基于改进的支持向量回归机算法的磁记忆定量化缺陷反演

针对焊缝缺陷磁记忆检测中存在定量化反演难题,建立了基于改进的支持向量回归机定量反演模型.以预制不同尺寸未焊透和夹渣缺陷的Q235焊接试样为试验材料,进行磁记忆扫描检测发现:缺陷位置的磁记忆信号特征参数随尺寸变化而呈现一定的变化规律,但同时存在分散性和不确定性.鉴于磁记忆信号样本的有限性、分散性和非线性,首先将提取到的磁记忆特征参数进行归一化处理,引入支持向量回归机建立焊缝缺陷磁记忆定量反演模型,并进一步利用模拟退火算法对支持向量回归机参数进行优化,使目标函数达到全局最优而非局部最优.最后,考虑到由磁记忆信号逆向反推缺陷的三维尺寸,存在解的不确定性,为此在缺陷单维尺寸反演模型的基础上,通过构建多层结构的支持向量回归机进行多尺寸反演输出,建立了基于模拟退火支持向量回归机的焊缝缺陷磁记忆定量反演模型,结果表明:未焊透缺陷尺寸反演最大相对误差为7.96%,夹渣缺陷为4.97%,为焊缝缺陷的磁记忆反演与定量化评价提供一种新的思路.      

基于磁记忆检测的桥钢箱梁翼缘损伤状态力磁关系

金属磁记忆检测技术由于其能够快速便捷的对铁磁性构件的损伤进行识别,且被认为具有识别隐性损伤的能力,而被广泛研究。为推进金属磁记忆检测技术在桥钢箱梁损伤检测方面的应用,对桥钢箱梁进行了静力受弯试验,提取其变形最严重的上翼缘磁信号分布,建立了损伤区域力与磁信号和磁信号梯度的关系曲线,并提出用磁场梯度指数来表征钢梁的受力和损伤状态。结果表明:上翼缘磁信号曲线与应力变化形态正好相反,磁信号曲线在进入塑性后发生反转变为负值,且随应力变化的速度增快,可以判断构件进入塑性状态,即将发生损伤;磁场梯度曲线在损伤最严重的区域出现最大值,且随着荷载的增大,磁梯度最大值点不断向钢梁中间移动,由此可以进行破坏状态的预警;磁场梯度与应力关系曲线可将构件整个受力过程明显的区分为初始、屈服、塑性、损伤4个状态;可以用磁场梯度指数来进行构件受力状态与损伤状态的表征。该研究可为金属磁记忆检测技术在桥钢箱梁损伤状态的定量评估和预警方面的应用提供依据和参考。      

基于IPSO-GRU深度学习算法的海底管道缺陷尺寸磁记忆定量反演模型

针对海底管道缺陷磁记忆定量反演的难题,提出一种基于改进粒子群优化的门控循环神经网络模型,即IPSO-GRU模型。以两端焊有盲板的X52管道作为实验材料,其上预制有不同直径、深度的缺陷,采用TSC-5M-32磁记忆检测仪,外接11-6W非接触探头,进行水下磁记忆检测试验,提取不同缺陷尺寸的磁记忆信号特征值。考虑到磁记忆信号特征值随缺陷尺寸呈复杂的非线性变化,引入门控循环神经网络,利用其双门结构能够记忆缺陷处的信号特征,非线性回归拟合能力强的特点,构建海底管道缺陷定量反演模型,进一步考虑到模型超参数选择的随机性,采用改进粒子群算法进行超参数寻优。验证结果表明:该模型对缺陷深度反演平均精度达96%;对缺陷直径反演平均精度达93%,为海底管道缺陷的磁记忆定量化识别与反演提供了新的思路和方法。      

基于粒子群优化的过程神经网络学习算法

基于粒子群优化为过程神经元网络提出了一种新的学习算法.新算法在对网络输入函数和连接权函数进行正交基函数展开后,将网络中的结构参数和其他参数整合成一个粒子,再用粒子群优化算法进行全局优化.新算法不依赖于函数梯度信息,不需要手动调节网络结构.粒子群优化具有良好的全局优化性能和收敛性能,保证了过程神经元网络的全局学习能力和新学习算法的收敛能力,更好地发挥过程神经网络的逼近性能.两个实际预测问题的实验结果表明,基于粒子群优化的学习算法比现有的基于梯度的基函数展开方法以及误差反传神经网络模型具有更好的预测精度.     

