approach for water level monitoring due to its low cost and high coverage. However, a limitation of current GNSS-R technology is the extended time required to record signals, which hinders its potential for real-time application. This paper introduces a novel real-time water level monitoring method based on GNSS dual-antenna attitude measurement and develops a model to invert water level based on baseline vector. This method uses double-difference observations to eliminate errors caused by various factors, such as satellite and receiver clock, and ionospheric and tropospheric delay. To avoid the impact of detecting and correcting cycle slips during real-time operations, a single-epoch calculation method is introduced. In order to verify the stability and reliability of our method, field tests were carried out at Dongshahe Station in Beijing. We obtained water level data with a time resolution of 1 Hz through field experiments. Experimental data collected from 12 May to 8 June 2022 and from 4 July to 8 August 2022 showed good agreement with on-site water gauge measurements, with root mean square errors of 2.77 cm and 2.54 cm, respectively. Experimental results demonstrate that this method can achieve high-precision, high-temporal-resolution water level monitoring.

Predictive Cruise Controller for Electric Vehicle to Save Energy and Extend Battery Lifetime

Electric vehicles are considered the most effective solution to the petroleum crisis and reduction of air pollution. In order to enhance energy efficiency and battery lifetime, this paper designs a predictive cruise controller (EC) for electric vehicles. Road information and traffic preview are employed for velocity planning while minimizing energy usage, maintaining battery health, and avoiding collision with a lead vehicle. To enable real-time implementation, we apply a model predictive control (MPC) framework formulated in space domain, and approximation and relaxation are introduced to obtain a smooth nonlinear program. Simulation results indicate that the proposed controller yields suboptimal performance as compared to the globally optimal solution. For higher practicability on real-life scenarios, we develop an enhanced EC that is capable of optimizing the stopping of the ego vehicle. According to the car-following studies where the lead vehicle is driven using real-life data, the enhanced EC achieves <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$7.14 \%$</tex-math></inline-formula> energy saving and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$29.81 \%$</tex-math></inline-formula> battery life extension when compared to the intelligent driving model. The computation time of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$40 \,\mathrm{ms}$</tex-math></inline-formula> per MPC update also demonstrates its potential for real-time applications.

Application of Generative Adversarial Network-Based Inversion Algorithm in Imaging 2-D Lossy Biaxial Anisotropic Scatterer

An effective quasi-real-time inversion algorithm based on the super-resolution generative adversarial network (SR-GAN) is proposed to quantitatively image the 2-D biaxial anisotropic scatterers. The SR-GAN was originally proposed for the purpose of super-resolution image reconstruction, which exactly fits the need for inverse problem. In addition, Visual Geometry Group (VGG) loss is introduced to extract the high-level features of the object instead of the low-level pixel-wise error measures. The angle-dependent reconstruction effect due to the dipole radiation as in the traditional inversion methods is effectively resolved by the machine learning method. Numerical results using both synthetic data and experimental data are given to validate the effectiveness of the proposed method. Both the imaging quality and resolution are greatly improved by the proposed SR-GAN algorithm, compared to the traditional iterative inversion algorithm. In addition, the computational time is reduced significantly and the quasi-real-time imaging is finally realized, which promises a potential real-time application of the inverse scattering method.

Three staged integrated community-based water filter system for potable water by effective removal of contaminants from ground water

Potable water scarcity is a dire problem faced worldwide due to contaminated ground water. Microbial contaminants, heavy metals chromium (Cr) and metalloid such as arsenic (As) are some of the major pollutants of ground water and pose serious health hazards. The present work is focused to design a multifunctional three compartment integrated water filtration system for community use and its validation at both lab and pilot scale. The top compartment is mainly for removal of flocculating contamination with improvement of taste and odor of drinking water. The second compartment consists of functionalized sand layers for removal of metal and microbial contaminants and the third compartment comprises of nanoparticle embedded ceramic candles to further eliminate any remaining contaminants. The developed integrated water filter containing various types of materials were characterized through flow rate, scanning electron microscopy (SEM) and energy dispersive spectrometry (EDS) before and after filtration. Moreover, the analysis of various physiological parameters such as pH, turbidity, total dissolved solids and hardness pre and post filtration elucidated stable and improved results. Micro-computed tomography (micro-CT) analysis of baked ceramic candle showed substantial increased porosity of ~83.11 ± 2.17 % in comparison to unbaked candle (26.09 ± 1.33 %). The inductively coupled plasma mass spectrometry (ICP-MS) analysis demonstrated significant reduction of As and Cr levels before and after filtration, which was further confirmed by EDS mapping. Microbial testing showed hierarchal decrease in microbial load after each stage at lab scale. This low-cost, multistage, gravity-based filtration system proved effective in removal of heavy metal and microbial contaminants in water and provides the potable filtrate for daily use. Thus, the integrated system with simple design without requiring major infrastructure and promising results may lead to its potential real-time application for supplying potable water to large populations, particularly in rural populations who cannot afford sophisticated water purification system.

