The geothermal power plants in Japan have a low plant capacity factor. In the geothermal power plant under study, a problem is that the pressure of the steam production wells frequently drops suddenly. The plant operators have recovered the steam production well to a stable state through the valve operation based on their experience when the well is unstable. However, the same operation, such as opening a valve, may result in a decrease in reservoir pressure which can be accelerated depending on the reservoir recharge conditions. The objective of this study was to examine wellhead valves operation to prevent sudden pressure drop. This study developed a machine learning (ML) model to predict time-series of steam enthalpy from inputs such as valve openings using Temporal Fusion Transformer (TFT) architecture. The ML model was confirmed to function as a simulator to predict the behavior of the steam enthalpy when the valve openings are varied. The ML model was also able to detect the rapid enthalpy decrease from 25 hours before the happened. Furthermore, we confirmed that the ML model can sequentially evaluate whether changing the valve openings stabilized the reservoir conditions. Future studies should validate and extend these findings from diverse perspectives.
Adaptive output feedback control based on almost strictly positive real (ASPR) properties of the system with an adaptive predictive control as a feedforward input has been proposed for linear continuous-time systems. The designed control system can guarantee stability and keep robust and higher control performance with input limitations even though the systems are uncertain. This paper proposes an adaptive predictive control strategy, which maintains the stability of the control system without imposing the input limitation. In the proposed method, we can obtain a stable control system for uncertain systems by estimating the unknown parameters. The stability of the proposed method is analyzed and its effectiveness is confirmed through numerical simulations.
A lot of research on path-planning technology for self-driving vehicles have focused on safety, time efficiency and the comfort of driving. The next needed step in the researches considering optimal behavior of vehicles is a control that takes into account the optimal behavior of the passenger. This paper considers a model predictive control method how to plan a path that restrains the passenger’s behavior. We evaluate ride comfort based on passenger’s motion perception characteristics.
In recent web conference systems, it has become possible to express the user’s gaze and attention through image processing. However, many users feel uncomfortable turning on the camera, and the sense of looked at and listened to worsens when the camera is turned off. Therefore, in this study, we investigated whether it is possible to improve the sense of gaze and listening by using thought bubbles even when the camera is off and there is no face information of the interlocutor. We have developed a video conferencing system that displays a pictogram image of a person with a thought bubble. The thought bubble displays the user’s face image. In addition, it is possible to extract the words that are the focal points of the user’s utterances and display them in a thought bubble. As a result, it was confirmed that the interlocutor’s sense of being looked at was enhanced by the presence of a thought bubble even when the interlocutor shows pictogram images of people. This result also suggests that thought bubbles can effectively show an interlocutor’s attention.
When using digital twins to create a simulation in the cyber space, their plant models must sequentially predict physical phenomena such as energy and water. However, increase in calculation costs renders their execution in real time challenging in case of a large scale, and they must be executed with high accuracy. Employing the mainstream technique of reducing the computational cost involves data-driven modeling, which suffers limited extrapolation accuracy as it does not have a physical structure. This study derived the simultaneous physical equations of the microgrid electric power network for three-phase voltage imbalance of AC electric power. Further, simulations were performed based on the equations. Using this power network model, we demonstrated that reduced-order modeling techniques with physical structure can offer better accuracy than models without a physical structure.
Forward dynamics simulations have the advantage of assessing performance of novel exoskeleton designs at a low cost. For developing a new passive lower-limb exoskeleton, the simulation needs to represent the sitting posture in which the wearer performs working tasks while maintaining balance with the whole body. The present study constructed a forward dynamics simulation for analyzing and developing a new passive lower-limb exoskeleton; the validity of the simulation was investigated using experimental data. The present method computes the interactions between the exoskeleton and wearer, such as reaction forces, physical posture, and physical load, based on the forward dynamics simulation driven by nonlinear model predictive control (NMPC). The NMPC cost function consisted of the physical load and the fitness of working task with constraints to evaluate balance. As a result, the present simulation represented the characteristic posture when sitting on the exoskeleton in which the wearer performs the working task while maintaining balance with the whole body. However, the simulation computed an upright posture of the lumbar joint that differed from the experimental results and needs to be improved. In future work, the simulation will be modified for representing the valid physical posture when wearing the exoskeleton, such as simulating the physical motion of the same working task as in the experiment, and modeling the interaction between the human, exoskeleton, and ground.
For the mobile robot to move to the target position, it is necessary to safely avoid collisions with surrounding objects. In this study, we propose a new dynamic obstacle avoidance algorithm for mobile robots using AR markers and cameras. The proposed method predicts the trajectory from the change in the position information of the dynamic obstacle, and controls the velocity of the mobile robot so that the mobile robot keeps a certain distance from the dynamic obstacle and avoids collision. Simulation and experiments demonstrate the effectiveness of the proposed method.
