This study considers an extension model of the Two-Echelon Location Routing Problem with Simultaneous Pickup and Delivery (2E-LRPSPD), namely the Two-Echelon Location Routing Problem with Simultaneous Pickup and Delivery and Parcel Locker (2E-LRPSPD-PL). 2E-LRPSPD-PL deals with the accumulation of routing plan and facilities' locations decisions, i.e., satellite and parcel lockers. The first echelon of 2E-LRPSPD-PL consists of one depot, opened satellites, and opened parcel lockers. While the second echelon is described by opened satellites which create routes to fulfill demands of home service customers and by opened parcel lockers which accommodate the demands of parcel locker customers. We propose two-indexed Mixed Integer Linear Programming (MILP) formulation to describe the problem. CPLEX solver is employed to solve small to medium test instances of the 2E-LRPSPD-PL. The computational result shows that CPLEX effectively results in an optimum solution for small instances of 2E-LRPSPD-PL. Furthermore, in the execution of small to medium test instances, some scenarios regarding the number of parcel locker customers are considered to analyze their impact toward the model.
An air cargo terminal is an essential gateway to move goods for import shipping or to load goods for export shipping. The air cargo terminal provides high-quality physical cargo handling services for import, export, and transshipment cargo, with extensive space and standard operating procedures. Terminals must increase their import throughput to use available space and infrastructure efficiently. This study presents an approach merging ordinal optimization and golden jackal optimization to solve the optimal allocation of container freight stations in air cargo. Golden jackal optimization is a metaheuristic method replicating the golden jackals’ natural hunting behaviors. The developed approach is adopted to decide the optimal allocation of shipping/receiving dock in an air cargo terminal such that the long-term average waiting time of trucks is minimized. Simulation results show that the presented approach could improve the convergence speed and local and global search capabilities.
In the modern era, research on Brain-Computer Interfaces (BCIs) is thriving, with increasing focus on the application of Motor Imagery (MI). This study has developed a game aimed at collecting MI signals from subjects imagining movements of their left hand, right hand, and being in a rest state, capturing the Electroencephalography (EEG) signals 1 second before each imagined movement. To achieve real-time interpretation of these signals, this research employed the Transformer architecture for processing EEG signals. Thanks to its self-attention mechanism's strong adaptability to sequential data, the Transformer not only effectively eliminated noise but also successfully extracted key features. Subsequently, these Transformer-processed signals were inputted into EEGNet for a three-class classification. In tests conducted with eight participants, our model achieved an average accuracy of 71.32%, marking a 9.17% improvement over methods that only used EEGNet. This outcome not only demonstrates the potential of combining Transformer and EEGNet in BCI applications but also offers a new direction for future research.
This study introduces a coverage control algorithm designed to optimize the monitoring and tracking of targets utilizing camera-equipped drones. The algorithm tackles the challenge posed by the limited sensing range of cameras by harnessing the flexibility of drones to achieve comprehensive target coverage effectively. In addition to enhancing coverage, the algorithm places emphasis on augmenting estimation accuracy to bolster system stability and facilitate future applications. It specifically targets estimation errors that are typically overlooked in conventional coverage control methods, with the aim of refining accuracy. A crucial criterion in this endeavor is the determination of the state covariance lower limit, which can be assessed through observability evaluation. In consideration of the probability of target occurrence, the primary objective is to ensure continuous coverage of areas with high information density within the field of view of drone-equipped cameras. Observability plays a pivotal role in promoting efficient collaboration through the exchange of information and positional data. To minimize errors, factors such as target locations, estimation accuracy, and camera characteristics on each drone are taken into account. In scenarios involving communication between drones, the exchange of observational data enables the estimation of target groups. The algorithm leverages observability as a decision-making indicator along with a Control Barrier Function to ensure that drones steer clear of losing track of targets. The effectiveness of the developed controller is corroborated through simulations.
In this paper we focus on Bluetooth RSSI positioning and develop a new Bluetooth positioning framework to reduce the problem of wasting resources caused by the lack of flexibility in the number of Bluetooth beacons required for large-scale positioning using the current framework. We break the traditional concept of the one being located, so that it can also function as a beacon after obtaining the coordinate position, providing more information to assist calculations, thus expanding more locatable areas and devices. This greatly reduces the cost and difficulty of densely deploying beacons. Utilizing the characteristics of low-power Bluetooth and its scanning and broadcast functions, we made a network structure similar to Bluetooth MESH, allowing information to flow layer by layer between devices, and ultimately allowing the central computing unit to locate the locations of all devices within the framework. Finally, this paper increased the number of devices in the experiment, and found that the increase in the number of devices rises the possibility of being located, and shows a logarithmic growth curve.