40Cr钢应力磁化过程中的磁化反转特征

金属磁记忆检测技术是一种可早期检测铁磁构件应力集中程度的新方法,但进一步定量评价和广泛应用的瓶颈问题是复杂的应力磁化反转特征.对40Cr钢圆棒试件在不同最大拉力下进行反复加载-卸载拉伸试验,测定试件表面某确定点处漏磁场与拉应力的关系.试验结果表明,当试件处于弹性变形阶段时,漏磁场强度与拉应力的变化规律为线性关系;当试件受力超过屈服强度时,漏磁场强度与拉应力的变化规律变为折线,表现为先减小后增大再减小的应力磁化反转现象.随着最大拉力的增大,应力磁化反转极值点位置向较高拉应力方向移动,漏磁场强度最大变化量△Bmax也逐渐增大.     

金属磁记忆检测技术研究新进展与关键问题

金属磁记忆检测技术是一种适用于铁磁材料的新兴的无损检测技术,主要优势在于无需外加激励磁场源,即在天然地磁场的激励作用下,通过测量材料表面的漏磁信号,就能够对铁磁构件的早期损伤进行检测,避免结构或构件发生突然的脆性破坏。针对近10余年金属磁记忆检测技术的研究现状,概述了该技术的理论基础,总结了该技术理论研究、试验研究以及工程应用新进展,探讨了磁记忆检测技术的损伤评判准则,分析了影响磁记忆检测信号的因素,基于此,提出了磁记忆检测技术目前存在的问题和未来的研究发展方向。      

不同材料在不同应力状态下的磁记忆信号特征

磁记忆是力磁效应的一种表现,目前关于力磁效应的基本规律及机理研究还没有统一的定论。为了进一步对磁记忆效应的机理及不同材料不同应力状态下磁记忆信号特征进行试验性的研究,对不同铁磁材料(Q235,20号钢,45号钢)圆棒试件进行静载拉伸试验,分别测量某固定点处在线加载和在线卸载2种不同应力状态下的磁记忆信号。试验结果表明:3种不同材料在线加载时磁记忆信号各表现出了不同的力磁效应;在线卸载时测得的磁记忆信号更能有利地分析构件的应力阶段;弹性阶段时磁荷梯度GF变化幅度不大,接近屈服阶段时,磁荷梯度变化幅度变大,可以利用磁荷梯度对低碳钢材料应力定量评价。试验结果为磁记忆效应机理研究提供了一定的依据,也为金属磁记忆检测的定量检测提供了潜在可能性。     

基于自适应搜索的免疫粒子群算法

经典粒子群算法由于多样性差而陷入局部最优,从而造成早熟停滞现象.为克服上述缺点,本文结合人工免疫算法,提出一种基于自适应搜索的免疫粒子群算法.首先,该算法改善了浓度机制;然后由粒子最大浓度值来控制子种群数目以充分利用粒子种群资源;最后对劣质子种群进行疫苗接种,利用粒子最大浓度值调节接种疫苗的搜索范围,不仅避免了种群退化现象,而且提高了算法的收敛精度和全局搜索能力.仿真结果表明该算法求解复杂函数优化问题的有效性和优越性.      

