基于深度学习的小样本声目标识别方法

声目标分类识别是声源识别领域的核心问题,然而,在应用深层神经网络进行声目标分类识别时,从少量样本中学习(样本复杂度较低)是一个具有挑战性的问题;针对此问题,提出了一种基于深度学习的小样本声目标识别方法,该方法将手工设计特征和对数梅尔声谱特征结合到一起,扩充了深度学习模型的可利用特征量,提高了声信号识别效率和精度;在实验验证中,该方法在测试集上实现了87.6%的识别精度;更进一步地,用较少量的训练样本对该方法和其它几种主流的深度学习模型的性能进行了比较验证,结果表明,该方法只需要更少量的数据即可实现同样的识别精度,在声源探测领域具有一定应用价值。     

Simulation of Fatigue Stiffness Degradation in Prestressed Concrete Beams under Cyclic Loading

In order to investigate and research the fatigue cracking of prestressed concrete fatigue properties and loading and stiffness degeneration process,cyclic loading tests were carried out on six prestressed concrete beams and the stiffness degradation under fatigue was investigated. A simulation model of stiffness degradation is proposed based on the stiffness analysis of the fatigue-damaged section. The elastic modulus of damaged concrete and the effective residual area of steel were introduced as well as an adjusted three-stage concrete fatigue damage evolution model. The strip method was used to analyze concrete damage due to changing stress along the depth of the beam section. The simulation and test results were compared and a method of predicting fatigue deflection was presented based on the simulation model. The predicted results were compared with that of the neural network method. It is in good agreement for the simulation results with the test results. It is only less than 5% error for the simulation model which can reveal the two-stage degradation of prestressed concrete beams under cyclic loading. It is more precise for the simulation prediction method under proper conditions.     

基于新的残差沙漏网络的人脸对齐

针对传统沙漏网络每一层使用单一感受野来提取特征,缺乏对关键点整体和局部关联信息描述的问题,提出了新的残差沙漏网络（NRHG）方法。该方法通过增加新的卷积分支来增加网络的感受野以更好地提取到不同尺度下的特征信息,同时新的感受野增加了网络对整体信息的描述。针对网络不同层相应调整新增卷积分支感受野大小来平衡feature map分辨率和感受野之间的关系,在更好地保留从局部到整体的结构化信息的同时,突出了网络对局部细节特征信息的描述。最后沙漏网络之间采用中间监督,对每一个沙漏网络输出的结果都进行loss计算,以避免网络深度造成梯度消失的问题。通过在300-W、IBUG、COFW数据集上大量的实验证明了该方法的有效性,并且实验结果优于传统的沙漏网络。     

极端海况下FPSO系泊系统安全性评估与分析

针对近年极端海洋性气候频发所导致的FPSO系泊系统安全性问题,从服役多年系泊系统钢缆的剩余强度入手,评估影响船体运动的环境参数,给出系泊钢缆的安全系数。利用准静态的方法估算环境等级的范围,选取南海某型FPSO建立系泊系统耦合计算模型。应用AQWA-DRAFT与AQWA-NAUT模块,采用时域非线性方法对系泊系统进行耦合分析,在满足规范系数的前提下确定现服役系泊系统的定位能力,预报系泊系统所能承受的最大风浪流极值组合及对应的台风等级。通过计算可看出钢缆强度的下降直接影响着系泊系统的定位能力,导致其无法抵御百年一遇的台风,系泊系统可抵御台风等级降至12级。     

机械自紧火炮身管残余应力分析

为计算具有包兴格效应和卸载硬化效应的机械自紧火炮身管的残余应力,对卸载阶段的火炮身管进行弹塑性分析,得到火炮身管卸载阶段的弹塑性解和弹性解,再结合理想弹塑性材料模型加载阶段的弹塑性解,得到了火炮身管反向屈服半径、残余应力和残余径向位移计算公式.为验证公式的准确性,将理论值与实验值进行比较,误差很小.     

切削加工表面完整性建模现状与发展趋势

针对切削加工表面完整性建模方法种类繁多且相互关系揭示不明等问题,总结了各表面完整性指标,包括表面形貌、残余应力、加工硬化和相变的主要建模方法,并从各建模方法的应用范围以及主要成果的角度概括了表面完整性建模的研究现状.论述了解析建模与数值建模方法在预测表面完整性各指标时的优缺点及内在联系,重点分析了数值模拟技术在表面完整性建模中的作用.指出了国内外表面完整性研究体系的区别及联系,并对表面完整性建模的发展方向作了展望.     

余热余压发电输煤控制系统的设计

针对废气余热排入大气的浪费和污染现状,将输煤控制系统应用于余热余压综合节能技术改造项目工程中。该控制系统由PLC输煤控制系统、上位机监控系统以及工业电视监控系统组成。经过一年多的实际运行表明,本控制系统的架构合理、运行可靠,同时降低劳动强度,提高设备利用率,其平稳运行也为其它类似工程提供了实践借鉴。     

基于深度学习的相对地质时间体估计方法

常规自动解释方法难以正确追踪三维地震数据体中被断层错断的地震反射层位。为此,提出了基于深度学习的相对地质时间体估计方法。首先,针对相对地质时间体估计需求,设计一个由编码器—解码器框架组成的相对地质时间体估计网络;其次,采用结构相似性准则为损失函数,利用生成的准确标注合成训练数据集对相对地质时间体估计网络进行训练,使其具备准确地从地震数据体中估计相对地质时间体的能力;最后,通过提取多个恒定的相对地质时间体等值面实现多个地震反射层位的自动追踪。测试结果表明,该方法不仅在验证数据集上显示出优异性能,而且在实际地震数据体上也获得了较好效果;利用估计的相对地质时间体能够一次性获得多个能够表征地层空间形态的地震反射层位。     

Nondestructive measurements of residual stress in air plasma-sprayed thermal barrier coatings

Premature spallation of thermal barrier coatings (TBCs) is a critical issue during the service of gas turbines, and nondestructive evaluation is crucial to address this problem. Herein, a novel approach that indicates delamination by measuring the residual stress evolution of thermally grown oxide (TGO) for air plasma spraying (APS) TBCs is proposed and verified via the combination of photoluminescence piezo-spectroscopy (PLPS) and X-ray computed tomography. A mineral-oil-impregnating approach and a cold-mount low-shrinkage epoxy-mounting approach are used to alleviate the signal attenuation by pores and microcracks in APS TBCs, improving the detectable PLPS signal and X-ray transmission for stress measurement and delamination characterization, respectively. We have nondestructively measured the TGO residual stress mapping in APS TBCs and its evolution with oxidation. Furthermore, the evolution of TGO morphology and critical microcracks are obtained by X-ray computed tomography. The synchronous evolution of TGO residual stress, TGO thickness, and critical microcracks as a function of oxidation time is obtained and correlated. The transition point, as experimentally identified, at which the TGO stress starts to drop, agrees well with the critical moment of microcrack coalescence. This directly verifies that the TBC delamination can be effectively indicated by residual stress evolution of TGO in APS TBCs.     

残余应力集中及其对疲劳极限和短裂纹扩展的影响

缺口根部的残余应力存在应力集中效应,应力集中值不仅取决于缺口几何形状而且和材料的屈服强度有关。残余应力集中系数是一个分布值,其数值甚至可超过理论应力集中系数。软试样喷丸后缺口处的残余应力易于松弛及喷丸时的表面损伤使其对疲劳极限的贡献小。残余压应力使缺口处裂纹闭合力增大甚至可出现非扩展裂纹。