A daily diagnosis system for bridge monitoring and maintenance is developed based on wireless sensors, signal processing, structure analysis, and displacement analysis. The vibration acceleration data of a bridge are firstly collected through the wireless sensor network by exerting. Nonlinear independent component analysis (ICA) and spectral analysis are used to extract the vibration frequencies of the bridge. After that, through a band pass filter and Simpson's rule the vibration displacement is calculated and the vibration model is obtained to diagnose the bridge. Since linear ICA algorithms work efficiently only in linear mixing environments, a nonlinear ICA model, which is more complicated, is more practical for bridge diagnosis systems. In this paper, we firstly use the post nonlinear method to change the signal data, after that perform linear separation by FastICA, and calculate the vibration displacement of the bridge. The processed data can be used to understand phenomena like corrosion and crack, and evaluate the health condition of the bridge. We apply this system to Nakajima Bridge in Yahata, Kitakyushu, Japan.
In this paper, we propose an algorithm for automatic behavior quantification in laboratory mice to quantify several model behaviors. The algorithm can detect repetitive motions of the fore- or hind-limbs at several or dozens of hertz, which are too rapid for the naked eye, from high-frame-rate video images. Multiple repetitive motions can always be identified from periodic frame-differential image features in four segmented regions — the head, left side, right side, and tail. Even when a mouse changes its posture and orientation relative to the camera, these features can still be extracted from the shift- and orientation-invariant shape of the mouse silhouette by using the polar coordinate system and adjusting the angle coordinate according to the head and tail positions. The effectiveness of the algorithm is evaluated by analyzing long-term 240-fps videos of four laboratory mice for six typical model behaviors: moving, rearing, immobility, head grooming, left-side scratching, and right-side scratching. The time durations for the model behaviors determined by the algorithm have detection/correction ratios greater than 80% for all the model behaviors. This shows good quantification results for actual animal testing.
This paper considers both the non-stationarity as well as independence/uncorrelated criteria along with the asymmetry ratio over the electroencephalogram (EEG) signals and proposes a hybrid approach of the signal preprocessing methods before the feature extraction. A filter bank approach of the discrete wavelet transform (DWT) is used to exploit the non-stationary characteristics of the EEG signals and it decomposes the raw EEG signals into the subbands of different center frequencies called as rhythm. A post processing of the selected subband by the AMUSE algorithm (a second order statistics based ICA/BSS algorithm) provides the separating matrix for each class of the movement imagery. In the subband domain the orthogonality as well as orthonormality criteria over the whitening matrix and separating matrix do not come respectively. The human brain has an asymmetrical structure. It has been observed that the ratio between the norms of the left and right class separating matrices should be different for better discrimination between these two classes. The alpha/beta band asymmetry ratio between the separating matrices of the left and right classes will provide the condition to select an appropriate multiplier. So we modify the estimated separating matrix by an appropriate multiplier in order to get the required asymmetry and extend the AMUSE algorithm in the subband domain. The desired subband is further subjected to the updated separating matrix to extract subband sub-components from each class. The extracted subband sub-components sources are further subjected to the feature extraction (power spectral density) step followed by the linear discriminant analysis (LDA).
In this paper, we consider signal interpolation of discrete-time signals which are decimated nonuniformly. A conventional interpolation method is based on the sampling theorem, and the resulting system consists of an ideal filter with complex-valued coefficients. While the conventional method assumes band limitation of signals, we propose a new method by sampled-data H∞ optimization. By this method, we can remove the band-limiting assumption and the optimal filter can be with real-valued coefficients. Moreover, we show that without band-limited assumption, there can be the optimal decimation patterns among ones with the same ratio. By examples, we show the effectiveness of our method.
In the stabilization problem for linear boundary control systems of parabolic type, we have just recently obtained a criterion on the smallest number of the sensors. We show in this note that a similar result holds on the number of the actuators, the best case of which is equal to 1, necessary for stabilization.
The optimal route recommendation in navigation systems is often considered to be the optimal route recommendation between two locations, i.e., an origin and a destination. However, in practical scenarios, traveling to several intermediate destinations before the final destination needs to be considered. Conventional route search algorithms cannot consider such restrictions in the route search. In this paper, a method to find the optimal route via several intermediate destinations is proposed. The proposed method is divided into three steps. In the first step, the conventional route search algorithm is used to find the optimal traveling time and optimal route among the origin, intermediate destinations and final destination. In the second step, the visiting order of the intermediate destination is optimized using the population based RasID-D (RasID-DP) to minimize the total traveling time. Finally, the optimal route from the origin to destination is determined based on the results of the previous steps. The proposed method is evaluated based on the efficiency of the optimization of the visiting order of intermediate destinations. Simulation results show that RasID-DP based optimization can find better solutions efficiently.
We provide a useful method for calculating the state vector of a state equation efficiently in a max-plus algebraic system. For a discrete event system whose precedence relationships are represented by a directed acyclic graph, computing the transition matrix, which includes the Kleene star operation of a weighted adjacency matrix, is occasionally the bottleneck. On the other hand, the common objective is to compute the state equation, rather than the transition matrix itself. Since the state equation is essentially the multiplication of the transition matrix and vector, we propose algorithms for efficiently calculating the multiplication and left division of the Kleene star of an adjacency matrix and a vector.
This paper proposes a method for evaluating control performance of a switching control system for which the switching strategy consists of so-called maximal output admissible sets. The proposed method gives an upper bound of the control performance index. The key to calculate it is the use of a dwell time. Two types of the algorithms are presented so as to seek for the dwell time. The proposed method is summarized as a theorem that provides an upper bound of the control performance index. Finally, a numerical example is used to make sure of what the theorem states.
This paper presents a software tool, entitled ODQLab, to design dynamic quantizers for discrete-valued input control. ODQLab is a Matlab-based graphical tool, which enables us to obtain and verify dynamic quantizers without the knowledge of any sophisticated design theory. In this paper, we introduce the software tool with the underlying theory. Its effectiveness is demonstrated by design examples and experimental evaluations.
This paper proposes a novel ODE-type Smith predictor for nonlinear control systems with transmission delays. The presence of time delays in the transmission of the control signals and the sensor signals may severely affect the performance of the closed-loop system and prevent the successful application of established nonlinear control methodologies. First, we investigate a backstepping interpretation of Smith predictors for linear time-variant (LTV) systems in the state space. Second, we propose a new nonlinear Smith predictor, the error dynamics of which is locally stabilized by another Smith predictor for the LTV systems. The Smith predictor can reduce the computational cost of the controller, because integral equations, which should be solved at each time in previously proposed predictive controllers, are replaced by inner-product operations.