According to cancer registration data from the National Health Service in 2009 and death statistics from the Ministry of Health and Welfare in 2011, the incidence of breast cancer among Taiwanese women ranks first, and the number of cancers ranks third. The peak period for breast cancer is between the ages of 45 and 69. However, the age range for breast cancer seems to be increasing, so young women should not be ignored. The main clinical feature of breast cancer is a palpable breast lump. However, breast lumps are initially thought to be benign and rarely malignant. The interpretation of masses has become the basis for interpretation by allied health professionals and patients.
Identifying crack propagation direction and depth is crucial for steel structure safety. This study explores an orthogonal eddy current excitation technique for detecting multi-orientation slits in mild steels. Using 3-D finite element (FE) simulation, the normal magnetic response Bz distribution induced by orthogonal eddy current excitation is calculated. Two perpendicular inductor arrays mimic orthogonal eddy currents. Artificial vertical, horizontal, 45 degrees, and cross slits on a 12-mm mild steel plate are modeled, and the normal magnetic response contributed by these slits is simulated. Findings show that the orthogonal eddy current excitation technique causes a change in the magnetic field distribution for all types of slits, indicating its ability to resolve any orientation of the slits. Both x- and y-differential signals exhibit similar sensitivity in detecting vertical and horizontal slits. In contrast to the real component of the differential signal, the imaginary components showed a higher normalized differential value due to slits, suggesting a higher detection sensitivity of the differential imaginary components. The orthogonal eddy current induction technique would be a preferable choice to be implemented in an ECT probe.
Dynamic charging management of electrified ride-hailing services under a stochastic environment is a challenging research issue due to the interplay between vehicles’ decisions for serving customers versus charging operations. Existing studies assume constant energy prices and uncapacitated charging stations or do not explicitly consider vehicle queueing at charging stations, resulting in over-optimistic charging infrastructure utilization. This work develops a mixed integer linear program to optimize the sequential decision problems of dynamic ride-hailing systems to maximize the operator's profit under time-varying energy prices. We tested the proposed method on different scenarios using 2019 NYC yellow taxi data. The results show that the proposed methodology can (i) increase the profit and service rate (+7.7% and +7.2%, respectively), (ii) reduce total electricity cost for charging by at least 52.6% compared with the best benchmark approach (100 EVs and 4000 customers/day), (iii) significantly reduce vehicle waiting time at charging stations under a heterogeneous charging station environment.
Pain perception in the brain is inherently subjective, with the existing quantitative measures of somatosensory sensitivity in healthy adults largely confined to results from subjective behavioral surveys. In this study, we used an unsupervised K-mean algorithm to cluster the somatosensory sensitivities of healthy adults to different stimuli, and the Gaussian classifier and K-nearest neighbor classifier for 75 participants achieved peak accuracy of 0.982 and 0.924 for K-means with ${k}={4}$ . Using Multidimensional scaling (MDS) to perform confirmation of the cluster distribution relationships for the four types of generally hypersensitive (HS), generally non-sensitive (NS), predominantly thermally sensitive (TS), and predominantly mechanically sensitive (MS). The investigation and quantification of somatosensory stimulus types in healthy adults will bring a deeper understanding of the breadth and applicability of cognitive neuroscience. The effect of data fusion benefits was achieved by using the EEG and the precise spatial localization of MRI to investigate the connectivity coherence of functional brain networks across different somatosensory phenotypes. We found that the number of brain regions activated in the TS type has a maximum of 43 brain regions and NS type has a minimum of 31 brain regions.
Gastric polyp segmentation is essential in helping clinicians detect and remove polyps to prevent their potential progression to cancer. However, manually segmenting polyps is a laborious task and demands costly human resources. To overcome this issue, deep learning approaches such as SegNet have been applied for automated polyp segmentation. Though SegNet has proven to obtain good results, its restricted receptive field may not capture extensive contextual information from input data, thereby reducing its robustness. This study proposes an optimized SegNet model that incorporates dilation rate and dropout within the convolution layers to increase its receptive field and prevent overfitting during the model training. The model is trained, validated, and tested on colonoscopy images in the Kvasir-SEG dataset. The optimized SegNet model is cross-validated to evaluate its performance. The proposed model is evaluated by comparing it with the other models, and the results show that the optimized SegNet achieves 99.81% of accuracy, 99.24% of sensitivity, 99.93% of specificity, 99.23% of F1-score, and 0.99 of mIoU. Conclusively, the optimized SegNet model proves its superiority in segmenting gastric polyps compared to existing state-of-the-art models.