Railway level crossing often exhibits long queues and spillbacks due to the temporary closure of the railroad gate or when many vehicles arrive simultaneously. Existing research deals with traffic congestion issues at level crossings through architectural upgrades, traffic signal control, and other safety analysis. We propose an optimal vehicle control method to address the problem without infrastructure changes. Specifically, we plan vehicles’ optimal trajectory to arrive from a suitable distance to smoothly reach and stop at the railway crossing at a designed time to reduce queuing and idling at the level crossing stop line. An extended data-driven model is used to predict the natural arrival timing at the crossing line, which is used as the prediction horizon of the optimal vehicle control problem, ensuring a safe gap with the preceding vehicle. Our optimal control method can prioritize safety constraints, showing improved fuel efficiency and reduced idling and traffic spillovers at our chosen study area compared to traditional driving methods in existing scenarios.
In this paper, we analyze the local identifiability of parameters in linear systems whose system matrix is given by a graph Laplacian. Graph Laplacian is a matrix given by a graph and the weights of the edges, and the weights represent the parameter in those systems. It is important to detect whether the parameter is locally unidentifiable (non-LI) a priori, since any parameter estimation methods cannot estimate non-LI parameters. To this aim, we address a problem to find the condition so that the parameter is non-LI for any initial state. Then, we obtain a sufficient condition for the parameter to be non-LI. The condition is given based on the number of vertices and edges and the number of distinct eigenvalues of the graph Laplacian. We give a system that satisfies the condition.
In this study, we introduce a vehicle tracking system for drone imagery, utilizing the real-time object detection network YOLOv5 to get vehicle location and cropped images. The system analyzes the segmented regions' histograms, compares them with previous frames, and identifies identical objects for tracking. The algorithm is designed to compare objects within a specific radius using coordinate information, enhancing histogram comparison efficiency. The MOTA (Multi-Object Tracking Accuracy) showed 90%, but the limited environment of data usage and experiments must be considered. The findings suggest that the real-time performance of the vehicle tracking system can be applied in various fields such as traffic control, vehicle management, and accident response.
This paper provides an interactive shape memory alloy (SMA) actuator-based nonlinear fault tolerant control (FTC) for the flexible arm with an active compensating unit. In control system structure, two SMA wires in parallel are unified an interactive actuator to govern the arm vibration displacement. SMA actuator-based nonlinear FTC system is designed by using operator-theoretic approach. Meanwhile, multiple feedback loops (MFLs) with an active compensation unit is to obtain the desired tracking performance, which the compensation unit is to eliminate the effect of non-invertible hysteresis factor from SMA actuator. The robust stability of the designed control system can be guaranteed with robust stabilization condition. When the unknown fault behavior of the actuator occurs, the fault tolerant characteristics can ensure the safety and reliability of the control system. Finally, the simulation cases verify the effectiveness of the proposed scheme under the normal and faulty states of the actuator. The results show that even if the one-sided actuator loses the input signal, the designed control system can still remain stable the arm vibration displacement, as well as has a good fault-tolerant characteristics for actuator faults. Compared with the classic Proportional-Derivative (PD) controller, the proposed method not only can quickly the vibration suppression with less time but also with lower input energy. The effectiveness and reliability of the SMA actuator-based FTC system is confirmed in case of actuator faults.
In recent years, disasters caused by localized torrential rains of short duration due to linear precipitation belts have frequently occurred. Therefore, we are considering the early detection of localized heavy rainfall to the extent that a disaster occurs by sound information from a distant place, and as a preparation, we are trying to know the relationship between rain sound and rainfall. In this paper, we report on a prototype sensor for measuring the kinetic energy of raindrops using strain gauges, which is mounted on multiple sensor nodes.
In recent years, civil engineering and construction structures such as bridges and tunnels are aging, and periodic inspections are becoming necessary. Therefore, robotics is expected to inspect bridges more efficiently and safely, and robots for inspection of civil engineering and construction structures are widely researched. In this study, we propose a combination of a propeller mechanism and an EPM (Electro Permanent Magnet) wheel to allow the robot to move on bridges of different materials. On concrete and other parts, the robot is driven by a propeller. When it detects that the bridge material is iron, the propeller is turned off and moved by the magnetic force of the EPM wheel.
