Weight smoothing to improve network generalization

A weight smoothing algorithm is proposed in this paper to improve a neural network's generalization capability. The algorithm can be used when the data patterns to be classified are presented on an n-dimensional grid (n>/=1) and there exists some correlations among neighboring data points within a pattern. For a fully-interconnected feedforward net, no such correlation information is embedded into the architecture. Consequently, the correlations can only be extracted through sufficient amount of network training. With the proposed algorithm, a smoothing constraint is incorporated into the objective function of backpropagation to reflect the neighborhood correlations and to seek those solutions that have smooth connection weights. Experiments were performed on problems of waveform classification, multifont alphanumeric character recognition, and handwritten numeral recognition. The results indicate that (1) networks trained with the algorithm do have smooth connection weights, and (2) they generalize better.     

模拟信号的恢复及其误差研究

数模混合仿真中模拟信号的恢复及其误差对整个仿真过程具有重要影响。本文探讨了用零阶保持和一阶保持进行信号恢复的方法，比较了它们的误差及工作特点。特别指出了一阶保持中，由于存在积分器复原时间及数字机操作时间的影响而会带来附加的误差，因此一阶保持方法较难达到其理论的精度。     

在经典极限下量子斯特林制冷机的优化

经济利润率是评价一个实际热力装置的主要指标之一。将有限时间热力学，非平衡量子统计理论和yong经济学相结合，导出了量子斯特林制冷机的最大利润率以及对应的性能界限，其结果与实际斯特林制冷机的优化设计和模型评估提供了一个最佳的预选方案。     

Magnetization and rotation of MTG HTSC ring in magnetic field

The magnetization of a melt-texture growth (MTG) HTSC ring has been studied. It is shown that the magnetic field inside the ring is larger than the external field under a certain range of external magnetic fields. We have also investigated the magnetic field dependence of the response of a detective coil near a rotating superconducting ring. The responses of the MTG sample are different for different cooling methods.     

A Linear Cross-Coupled Control System for High-Speed Machining

We present a linear cross-coupled controller to improve highspeed contouring accuracy independently of tracking accuracy in a biaxial machine tool feed drive servomechanism. Unlike conventional cross-coupled controllers, the cross-coupled controller presented here is a linear system, so it is very easy to perform the stability and steady-state error analysis, and to optimise the controller parameters. The proposed controller is evaluated experimentally on a CNC LOM machine and compared to an uncoupled controller and a conventional cross-coupled controller. Controller performance is evaluated for a circular contour at a feedrate of 30 m min _1 . The experimental results show that the proposed controller can greatly reduce the contour error at large feedrates. The linear cross-coupled controller is simple to implement and is practical.     

既有钢筋混凝土框架结构抗震加固若干问题探讨

针对钢筋混凝土框架结构抗震加固问题的复杂性 ,本文深入分析了抗震加固结构的受力特征 ,探讨了卸荷对加固构件承载力的影响 ,并针对抗震加固后的“隐含危险转移”问题提出了相应的对策。最后 ,提出了抗震加固的基本原则 ,可供结构抗震加固时参考     

An experience of tunnelling in mudstone area in southwestern Taiwan

Many tunnels will be constructed in southwestern Taiwan in the upcoming decade to meet the huge demands of transportation, energy and water infrastructure projects. Mudstone strata cover more than a 1000 km² area, consisting mainly of massive mudstone or alternation of mudstone and sandy layer, exhibiting unfavorable geological and hydrological characteristics, making it extremely difficult for tunnelling. This paper presents the lessons learned from three tunnelling projects in the 1990s in a mudstone area. In addition to discussing the rock behavior in the vicinity of tunnel and its failure patterns, the monitoring data during tunnelling are presented as well. Finally, the effective method of design and construction are recommended for tunnelling in mudstone area.     

Finite-difference time-domain simulation of scattering from objectsin continuous random media

A three-dimensional (3D) finite-difference time-domain (FDTD) scheme is introduced to model the scattering from objects in continuous random media. FDTD techniques have been previously applied to scattering from random rough surfaces and randomly placed objects in a homogeneous background, but little has been done to simulate continuous random media with embedded objects where volumetric scattering effects are important. In this work, Monte Carlo analysis is used in conjunction with FDTD to study the scattering from perfectly electrically conducting (PEC) objects embedded in continuous random media. The random medium models under consideration are chosen to be inhomogeneous soils with a spatially fluctuating random permittivities and prescribed correlation functions. The ability of frequency averaging techniques to discriminate objects in this scenarion is also briefly investigated. The simulation scheme described in this work can be adapted and used to help in interpreting the scattered field data from targets in random environments such as geophysical media, biological media, or atmospheric turbulence