铁磁试件应力磁化过程中的磁化反转效应

在反复加载-卸载条件下,测量了35号冷轧钢试件在不同的最大拉应力作用下试件表面某确定点漏磁场随拉应力的对应关系,并进行了讨论和分析。建立了两种应力磁效应的数学模型,即应力磁化反转效应的应力等效磁场模型和应力磁化反转效应的磁导率模型。理论模型都能很好地反映应力磁化耦合的反转现象,和试验数据基本一致。理论模型的建立进一步证明了金属磁记忆定量检测的可能性。     

Fe-VC复合材料与45~#钢的钨极氩弧焊焊接

采用钨极氩弧焊(TIG)和奥氏体不锈钢焊丝ER309L作为填充金属,对Fe-VC复合材料与45#钢的可焊性进行了研究.借助X衍射仪、扫描电镜分析焊缝金属的物相结构和组织形貌;应用电子探针测试了合金元素在焊缝中的成分分布;按照国家标准,测试了焊接接头的拉伸强度.结果表明:复合材料与45#钢实现了良好的冶金结合,合金元素在焊缝中呈梯度分布,拉伸试样的断裂位置均在复合材料处,表明所采用的焊接方法可靠,能够满足异种金属的焊接要求.     

基于PSO-RBF神经网络模型的岩爆倾向性预测

鉴于岩爆机理的复杂性以及岩爆发生前后信号提取困难的现状,对高应力区进行岩爆倾向性预测研究具有现实意义。为提高岩爆预测的准确性,基于岩爆预测多维非线性的特点,选取4个影响岩爆发生的核心指标作为判决依据,结合粒子群优化算法(PSO)与径向基神经网络(RBF)建立了PSO-RBF神经网络岩爆预测模型。采用试错法确定隐含层节点数后,进一步利用国内外典型工程数据对模型参数隐含层基函数中心ci,隐含层节点宽度σi以及隐含层与输出层间权重因子w进行学习优化以获取最优参数,并将所建立的模型应用于实际工程的岩爆倾向性预测。结果表明:利用该模型预测的岩爆等级与实际岩爆情况基本相符,相对误差率为10%,精度较以往预测方法有显著提高。     

Minimum Weight Design of Fillet Welds Using Convex Formulation

This paper presents a discrete formulation of the nonlinear minimum weight design of welded connections, in which global throat stresses along every fillet are approximated by piecewise linear shape functions. With this formulation it is possible to find the set of throat thicknesses of a welded connection that gives the minimum amount of weld metal necessary to support an arbitrary set of external actions and a stress field in equilibrium with these actions and satisfying a strength criterion. The formulation adopted involves a linear form of objective function and equilibrium equations, which, together with the convex character of the nonlinear constraints, gives a full convex set of constraints and consequently a global optimum of the objective function. Moreover, the structure of the set of constraints makes it possible to demonstrate that stress and throat thickness values are unique at the optimum, which allows the optimum stress distributions to be simplified, to establish practical design recommendations. Realistic conditions can be imposed on the stress field. The problem is solved with a very accessible and well-known scientific library, which allows this method to be implemented by anyone involved in welding practice. The method proposed is developed for the case of a planar connection. Two practical applications using the strength criterion of Eurocode 3 are included, discussing the results obtained for two types of common arrangements and showing that numerical results are consistent and CPU times reasonably low.     

Estimating survival curves with left-truncated and interval-censored data under monotone hazards

We show that the nonparametric maximum likelihood estimator (NPMLE) of a survival function may severely underestimate the survival probabilities at very early times for left-truncated and interval-censored data. As an alternative, we propose to compute the (nonparametric) MLE under a nondecreasing hazard assumption, the monotone MLE, by a gradient projection algorithm when the assumption holds. The projection step is accomplished via an isotonic regression algorithm, the pool-adjacent-violators algorithm. This gradient projection algorithm is computationally efficient and converges globally. Monte Carlo simulations show superior performance of the monotone MLE over that of the NPMLE in terms of either bias or variance, even for large samples. The methodology is illustrated with the application to the Wisconsin Epidemiological Study of Diabetic Retinopathy data to estimate the probability of incidence of retinopathy.     

