Carbon capture and storage is a vital strategy for mitigating rising atmospheric carbon dioxide, and metal–organic frameworks (MOFs) have gained attention as promising sorbents. Numerous simulations have examined factors governing CO₂ capture in MOFs—such as diffusion in MOF-74 under varying temperatures and process modeling of MOF-5—but most were limited to specific structures or conditions, hindering a systematic understanding of diffusion across diverse MOFs. Conventional computational methods also face constraints: density functional theory mainly provides static energy evaluations, while molecular dynamics relies on fixed force fields with poor transferability and an inability to describe reactive events. To overcome these limitations, this study employs molecular dynamics simulations driven by neural network potentials to evaluate CO₂ diffusivity in 17 types of MOFs. Results reveal significant variation in transport behavior, with zeolitic-imidazolate framework-3 showing the highest diffusivity and MOF-74 the lowest—an approximately 19-fold difference. These findings highlight the capability of neural-network-based molecular dynamics to deliver consistent and quantitative assessments of CO₂ transport in MOFs, providing a reliable framework for the rational design of next-generation capture materials.

Commercial exoskeletons currently utilize multiple sensors, including inertial measurement units, electromyography sensors, and torque/force sensors, to detect human motion. While these sensors improve motion recognition by leveraging their unique strengths, they can also lead to discomfort due to direct skin contact, added weight, and complex wiring. In this paper, we propose a simplified motion recognition method that relies solely on encoders embedded in the motors. Our approach aims to accurately classify various movements by learning their distinctive features through a deep learning model. Specifically, we employ a convolutional neural network algorithm optimized for motion classification. Experimental results show that our model can effectively differentiate between movements such as standing, lifting, level walking, and inclined walking, achieving a test accuracy of 98.76%. Additionally, by implementing a sliding window maximum algorithm that tracks three consecutive classifications, we achieved a real-time motion recognition accuracy of 97.48% with a response time of 0.25 seconds. This approach provides a cost-effective and simplified solution for lower limb motion recognition, with potential applications in rehabilitation-focused exoskeletons.

3D ground reaction force (GRF) estimation during walking is important for gait and inverse dynamics analyses. Recent studies have estimated 3D GRF based on kinematics measured from optical or inertial motion capture systems without force plate measurement. A neural network (NN) could be used to estimate ground reaction forces. The NN network approach based on segment kinematics requires the selection of optimal inputs, including kinematics type and segments. This study aimed to select optimal input kinematics for implementing an NN for each foot’s GRF estimation. A two-stage NN consisting of a temporal convolution network for gait phase detection and a gated recurrent unit network was developed for GRF estimation. To implement the NN, we conducted level/inclined walking and level running on a force-sensing treadmill, collecting datasets from seven male participants across eight experimental conditions. Results of the input selection process indicated that the center of mass acceleration among six kinematics types and trunk, pelvis, thighs, and shanks among 15 individual segments showed the highest correlations with GRFs. Among four segment combinations, the combination of trunk, thighs, and shanks demonstrated the best performance (root mean squared errors: 0.28, 0.16, and 1.15 N/kg for anterior-posterior, medial-lateral, and vertical components, respectively).

This study proposed a conditional generative adversarial network (cGAN) model for predicting steel plate deformation based on heating line positions in a line heating process. A database was constructed by performing finite element analysis (FEA) to establish relationships between heating line positions and deformation shapes. Deformation shapes were converted into color map images. Heating line positions were used as conditional labels for training and validating the proposed model.
During the training process, generator and discriminator loss values, along with MSE and R² metrics, converged stably, demonstrating that generated images closely resembled the actual data. Validation results showed that predicted deformation magnitudes had an average relative error of approximately 3% and a maximum error of less than 7%. These findings confirm that the proposed model can effectively predict steel plate deformation shapes based on heating line positions in the line heating process, making it a reliable predictive tool for this application.

Predicting elastic modulus of a porous structure is essential for applications in aerospace, biomedical, and structural engineering. Traditional methods often struggle to capture complex relationships between material properties, design variables, and mechanical behavior. This study employed artificial neural networks (ANNs) to predict the elastic modulus of a porous structure based on various material and design parameters. An ANN model was trained on a dataset generated via finite element analysis (FEA) simulations, covering diverse combinations of material properties and design variables (e.g., porosity, structure types). The model demonstrated high accuracy in predicting the elastic modulus on a separate test dataset. Key findings included identification of significant design variables influencing the elastic modulus and the ANN model"s ability to generalize predictions to new data. This approach showcases that ANN is a powerful tool for designing and optimizing porous structures, providing reliable mechanical property predictions without extensive experimental testing or complex simulations. The proposed method can enhance design efficiency and pave the way for developing advanced materials with tailored mechanical properties. Future research will extend the model to predict other mechanical properties and incorporate experimental validation to verify ANN predictions.

