The number of residents in care facilities has increased with the increase in the number of elderly people in Japan. In the previous study, we reported that meal assistance was the most difficult service performed by the care staff. We also reported that swallowing detection using an accelerometer was easier than detection using EMG methods. In this study, we determined a suitable position of the accelerometer to detect a swallowing. The integrated acceleration (iAcc) value at the sternohyoid muscle was significantly larger than that at other positions. We developed a device that equipped an apron with an accelerometer. The device was evaluated, and the iAcc value during swallowing was significantly larger than that during rest in both the young and the elderly people.
This paper presents a novel machine-learning-based method for bed-leaving detection using Elman-type Feedback Counter Propagation Networks (EF-CPNs), which is particularly effective for processing time-series signals. In our earlier study, we have proposed a method based on CPNs, a form of supervised model of Self-Organizing Maps (SOMs), to produce category maps to learn relations among input and teaching signals. In this study, we introduce a feedback loop in CPNs as the second Grossberg layer so that the time-series features can be learnt. Moreover, we develop an original caster-stand sensor using piezoelectric films to measure, via bed legs, weight changes of a subject on a bed. The developed sensor has the features that it does not require a power supply for operations and can be easily installed on existing beds. We evaluate our sensor system by examining 10 people in an environment representing a clinical site. The mean recognition accuracy is 81.0%, while the mean recognition accuracy for the most important behavior terminal sitting is 98.0%. In view of the fact that most falsely recognized patterns belong to the categories of sleeping and sitting which are not so important for bed-leaving detection, we believe that the developed system can be applied to an actual environment as a novel sensor system requiring no restraint of patients.
In recent years, numerous studies on motion classification using surface electromyography (sEMG) have been conducted to realize a myoelectric hand system. This study considered the practical issues in a motion classification method using sEMG, which will be used to control a myoelectric hand. One such issue is the change in the sEMG characteristics that is caused by a change in the position of the sEMG electrode, owing to the replacement of the electrode. Because such changes influence the performance of a motion classifier, it is generally necessary to be relearn the classifier that the position of the electrode may change. To solve this problem, we propose a new classifier update method using a semi-supervised learning technique. In this method, the data measured under electrode position change is categorized into each motion by a semi-supervised learning technique using the relationship between the unsupervised data and the known categorized supervised data measured at a reference position; the classifier is then recalculated using categorized data. The experimental result shows that the performance of our proposed method is maintained even if the position of the electrode is changed.
The purpose is to estimate the elasticity of the anterior tibial muscle in walking from the evoked mechanomyogram by eliminating the motion acceleration with a Kalman filter. The muscle elasticity in walking was 22 times higher than that in resting.
In this paper, we describe a versatile navigation system for autonomous mobile robot that we have developed. As a development background, there is an increasing demand for autonomous mobile robot to play an active part in the human living environment in recent years. That is indicated to need a more versatile to the navigation system. In here, the versatile system that it can be used for indoor or outdoor without limiting the operating environment. To realize this, we propose a navigation system using an electronic map which is easy to access. To perform the localization by comparing the information of shape about the surrounding environment acquired from LIDAR data with electronic map data around the robot's position. In order not to approach into a local hazardous area where are not registered on the electronic map, robots have to detect and avoid the area on the fly. To detect the hazardous area, traversability analysis should be performed in real-time with a 3D point cloud acquiring form the LiDAR data. The collision avoidance maneuver may be carried out by a real-time path planning based on the dynamics of the robot and the results of the traversability analysis. Several experiments along with the route which contains both indoor and outdoor environments in the distance of more than 1,300m travel are carried out to examine the system's versatility.
Reduction of noise in a car cabin due to road irregularities has been a subject of research for many years. For reducing such a noise, we propose the H2-based Active Structural Acoustic Control (ASAC) method that uses detailed acoustic models obtained by system identification, from the actuators and the shakers to the driver's ears sound pressure level. During experiments in a semi-anechoic room with a compact car, we verify the effectiveness of the proposed approach by obtaining overall noise reduction over the frequency range of 80∼400Hz at the driver's ears.
A cooling control method on the basis of a model predictive control (MPC) for a modular data center utilizing fresh air is proposed. The proposed control method reduces the total energy consumption of information communication technology (ICT) equipment and cooling facilities in the data center while considering a relationship between energy savings and the temperature management of ICT equipment. The proposed method controls the maximum temperature of the central processing unit (CPU) in all of the servers by using the data center's cooling fans. In designing the proposed method, a prediction model was developed to represent the CPU temperature by the revolution speed of the data center's fans, the fresh-air temperature, the utilization of servers, and other factors. Furthermore, the proposed control method was applied to the actual modular data center. The energy consumption of the proposed method is compared with that of a conventional method, which has controlled the temperature difference between the inlet and outlet of the server racks on the basis of proportional integral (PI) control. Actual comparison experiments with the conventional method are provided to validate the effectiveness of the proposed method. The results show that the proposed method realizes energy savings of more than 20% when compared with the conventional method in the actual modular data center.
In this paper, load frequency control problem of a microgrid is discussed. The control is conducted to retain power demand supply balance by controling output of generators in power system. However, it becomes harder to retain the balance because of large-volume injection of renewable energy. For the problem, two approaches are considered in this research field. One is to use dispersion type power sources like batteries, and the other is to apply new control theories. Therefore, we focus on a microgrid which holds a wind turbine generator, a diesel generator and a battery, and propose a technique based a static H∞ control. In the beginning, we design the generalized plant considering the difference of the response speed of a diesel generator and a battery. Then, static H∞ control is applied to the generalized plant. Moreover, an LMI condition to avoid control gains to cause too big control input are combined with static H∞ control. In numerical simulation, we validate that the controller can suppress frequency deviation and deviation of residual capacity of the battery. Futhermore, we show that the approach can design controllers with lower dimensions, and avoid undesirable control gains. In experiment with a generator and a variable resistance, the effectiveness of the controller is also confirmed.
When simple adaptive control (SAC) is applied to vibration systems which include antiresonance modes, there is a problem in that undesirable vibration input is generated. A solution is to use a parallel feedforward compensator (PFC) which makes expansion system consisting of the plant and the PFC into desirable frequency response. On the other hand, the PFC must provide ASPR property to the expansion system. PFC design method based on frequency response fitting is proposed. The matching condition is described using LMI/BMI conditions, and a desirable PFC is obtained by solving an optimization problem described by LMI/BMI conditions. Effectiveness of the proposed method is verified by numerical simulations.
In this paper, we report relative comparison on leaning speed of fish in an antagonistic relation of prey and predator—the prey is fish and the predator is a robot seeking to catch the fish by a net attached robot's hand by means of visual servoing. It was confirmed that the fish have found escaping strategy by itself, e.g., staying at corners of a pool where the net is inhibited from closely approaching to the corners to avoid the net clashing to the pool wall. The effectiveness of the conceived escaping strategies by fish has been measured as learning speed that describe decreasing tendency of how many fish could be caught in constant time when the fish caught be released immediately to the same pool. To overcome such fish's ability to conceive new strategies for escape, in this paper, chaos and randomness have been added to the net motion. The effectiveness of chaos and randomness are experimentally examined to judge whether they can decrease the fish's learning speed.