Control of the Molten Steel Level in the Top Side-Pouring Twin-Roll Casting Process Based on Fuzzy Rules Optimized by Particle Swarm Optimization Algorithm

Twin-roll strip casting is a near-net-shape casting technology that can produce thin steel strips directly from molten steel. Stably controlling the molten steel level is regarded as an important issue to ensure strip quality and casting process stability. As the control of the molten steel level is a time-varying, nonlinear, and multidisturbance complex system, it is difficult to establish an accurate process model for designing a model-based controller. Top side-pouring twin-roll casting is a new kind of twin-roll strip casting technology. This study introduces the control system of the top side-pouring twin-roll casting process. A fuzzy logic controller (FLC) with its fuzzy rules optimized by particle swarm optimization (PSO) is developed to regulate the molten steel level. Simulation results show that the performance of the FLC can be improved while its fuzzy rules are optimized by PSO. The objective function of PSO has a great influence on the optimization of the fuzzy rules. The top side-pouring twin-roll casting experiments are carried out using the FLC with its fuzzy rules optimized by PSO; the results show that strip quality and casting process stability are guaranteed.     

Type IV Creep Damage Behavior in Gr.91 Steel Welded Joints

Modified 9Cr-1Mo steel (ASME Grade 91 steel) is used as a key structural material for boiler components in ultra-supercritical (USC) thermal power plants at approximately 873 K (600 °C). The creep strength of welded joints of this steel decreases as a result of Type IV creep cracking that forms in the heat-affected zone (HAZ) under long-term use at high temperatures. The current article aims to elucidate the damage processes and microstructural degradations that take place in the HAZ of these welded joints. Long-term creep tests for base metal, simulated HAZ, and welded joints were conducted at 823 K, 873 K, and 923 K (550 °C, 600 °C, and 650 °C). Furthermore, creep tests of thick welded joint specimens were interrupted at several time steps at 873 K (600 °C) and 90 MPa, after which the distribution and evolution of creep damage inside the plates were measured quantitatively. It was found that creep voids are initiated in the early stages (0.2 of life) of creep rupture life, which coalesce to form a crack at a later stage (0.8 of life). In a fine-grained HAZ, creep damage is concentrated chiefly in an area approximately 20 pct below the surface of the plate. The experimental creep damage distributions coincide closely with the computed results obtained by damage mechanics analysis using the creep properties of a simulated fine-grained HAZ. Both the concentration of creep strain and the high multiaxial stress conditions in the fine-grained HAZ influence the distribution of Type IV creep damage.     

Fast numerical algorithms for fitting multiresolution hybrid shape models to brain MRI

In this paper, we present new and fast numerical algorithms for shape recovery from brain MRI using multiresolution hybrid shape models. In this modeling framework, shapes are represented by a core rigid shape characterized by a superquadric function and a superimposed displacement function which is characterized by a membrane spline discretized using the finite-element method. Fitting the model to brain MRI data is cast as an energy minimization problem which is solved numerically. We present three new computational methods for model fitting to data. These methods involve novel mathematical derivations that lead to efficient numerical solutions of the model fitting problem. The first method involves using the nonlinear conjugate gradient technique with a diagonal Hessian preconditioner. The second method involves the nonlinear conjugate gradient in the outer loop for solving global parameters of the model and a preconditioned conjugate gradient scheme for solving the local parameters of the model. The third method involves the nonlinear conjugate gradient in the outer loop for solving the global parameters and a combination of the Schur complement formula and the alternating direction-implicit method for solving the local parameters of the model. We demonstrate the efficiency of our model fitting methods via experiments on several MR brain scans.     

Vibration-based Damage Detection of Structures by Genetic Algorithm

Vibration-based methods are being rapidly applied to detect structural damage. The usual approaches incorporate sensitivity analysis and the optimization algorithm to minimize the discrepancies between the measured vibration data and the analytical data. However, conventional optimization methods are gradient based and usually lead to a local minimum only. Genetic algorithms explore the region of the whole solution space and can obtain the global optimum. In this paper, a genetic algorithm with real number encoding is applied to identify the structural damage by minimizing the objective function, which directly compares the changes in the measurements before and after damage. Three different criteria are considered, namely, the frequency changes, the mode shape changes, and a combination of the two. A laboratory tested cantilever beam and a frame are used to demonstrate the proposed technique. Numerical results show that the damaged elements can be detected by genetic algorithm, even when the analytical model is not accurate.