CNN is one of the deep learning technologies useful for image-based pattern recognition and classification. For machining processes, this technique can be used to predict machining parameters and surface roughness. In electrical discharge machining (EDM), the machined surface is covered with many craters, the shape of which depends on the workpiece material and pulse parameters. In this study, CNN was applied to predict EDM parameters including capacitor, workpiece material, and surface roughness. After machining three metals (brass, stainless steel, and cemented carbide) with different discharge energies, images of machined surfaces were collected using a scanning electron microscope (SEM) and a digital microscope. Surface roughness of each surface was then measured. The CNN model was used to predict machining parameters and surface roughness.

In the semiconductor manufacturing industry, efficient operation of wafer transfer robots has a direct impact on productivity and product quality. Ball screw misalignment anomalies are a critical factor affecting precision transport of robots. Early diagnosis of these anomalies is essential to maintaining system efficiency. This study proposed a method to effectively diagnose ball screw misalignment anomalies using 1D-CNN and 2D-CNN models. This method mainly uses binary classification to distinguish between normal and abnormal states. Additionally, explainable artificial intelligence (XAI) technology was applied to interpret diagnostic decisions of the two deep learning models, allowing users to convince prediction results of the AI model. This study was based on data collected through acceleration sensors and torque sensors. It compared accuracies of 1D-CNN and 2D-CNN models. It presents a method to explain the model"s predictions through XAI. Experimental results showed that the proposed method could diagnose ball screw misalignment anomalies with high accuracy. This is expected to contribute to the establishment of reliable abnormality diagnosis and preventive maintenance strategies in industrial sites.

Elderly monitoring systems are gaining significant attention in our increasingly aging society. Existing monitoring systems, which utilize RGB and infrared cameras, often encounter errors when recognizing human-like objects, photos, and videos as actual humans. Additionally, privacy concerns arise due to this issue. However, these challenges can potentially be overcome by employing thermal images. Thus, our study aimed to investigate the feasibility of identifying and categorizing human postures depicted in thermal images using deep learning models and algorithms. To conduct our experiment, we developed a system that utilizes a thermal pose algorithm and a convolutional neural network. As a result, we achieved an average accuracy of 88.3%, with the highest accuracy reaching 91.2%.

In order to monitor the machining status of a machine tool, it is necessary to measure the signal of the machine tool and establish the relationship between the machining status and the signal. One effective approach is to utilize an AIbased analysis model. To improve the accuracy and reliability of AI models, it is crucial to identify the features of the model through signal analysis. However, when dealing with time series data, it has been challenging to identify these features. Therefore, instead of directly applying time series data, a method was used to extract the best features by processing the data using techniques such as RMS and FFT. Recently, there have been numerous reported cases of designing AI models with high accuracy and reliability by directly applying time series data to find the best features, particularly in the case of AI models combining CNN and LSTM. In this paper, time series data obtained through a gap sensor are directly applied to an AI model that combines CNN, LSTM, and MLP (Multi-Layer Perceptron) to determine tool wear. The machine tool and tool status were monitored and evaluated through an AI model trained using time series data from the machining process.

Brain-computer interface (BCI) is a technology used in various fields to analyze electroencephalography (EEG) signals to recognize an individual"s intention or state and control a computer or machine. However, most of the research on BCI is on motor imagery, and research on active movement is concentrated on upper limb movement. In the case of lower limb movement, most of the research is on the static state or single movements. Therefore, in this research, we developed a deep-learning model for classifying walking behavior(1: walking, 2: upstairs, 3: downstairs) based on EEG signals in a dynamic environment to verify the possibility of classifying EEG signals in a dynamic state. We developed a model that combined a convolutional neural network (CNN) and a bidirectional long short-term memory (BiLSTM). The model obtained an average recognition performance of 82.01%, with an average accuracy of 93.77% for walking, 76.52% for upstairs, and 75.75% for downstairs. It is anticipated that various robotic devices aimed at assisting people with disabilities and the elderly could be designed in the future with multiple features, such as human-robot interaction, object manipulation, and path-planning utilizing BCI for control.

