Semi-supervised clustering employs limited supervision in the form of labeled instances or pairwise instance constraints to aid unsupervised clustering and often significantly improves the clustering performance. Despite the vast amount of expert knowledge spent on this problem, most existing work is not designed for handling high-dimensional sparse data. This paper thus fills this crucial void by developing a Semi-supervised Clustering method based on spheRical K-mEans via fEature projectioN (SCREEN). Specifically, we formulate the problem of constraint-guided feature projection, which can be nicely integrated with semi-supervised clustering algorithms and has the ability to effectively reduce data dimension. Indeed, our experimental results on several real-world data sets show that the SCREEN method can effectively deal with high-dimensional data and provides an appealing clustering performance.

No abstract

A feature extraction for latent fault detection and failure modes classification method of board-level package subjected to vibration loadings is presented for prognostics and health management (PHM) of electronics using adaptive spectrum kurtosis and kernel probability distance clustering. First, strain response data of electronic components is filtered by empirical mode decomposition (EMD) method based on maximum spectrum kurtosis (SK), and fault symptom vector is developed by computing and reconstructing the envelope spectrum. Second, nonlinear fault symptom data is mapped and clustered in sparse Hilbert space using Gaussian radial basis kernel probabilistic distance clustering method. Finally, the current state of board level package is estimated by computing the membership probability of its envelope spectrum. The experimental results demonstrated that the method can detect and classify the latent failure mode of board level package effectively before it happened. board-level package, vibration loading, spectrum kurtosis, kernel probabilistic distance clusteringCitation: Tang W, Jing B, Huang Y F, et al. Feature extraction for latent fault detection and failure modes classification of board-level package under vibration loadings.

Infrasound is a type of low frequency signal that occurs in nature and results from man-made events, typically ranging in frequency from 0.01 Hz to 20 Hz. In this paper, a classification method based on Hilbert-Huang transform (HHT) and support vector machine (SVM) is proposed to discriminate between three different natural events. The frequency spectrum characteristics of infrasound signals produced by different events, such as volcanoes, are unique, which lays the foundation for infrasound signal classification. First, the HHT method was used to extract the feature vectors of several kinds of infrasound events from the Hilbert marginal spectrum. Then, the feature vectors were classified by the SVM method. Finally, the present of classification and identification accuracy are given. The simulation results show that the recognition rate is above 97.7%, and that approach is effective for classifying event types for small samples.

Dynamic spectra (DS) can greatly reduce the influence of individual differences and the measurement environment by extracting the absorbance of pulsating blood at multiple wavelengths, and it is expected to achieve noninvasive detection of blood components. Extracting high-quality DS is the prerequisite for improving detection accuracy. This paper proposed an optimizing differential extraction method in view of the deficiency of existing extraction methods. In the proposed method, the sub-dynamic spectrum (sDS) is composed by sequentially extracting the absolute differences of two sample points corresponding to the height of the half peak on the two sides of the lowest point in each period of the logarithm photoplethysmography signal. The study was based on clinical trial data from 231 volunteers. Single-trial extraction method, original differential extraction method, and optimizing differential extraction method were used to extract DS from the volunteers’ experimental data. Partial least squares regression (PLSR) and radial basis function (RBF) neural network were used for modeling. According to the effect of PLSR modeling, by extracting DS using the proposed method, the correlation coefficient of prediction set ( Rp) has been improved by 17.33% and the root mean square error of prediction set has been reduced by 7.10% compared with the original differential extraction method. Compared with the single-trial extraction method, the correlation coefficient of calibration set ( Rc) has increased from 0.747659 to 0.8244, with an increase of 10.26%, while the correlation coefficient of prediction set ( Rp) decreased slightly by 3.22%, much lower than the increase of correction set. The result of the RBF neural network modeling also shows that the accuracy of the optimizing differential method is better than the other two methods both in calibration set and prediction set. In general, the optimizing differential extraction method improves the data utilization and credibility compared with the existing extraction methods, and the modeling effect is better than the other two methods.