For the acquisition of sports skills, it is important not only to support with visual information such as video information, but also through interactive support methods using somatosensory information. In this study, we focus on the sport of skateboarding and, develop a device to measure the center of pressure (COP) on the skateboard surface in order to support skill acquisition. This paper describes the fabrication of a skateboard system equipped with resistance pressure distribution sensors. Then, regarding the evaluation of the development system, we verified the measurement performance and the difference in the skateboard skill level when going straight. Experimental results showed that there was a difference in the magnitude of his COP blur during straight-ahead skateboarding between experienced and inexperienced skateboarders.
Data-driven control including V-Tiger has been actively studied in recent years. V-Tiger can predict the feedback loop response, but is sensitive to noise. Therefore, this paper introduces High-order ARX model identification and Confidence interval analysis to predict confidence interval of stability margins of control system designed by V-Tiger. The effectiveness is verified using pneumatic artificial muscle.
Multi-objective reinforcement learning (MORL) for robot motion learning is a challenging problem not only because of the scarcity of the data but also of the high-dimensional and continuous state and action spaces. Most existing MORL algorithms are inadequate in this regard. However, in single-objective reinforcement learning, policy-based algorithms have solved the problem of high-dimensional and continuous state and action spaces. Among such algorithms is policy improvement with path integral (PI2), which has been successful in robot motion learning. P$\mathrm{I}^{2}$ is similar to evolution strategies (ES), and multi-objective optimization is a hot topic in ES algorithms. This paper proposes a MORL algorithm based on P$\mathrm{I}^{2}$ and multi-objective ES, which can handle the problem related to robot motion learning. The effectiveness is shown via numerical simulations.
As the global aging trend intensifies, the health of older people has become a focal point for governments and international organizations. Posture state detection is particularly critical in elderly health management, but the current detection technology’s accuracy is still insufficient to meet application requirements. This paper proposes an improved method based on feature engineering to enhance the accuracy of posture state detection for older people. This method combines simulated data and automatic feature extraction techniques to strengthen machine learning algorithms’ learning ability to capture motion features from cameras. With this innovative approach, we can lower the entry barrier for machine learning technology, enabling more people to utilize machine learning techniques to solve practical problems. Additionally, this solution effectively reduces the risk of privacy leakage for elderly individuals. This research is dedicated to improving the accuracy and security of posture state detection for older people, protecting their privacy, and promoting their health and well-being. Through this innovative method, we hope to contribute to elderly health management and address the challenges the global aging population brings.
This study shows how to prevent the rank deficiency in the thrust allocation matrix by giving each rotor of a 2-DOF tiltable coaxial rotor a difference in rotational speed to generate counter-torque. In addition, in order to avoid a singular attitude at a pitch angle of 90 degrees when moving along a wall, we propose a method of expressing the attitude angle of the UAV as a quaternion and controlling translational movement while maintaining the singular attitude and controlling the force through impedance control. The usefulness of this method is verified by simulating wall running in an environment that includes a wall model.
The safety and comfort of autonomous vehicles are currently important research directions. As an evaluation indicator of vehicle comfort during driving, how to reduce the probability of motion sickness among passengers inside the vehicle has become a key research topic in the field of autonomous vehicles. This article proposes a novel method that considers the vertical stability of the vehicle and reduces the incidence of carsickness based on the active suspension of the vehicle. Active suspension, as a core part of the vehicle chassis-by-wire, plays a role in reducing ground roughness excitation and enhancing vehicle handling stability. Firstly, a dynamic model of the vehicle’s active suspension was established based on the actual vehicle, and then an adaptive fuzzy PID algorithm was designed to optimize the performance indicators of the active suspension, such as body acceleration and suspension dynamic deflection. A quantitative model of the motion sickness mechanism was also established. By comparing different algorithms, it is shown that the proposed method can more effectively reduce the incidence of motion sickness among passengers in the vehicle. This control method can effectively improve vehicle comfort and contribute to the promotion and application of autonomous vehicles.
This paper proposes a method that can improve the performance of multi-object tracking by modifying the Kalman filter which is used to estimate the state of the objects in the video sequence and matching strategy to prevent objects obtained at the current frame in the video sequence from being associated with the wrong trajectory. The Kalman filter used in multi-object tracking is conventionally structured to use only bounding box location information as an observation system. To improve the accuracy of the filter estimation, we added velocity information derived from the difference of positional measurements to the observed data and introduced a process in which the observation noise covariance varies with the confidence score of the object detection. This enables tracking that is more nearly ground-truth. We also introduced gating as an improvement of the matching strategy. Gating is configured to compare the speeds that each track of the object and the matching candidate of the results of object detection with the Euclidean distance, and if this distance exceeds a threshold, the matching is canceled. The proposed method is introduced ByteTrack, which is the base of several object tracking methods reported in recent years. As a result, we achieved 90.1 MOTA, 83.7 IDF1, and 0.117 MOTP on the training set of MOT17.
