Key Engineering Materials Vols. 467-469

Abstract: Aiming at the problem of diagnosis difficulty caused by too many factors of mechanical system, a kind of diagnosing method based on fault phenomenon was presented. The research on mechanical system fault phenomenon space arrived at conclusion that the emergency of each fault phenomenon subject to 0-1 distribution. Therefore, phenomenon vector corresponding to each fault formed cluster whose accumulation point is expectation of vector. After exclusion of abnormal vectors, the distance discrimination was used to fault diagnosis to establish expert system based on fault phenomenon vector. The confirmed result was return back to fault database so that the system achieve self-learning of real-time diagnosis experiences. Finally, the example on X-type hydraulic excavator proves that the diagnostic method has characteristics of good real-time, simple operation and high diagnostic accuracy.

Abstract: GMM with modified weight is proposed, and it is applied into the audio segment classification. Since GMM is training only by intra-class data, during classification, there may be some components in different GMM overlapping each other. That means the likelihoods of the overlap components are similar, which will induce the confusion during classification. To avoid this, the weight in GMM is modified according to the distance of components. For that component easily to confuse, the weight is deduced to decrease the effect of the component in whole result. Instead, for that component with well separability, the weight is enhanced. Based on the ratio of sub-band energy and the proposed model, an optimized likelihood is further put forward, in which the length of clip can be adjusted. From experiments, it can be drawn that GMM with modified weight has better performance than traditional GMM, and combined with optimized likelihood, the performance can be further enhanced.

Abstract: The binding properties of chloride ion of sulphoaluminate cement and portland cement were studied in different age,different water-cement ratio and different chloride ion content. The results show that binding rate tends to increase as the age and water-cement ratio increase for above two type cements. But the trend is opposite with the increase of total amount of incorporation chloride ion(TAIC), and the early strength of sulphoaluminate cement decreases slightly.

Abstract: As the real-time image acquiring and processing need to be dealt with high speed, a image acquisition and preprocessing system is discussed in this paper. It is built on FPGA( field programmable gate array ) with pipelined and parallel technology. The configurable macro function modules provided by Altera company achieve the Sobel edge detection algorithm. The real-time display the image after edge detection works properly and The new design method shorten the development cycle.

Abstract: One of the major difficulties of wireless sensor network (WSN) applications is how to efficiently manage the large amount of data produced by sensors. The differences from standard database source pose challenges of sensor data management. In this paper, we propose a novel sensor data management architecture based on our extensive discussion on existing works. The hierarchical system model consists of sensor network layer and proxy network layer. Sensor network layer performs limited computation and communication while being managed intelligently by the proxy network. The proxy network receives sensor data, manages sensor data and processes queries in a distributed manner. We also provide insight into possible research directions in this area.

Abstract: Localization capability is usually required and designed for wireless networks. Although many localization algorithms have been proposed, the refinement issue of guaranteeing the location accuracy is still in its early stage of development. This paper compares different ranging technologies for localization measurement. Specifically, we analyze infrared, ultra sonic, radio frequency and ultra wideband as different choices. After a comparative study the paper recommends ultra wideband as the best candidate for accurate range-based localization system in the short-range wireless network scenario.

Abstract: Fast determination of mineral nutrition contents of fruit trees is essential for orchard precision fertilizing management. A multi-spectral imaging system was developed and tested for the measurement of leaf nitrogen content of fruit trees in the study. Images taken using this system included visible images(R-G-B) and near-infrared image(NIR). These images were further processed into several indices such as RVI, NDVI, GNDVI, -log(R) and –log(G). Total 185 leaf samples were picked from Huang-hua pear trees which were planted in three orchards with different nitrogen fertilizing levels. Among them, 135 samples were randomly sorted out as calibration set with the remaining 50 as prediction set. A SPAD-502 chlorophyll meter was used for nitrogen reference measurement. In calibration modeling, leaf front and back faces were photographed respectively. Calibration models were developed based on single variant as well as multiple variants. The result shows that calibration models based on leaf front face are better than those based on leaf back face. Among others, R and G are the most important factors for nitrogen determination with less contribution of B and NIR. Based on the images of leaf front face, R, G, RVI, NDVI, GNDVI, -log(R) and –log(G) were found significantly correlated with nitrogen content with correlation coefficients of prediction (r_pre^2) of 0.7516, 0.7396, 0.7332, 0.7220, 0.7588, 0.7598 and 0.7379 respectively. The linear combinations of R-G-B-NIR, RVI-NDVI-GNDVI and NDVI-GNDVI achieved better prediction accuracy with r_pre^2 of 0.8157, 0.7775 and 0.7661 respectively. To further improve the prediction accuracy, a three-layer BP-ANN was developed with the three combinations as its input data. The result shows that BP-ANN has an excellent performance to predict nitrogen contents. BP-ANN with the input of R-G-B-NIR performs best with r_pre^2 of 0.9386 and maximum error of 3.52(SPAD). The study suggests that multi-spectral imaging system integrated with prediction model of BP-ANN with original reflectance intensity of R-G-B-NIR channels as its input data is promising for in situ measurement of nitrogen content of fruit tree.

Abstract: . Instrumental measurement of soil properties are affected by several factors including soil texture. The classification of soil texture can help improve the accuracy of calibration models for soil measurement. In the study, the effect of soil particle size on the selection of preprocessing methods for principle component analysis (PCA) of soil classification was examined. Total 178 soil samples with five textures were collected from Silsoe Farm, Cranfield, England. After dried and ground, soil samples sieved by a 2mm sieve were named by Mixed Group. A Vis/NIR spectrophotometer (LabSpec2500, spectral range 350-2200nm, ASD) was used for spectral scanning of soil samples. After that, all samples were sieved by a 1mm sieve and divided into two groups: one with particle size less than 1mm named by Thin Group and another with particle size between 1mm and 2mm named by Thick Group. Preprocessing methods of moving average with segment size of 5(MA5), standard normal variation (SNV) and 1st Savitzky-Golay derivatives with smoothing points of 3(Der1(3)) were examined. PCA was applied to evaluate the discriminative capacity of MA5, MA5+SNV and MA5+Der1(3). The score plots of 1st~2nd and 2nd~3rd PCs show that MA5+Der1(3) is the best preprocessing method not only for Thick Group and Thin Group, but also for Mixed Group. MA5+SNV is suitable for Thick Group and Thin Group but does not perform well for Mixed Group. Only MA5 does not perform well in any of three groups. The study suggests that pre-processing with 1st derivative is an essential step for soil classification with various particle sizes using Vis/NIR spectroscopy.

Abstract: Globally exponentially stability (GES) of a class of discrete- time recurrent neural networks with unsaturating linear activation functions is studied. Based on matrix eigenvalue, a new definition of GES is presented. By applying matrix theory, some conditions for GES are obtained. Simultaneously, those conditions are proved without energy functions.

Abstract: This paper presents the theory of computer automatic weed identification , some methods are discussed in detail including spectrum analysis, color identification, texture assessment and figure analysis methods individually. Besides, the research situation in this field is reviewed and the application trend direction in our country is prospected.