The key components of smart manufacturing, a central concept in the era of the 4th Industrial Revolution, consist of digital twin technology, AI, and computer vision technology. In this study, these technologies were utilized to govern the Poppy robot, a humanoid robot designed for educational and research purposes. The digital twin creates a virtual environment capable of real-time simulation, analysis, and control of the robot’s motions. The digital twin of the robot was constructed using Unity, a 3D development program. Motion data was captured while simulating the physical structure and movements of the virtual robot. This data was then fed into a Tensorflow-based deep neural network to generate a regression modelthat predicts motor rotation based on the position of the robot’s hand. By integrating this model with a Python-based robot control program, the robot’s movements could be effectively managed. Additionally, the robot was controlled using Openpose, a computer vision algorithm that predicts characteristic points on a human body. Position data for human joint points was collected from 2D images, and the motor angle was calculated based on this data. By implementing this approach on an actual robot, it became possible to enable the robot to replicate human movements.

In this study, we proposed an AI-algorithm for face mask recognition based on the MobileNetV2 network to implement automatic door control in intensive care units. The proposed network was constructed using four bottleneck blocks, incorporating depth-wise separable convolution with channel expansion/projection to minimize computational costs. The performance of the proposed network was compared with other networks trained with an identical dataset. Our network demonstrated higher accuracy than other networks. It also had less trainable total parameters. Additionally, we employed the CVzone-based machine learning model to automatically detect face location. The neural network for mask recognition and the face detection model were integrated into a system for real-time door control using Arduino. Consequently, the proposed algorithm could automatically verify the wearing of masks upon entry to intensive care units, thereby preventing respiratory disease infections among patients and medical staff. The low computational cost and high accuracy of the proposed algorithm also provide excellent performance for real-time mask recognition in actual environments.

Estimating energy expenditure is essential in monitoring the intensity of physical activity and health status. Energy expenditure can be estimated based on wearable sensors such as inertial measurement unit (IMU). While a variety of methods have been developed to estimate energy expenditure during day-to-day activities, their performances have not been thoroughly evaluated under walking conditions according to various speeds and inclines. This study investigated IMU-based neural network models for energy expenditure estimation under various walking conditions and comparatively analyzed their performances in terms of sensor attachment locations and training/testing datasets. In this study, two neural network models were selected based on a previous study (Slade et al., 2019): (M1) a multilayer perceptron using sensor signals during each gait cycle, and (M2) a recurrent neural network using sensor signal sequences of a fixed window size. The results revealed the following: (i) the performance of the foot attachment model was the best among the five sensor attachment locations (0.89 W/kg for M1 and 1.14 W/kg for M2); and (ii) although the performance of M1 was superior to that of M2, M1 requires accurate gait detection for data segmentation by each stride, which hinders the usefulness of M2.

Bone plates are a medical device used for fixing broken bones, which should not have a crack and hole defect. Defect detection is very important because bone plate defect is very dangerous. In this study, we proposed a defect detection model based on a parallel type convolution neural network for detecting bone plate crack and pore deformation. All size filters were different according to the defect shape. A convolution neural network detected pore defects. Another convolution neural network detected the crack. Two convolution neural networks simultaneously detected different defect types. The performance of the defect detection model was measured and used for the F1- score. We confirmed that performance of the defect detection model was 98.4%. We confirmed that the defect detection time was 0.21 seconds.

Recently, the estimation of joint kinetics such as joint force and moment using wearable inertial sensors has received great attention in biomechanics. Generally, the joint force and moment are calculated though inverse dynamics using segment kinematic data, ground reaction force, and moment. However, this approach has problems such as estimation error of kinematic data and soft tissue artifacts, which can lead to inaccuracy of joint forces and moments in inverse dynamics. This study aimed to apply a recurrent neural network (RNN) instead of inverse dynamics to joint force and moment estimation. The proposed RNN could receive signals from inertial sensors and force plate as input vector and output lower extremity joints forces and moments. As the proposed method does not depend on inverse dynamics, it is independent of the inaccuracy problem of the conventional method. Experimental results showed that the estimation performance of hip joint moment of the proposed RNN was improved by 66.4% compared to that of the inverse dynamics-based method.