This paper presents a text-independent speaker verification system leveraging lightweight 3D Convolutional Neural Networks (3D-CNN). Our system independently operates of text and focuses on classifying speakers based on their extracted features. We employ lightweight 3D-CNN to capture the nuances within speech samples from the same speaker. Initially, speaker speech data is used for enrollment, generating corresponding speaker features that form the basis of the speaker model, also referred to as the identity discriminator. Subsequently, speech data requiring verification is utilized as evaluation data, and the resulting speaker features are compared with the speaker model using cosine similarity. Experimental findings show that our system achieves 14.3% on Equal Error Rate (EER). Additionally, the performance of the lightweight 3D-CNN system remains consistent compared to the 3D-CNN system. At the same time, the proposed highlighting system can effective to reduce on computational cost.
In this article, we propose a novel collision-avoidance control design for human side-following of a mobile robot. This design allows a mobile robot to choose a suitable side-following position according to the acquired point cloud data, and control the robot to move toward the best-possible following position while avoiding obstacles. The idea is to keep the user and robot not to lose the sight of each other. Based on this concept, the robot chooses the best position in the visibility field for continuous following while avoiding obstacles. The visibility field is built based on the extended ray-casting algorithm. Combining the extended ray-casting with the mapping of the moving grid, the field of view of the user and the robot can be aligned. The proposed intelligent visibility algorithm(ISA) allows the robot to find the best possible following position at each time instant. This method enables the robot to use a single 2D LiDAR to detect the user and obstacles for real-time side following without colliding with any obstacles. The experimental results on a self-made laboratory mobile robot verified the proposed method. It is demonstrated that the robot can follow the user side-by-side in indoor environments with the capacity to continue side-following after avoiding static and dynamic obstacles.
Deep learning technologies have emerged as optimum solutions for many industrial applications, especially failure-classified of rotating machine problems. The deep learning approach can automatically identify failure types and provide recommendations for unsafe conditions of diesel generators. This research proposed the failure diagnosis method based on long short-term memory to provide the maintenance indicators based on different collected variables. The accumulated datasets in the laboratory evaluate the model capability of failure classified with various evaluating metrics. The classified accuracy is compared with the traditional recurrent neural network (RNN), which demonstrates the classified ability to accurately identify various failure types of diesel generators in the fourth industrial revolution.
This paper presents the design of a brain-computer interface (BCI) system that receives and analyses electroencephalography (EEG) signals to determine commands for drone control. To achieve this, we used an 8-channel wearable EEG headset (Fp1, Fp2, Fz, C3, C4, Pz, O1, O2) to establish a personal database. It is primarily used to classify the intention for left or right eyes movement, allowing the subject to control the drone's direction. The model was applied during real-time testing. The data underwent preprocessing using a bandpass filter. The system recorded the subject's relaxed EEG data for 10 seconds to establish a baseline, which was subsequently used to determine the related drone commands. After the drone took off automatically, we recorded and analyzed the brainwave state every 3 seconds. We then transmitted the corresponding action commands to the DJI Tello drone to control its movements. Currently, we have designed seven different action commands based on EEG data for drone movements, including forward, backward, upward, downward, leftward, rightward, and landing. The accuracy of control commands is approximately 70%. The system aims to create a drone system that is operable by anyone. It is expected that this system can be applied to fields such as neurorehabilitation, exoskeleton manipulation, and the defense industry in the future.
Controlling the moisture content in paddy seeds during the drying process is crucial to ensure the quality of the seeds. Although the capacitance technique can provide a straightforward moisture content detection based on the change in the seeds' dielectric properties, capacitance drift in the sensing circuit due to the temperature can affect the detection accuracy. In this work, we investigate the capacitance drift of a custom-made LC resonant-based capacitance sensor caused by the temperature from 30 to 70 degrees C. The capacitance of the sensor's electrodes is simulated using a finite element method, and the sensor is implemented in a printed circuit board. A differential capacitance technique is proposed to mitigate the temperature drift in the capacitance sensor where the capacitance drift is reduced by four times to 4.9 fF/degrees C. The differential capacitance technique showed a promising approach to enable a simple temperature compensation technique in a paddy seeds' moisture content detection system using the LC resonant-based capacitance sensor.