Power Management of a Plug-in Hybrid Electric Vehicle Using Neural Networks with Comparison to Other Approaches

Many researchers spent much effort on the online power management strategies for plug-in hybrid vehicles (PHEVs) and hybrid electric vehicles (HEVs). Nowadays, artificial neural networks (ANNs), one of the machine learning techniques, have also been applied to this problem due to their good performance in learning non-linear and complicated multi-inputs multi-outputs (MIMO) dynamic systems. In this paper, an ANN is applied to the online power management for a plug-in hybrid electric vehicle (PHEV) by predicting the torque split between an internal combustion engine (ICE) and an electric motor (e-Motor) to optimize the greenhouse gas (GHG) emissions by using dynamic programming (DP) results as training data. Dynamic programming can achieve a global minimum solution while it is computationally intensive and requires prior knowledge of the entire drive cycle. As such, this method cannot be implemented in real-time. The DP-based ANN controller can get the benefit of using an ANN to fit the DP solution so that it can be implemented in real-time for an arbitrary drive cycle. We studied the hyper-parameters’ effects on the ANN model and different structures of ANN models are compared. The minimum training mean square error (MSE) models in each comparison set are selected for comparison with DP and equivalent consumption minimization strategy (ECMS). The total GHG emissions and state of charge (SOC) are the metrics used for the analysis and comparison. All the selected ANNs provide results that are comparable to the optimal DP solution, which indicates that ANNs are almost as good as the DP solution. It is found that the multiple hidden-layer ANN shows more efficiency in the training process than the single hidden-layer ANN. By comparing the results with ECMS, the ANN shows great potential in real-time application with the smallest deviation from the results of DP. In addition, our approach does not require any additional trip information, and its output (torque split) is more directly implementable on real vehicles.

Wearable Inertial Sensor-Based Limb Lameness Detection and Pose Estimation for Horses

Accurate objective, automated limb lameness detection and pose estimation play an important role for animal well-being and precision livestock farming. We present a wearable sensor-based limb lameness detection and pose estimation for horse walk and trot locomotion. The gait event and lameness detection are first built on a recurrent neural network (RNN) with long short-term memory (LSTM) cells. Its outcomes are used in the limb pose estimation. A learned low-dimensional motion manifold is parameterized by a phase variable with a Gaussian process dynamic model. We compare the RNN-LSTM-based lameness detection method with a feature-based multi-layer classifier (MLC) and a multi-class classifier (MCC) that are built on support vector machine/K-nearest-neighbors and deep convolutional neural network methods, respectively. Experimental results show that using only accelerometer measurements, the RNN-LSTM-based approach achieves 95% lameness detection accuracy and also outperforms the feature-based MLC or MCC in terms of several assessment criteria. The pose estimation scheme can predict the 24 limb joint angles in the sagittal plane with average errors less than 5 and 10 degs under normal and induced lameness conditions, respectively. The presented work demonstrate the successful use of machine learning techniques for high performance lameness detection and pose estimation in equine science. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —Automation technologies are increasingly used for precision agriculture but few have focused on monitoring individual animals in open field for precision livestock farming. Limb lameness detection and pose estimation in open field is labor-intensive, unsafe for farmers, and inefficient. The presented machine learning-enabled, wearable inertial sensor-based design provides an effective and efficient approach for horse limb lameness detection and pose estimation applications. We present an RNN-LSTM for lameness detection and an integrated manifold learning model is used to predict the horse limb joint angles in walk and trot gaits under normal and induced lameness conditions. We also present a systematic analysis and experiments to demonstrate the impacts of the wearable sensor locations and signal information on lameness detection and pose estimation performance. Several other machine learning-based lameness detection methods are also presented and compared. The extensive multi-horse testing results are presented to demonstrate the superior accuracy and higher performance than other types of machine learning methods. One attractive feature of the proposed design lies in its high performance and fast computational capability for potential real-time applications in open field.