China 
Africa 
Explore chapters and articles related to this topic
Semisupervised Clustering
Published in Charu C. Aggarwal, Chandan K. Reddy, Data Clustering, 2018
Amrudin Agovic, Arindam Banerjee
Local Tangent Space Alignment (LTSA): In LTSA, the tangent space at each point is approximated using local neighborhoods, and a global embedding is obtained by aligning the local tangent spaces. If XiN denotes the matrix of neighbors of xi, then it can be shown [44] that the principal components of XiN give an approximation to the tangent space of the embedding fi. Let gi1, …, gik be the top k principal components for XiN. Let Gi=[e/k,gi1,...,gid]T If Ni are the indices of the neighbors of xi, submatrices of the alignment matrix M are computed as M(Ni,Ni)←M(Ni,Ni)+I−GiGiT for i = 1, …, n. Finally, using M, which is guaranteed to be positive semidefinite, an embedding is subsequently obtained by minimizing the alignment cost: () minffTMf,s.t.f⊥1,‖f‖2=n.
Small-amplitude bogie hunting identification method for high-speed trains based on machine learning
Published in Vehicle System Dynamics, 2023
Jinying Guo, Gexiang Zhang, Huailong Shi, Jing Zeng
Kulkarni et al. [7] have proposed a vehicle running instability detection algorithm (VRIDA) that employs the frequency and phase information of the wheelset acceleration signals to identify hunting instability and suspension fault. Li et al. [8] have established an upper limit of 5 m/s2 for the SABH alarm in the stability monitoring system. Sun et al. [9] have utilised cross-correlation indicators (CCIs) to distinguish SABH and large amplitude bogie hunting (LABH) from the normal samples. Zeng et al. [10] have proposed a bogie instability alarm method that can detect SABH using the periodicity index of the phase trajectory (PIPT), a modification algorithm of the Largest Lyapunov exponent (LLE). Ning et al. [11] have proposed a feature extraction method based on Multiscale Permutation Entropy (MPE) and Local Tangent Space Alignment (LTSA) and have categorised 70 acceleration signals from field tests into four categories: normal, convergent SABH, divergent SABH, and hunting, through unsupervised machine learning. Wang et al. [12] use the energy concentration rate to evaluate the hunting stability of a freight bogie.
A novel measuring system for high-speed railway vehicles hunting monitoring able to predict wheelset motion and wheel/rail contact characteristics
Published in Vehicle System Dynamics, 2023
Jianfeng Sun, Enrico Meli, Xinwu Song, Maoru Chi, Weidong Jiao, Yonghua Jiang
Unlike the model-based methods, the signal-based techniques only use the output signals for analyses and focus on the variations of the measured signals. The signal processing methods, including time-domain analysis, spectral analysis, cross-correlation analysis, wavelet analysis, etc., can be used during the prediction phase. Hung et al. [13] proposed a technique for predicting wheel-climb derailment at low speed by using a time-domain method, and the bogie accelerations and angular velocities are collected as the input. Ning et al. [14] presented a feature extraction method based on Multiscale Permutation Entropy and Local Tangent Space Alignment to distinguish the features of the small amplitude hunting signals, and this method can be used to detect the beginning of hunting. Exploiting the cross-correlation techniques, Sun et al. [15] described a signal analysis-based hunting instability detection methodology, which is able to detect both small amplitude and large amplitude hunting motions.
Dimensionality reduction and classification for hyperspectral image based on robust supervised ISOMAP
Published in Journal of Industrial and Production Engineering, 2022
Shengfeng Ding, Colin Arthur Keal, Lizhou Zhao, Dan Yu
Common manifold learning algorithms include ISOMAP [22], laplacian eigenmaps [23], locally linear embedding [24], and local tangent space alignment [25]. At present, many scholars have successfully applied ISOMAP algorithm to reduce the dimension of hyperspectral data. Hyperspectral image data can be described as a manifold with simple geometric structure in low-dimensional embedding space [26]. The whole manifold can be established by selecting some data representing the whole scene and manifold coordinates of other data are obtained by interpolation in order to solve ISOMAP large computation [27]. The spectral angle, spectral information divergence, and geodetic distance were combined into ISOMAP. The results were better than traditional ISOMAP by calculating the neighborhood based on Euclidean distance in terms of redundant variance and spectral normalized eigenvalue [28]. To determine nonlinear structural characteristics of hyperspectral data, ISOMAP was used to reduce nonlinear dimension and the result was better than minimum noise fraction, which had better data compression performance [29].