A hybrid control algorithm, which combines channel selection and transmission power control for Sub-1GHz wireless sensor network systems, is proposed in this paper. By measuring and evaluating the status of candidate channels, the hybrid control algorithm can identify a channel with minimal interference for transmission. Furthermore, by analyzing link quality estimators and employing a fuzzy controller, the system can automatically adjust the transmission power of nodes to adapt to their surrounding environment. Unstable frequencies can be avoided through frequency hopping, either. The proposed algorithm helps nodes reduce power consumption and prolong the life of system while ensuring communication stability. The experiment results show that the proposed hybrid control algorithm can maintain the communication stability defined by IEEE802.15.4, achieving a packet error rate of less than 1% at various test locations. Additionally, the performance of the hybrid control algorithm is more significant when the environment changes rapidly. According to the estimation of nodes' power consumption, several nodes employing the proposed algorithm can maintain the battery life of over seven years, demonstrating the efficacy of the hybrid control algorithm presented in this paper.
This paper proposes a three-step neural network integration. In the first step, two neural networks are trained to produce a truth output and a falsity output. The second step uses cascade generalization to improve the results of the first step. A sequence of pairs of neural networks where the output of lower-order pair and the training data are used to train the next higher-order pair. The truth and falsity output from the first step are used separately in training using this technique. In the third step, two neural networks are trained using the results of the first step, the two results from the second step, and the training data to produce the final classification results. The rice dataset from UC Irvine Machine Learning Repository are used to test the proposed technique. The accuracy of using the three-step technique is better than using other ensemble techniques.
Firefighting pumps are a critical component of the firefighting system, which directly affects the safety operation of the buildings. Current fault-diagnosing methods for firefighting pumps have limitations and shortcomings due to their complex structure. To improve the failure diagnosing performances of artificial intelligence-based approaches, a GRU framework is developed to quickly identify failure conditions of firefighting pumps, which reduces labor expenses and enhances the quality of maintenance service. Firefighting pumps are installed with sensor devices to simulate various health conditions during the data acquisition. A deep learning approach has been developed to identify different failure types of firefighting pumps. The comparison with other state-of-the-art techniques, including recurrent neural network (RNN), demonstrates the effectiveness of the proposed method, which achieves the ultimate improvements of 17.72% loss, 12.36% MAE, 6.41% validating MAE, 29.36% MSE, and 23.92% validating MSE. The proposed framework has been successfully developed and deployed in Taiwan's firefighting pump manufacturing company.
This paper aims to propose a fuzzy PID controller based on quantum computing to enable the bipedal robot system to achieve autonomous balanced walking and obstacle avoidance. The hardware configuration of this system uses Arduino Mega 2560 as the main control board, and is equipped with an MPU-6050 gyroscope and a Pixy2.1 vision module to provide all the sensing data required for the bipedal robot to walk and avoid obstacles, and to obtain the robot location and environmental obstacle information in real time. The control strategy uses quantum computing to perform operations on the fuzzy PID controller to generate control commands so that the bipedal robot can achieve stable walking and obstacle avoidance. In addition, the study also compared the performance between quantum computing and traditional computing, thereby demonstrating the excellent performance of quantum computing in the control required for bipedal robots. Finally, this study verified the effectiveness of the proposed fuzzy PID controller for quantum computing through experiments in which the biped robot walked autonomously and balancedly on flat ground and narrow paths to avoid obstacles.
Stampede accidents have always been a problem that cannot be ignored in the world. Recently, numerous serious fatal accidents have occurred due to crowd pushing during events. The most famous one is the serious accident in Itaewon, South Korea in 2022. According to the South Korean government, 159 people were killed and 196 others were injured. The main cause of death in this crowding accident is not only due to being trampled, but also due to the high crowd density, which caused people to suffer from compressive suffocation, means that it was a long time for people stuck in there not to breathe. These stampede incidents occur mainly because the flow of people cannot be effectively managed and diverted. In view of this, this study develops an intelligent people flow guidance system in crowded spaces, which apply deep learning technology to identify pedestrians and predict the direction of people flow. Next, an intelligent retractable with autonomous movement, positioning and navigation functions is developed, so that the intelligent retractable to move to the correct position autonomously and pull up telescopic fences to guide and divert pedestrians, effectively achieving the effect of rapid evacuation in crowded areas.
