卷积神经网络在无纺布缺陷分类检测中的应用

目的针对传统无纺布缺陷分类检测中人工依赖性强、效率低等问题,提出一种能够满足工厂要求的卷积神经网络分类检测方法。方法首先建立包括脏点、褶皱、断裂、缺纱和无缺陷等5种共计7万张无纺布图像样本库,其次构造一个具有不同神经元个数的卷积层和池化层的神经网络,然后采用反向传播算法逐层更新权值,通过梯度下降法最小化损失函数,最后利用Softmax分类器实现无纺布的缺陷分类检测。结果构建了12层的卷积神经网络,通过2万张样本进行测试实验,无缺陷样本准确率可以达到100%,缺陷样本分类准确率均在95%以上,检测时间在35 ms以内。结论该方法能够满足工业生产线中对于无纺布缺陷实时分类检测的要求。     

基于ACFM焊缝体积型和簇群型缺陷检测仿真分析

交流电磁场检测技术(ACFM)对焊缝体积型缺陷和复杂缺陷检测识别需要完善和改进.本文利用ANSYS Maxwell建立了的电磁场仿真模型,采用有限元分析方法研究了长方体缺陷、椭圆形缺陷和圆柱体缺陷这3种不同的体积型缺陷形状对ACFM信号特征的影响,以及不同缺陷簇群的密集程度对检测结果的影响.通过对提取的Bx,Bz特征信号的影响进行分析与探讨,获得了缺陷形状及分布数量对信号曲线的影响规律,可以识别缺陷形状.缺陷簇群的密集分布会影响ACFM检测的准确度,识别不到缺陷的定量尺寸.     

基于小波变换的低碳钢连铸坯超声探伤

利用离散小波变换对粗晶钢材超声检测信号进行处理,比较了使用不同的小波基,并对不同尺度中突出体现的信息分别进行处理后重构信号的去噪效果。结果表明,对不同尺度系数作不同的阈值处理有利于缺陷的识别,这种方法对粗晶钢材料的检测有积极的意义。     

基于多尺度排列熵的舰船辐射噪声复杂度特征提取研究

针对复杂海洋环境中舰船辐射噪声的特征提取问题,提出了一种基于多尺度排列熵的舰船辐射噪声复杂度特征提取方法。分别利用基于EEMD的最强固有模态中心频率法、高低频能量差法和基于复杂度的排列熵与多尺度排列熵提取了五种不同类别、一定样本数量的舰船辐射噪声特征,并将四种特征提取方法所提取的舰船特征分别输入概率神经网络进行分类识别。研究发现,多尺度排列熵是一种一致性好、稳定性强的非线性特征参数,能够从多个维度描述信号的复杂度。实验结果表明,多尺度排列熵特征具有很好的可分性,以多尺度排列熵为特征进行舰船分类识别,识别率显著高于其他舰船辐射噪声特征提取算法。     

基于相控阵激光超声的裂纹衍射增强试验研究

针对超声衍射时差法(TOFD)存在检测精度较差、区域检测可靠性不够和信号信噪比(signal-to-noise ratio, SNR)低等问题,提出了一种基于光纤皮秒激光器和高速旋转镜的相控阵激光超声裂纹检测方法。利用有限元方法模拟热弹机制,建立二维瞬态激光超声力-固耦合模型产生横波与纵波在缺陷处发生的衍射现象,分析了裂纹尖端不同奇异点、声波不同中心频率和相控阵激励源不同位置对声波衍射的影响,通过衍射信号的信噪比和位移幅值两个计算指标来分析变化规律,并进行了试验验证。结果表明:数值模拟与试验结果有较好的一致性,相控阵激光源较传统单束激光源对衍射信号幅值和信噪比有明显的增强作用,纵波衍射信号信噪比较理想;衍射信号幅值随裂纹尖端奇异点增加和声波中心频率减小而增大;信噪比随尖端奇异点增加而增大,随声波中心频率一定范围增加无明显变化,随激光源距离的增加呈现先增加后减小的趋势;缺陷定量分析时计算出的裂纹长度与实际裂纹的误差均不超过6.8%。     

油气管道腐蚀缺陷分类识别技术研究

缺陷准确量化是管道漏磁检测领域中长期存在的一个难点,而对缺陷进行科学的分类是实现准确量化的重要前提。针对不同形态的缺陷,分析其特征参数对漏磁信号的影响因素,建立用于缺陷分类的BP神经网络模型,设计改进的Levenberg-Marquardt算法用于网络训练,并利用Ansoft Maxwell 3D建立仿真缺陷数据作为样本进行测试。结果表明:改进后的神经网络系统可实现对缺陷的有效分类,改善传统算法分类精度低、误差大的缺点,收敛速度大幅度提高。该方法已成功应用于大型油气田软件工程领域,为实现缺陷准确量化提供基础和依据。     

波纹管内部脱浆状态的超声检测方法

为了检测波纹管内部不同脱浆状态的缺陷问题,提出了一种采用变分模态分解（Variational Mode Decomposition,VMD）与蚁群神经网络相结合的超声检测方法。将检测到的回波信号进行变分模态分解,将分解后信号的多尺度样本熵作为特征参数,输入到蚁群神经网络中进行检测。实验结果表明利用VMD分解方法与蚁群神经网络相结合可以对波纹管内部横向缺陷进行有效判断。     

基于多尺度韦伯脸和梯度脸的复杂光照下人脸识别研究

针对复杂光照变化影响人脸识别准确性的问题,提出了一种基于多尺度韦伯脸和梯度脸相结合的复杂光照下人脸识别方法。首先定义了能够有效描述人脸纹理结构的多尺度韦伯脸,一定程度上减弱了不同光照条件的影响;其次融合多尺度韦伯脸和梯度脸提取人脸光照不变量;最后利用SVM多类分类算法实现人脸识别。使用CMU PIE与Yale B人脸库进行验证,结果表明:提出的算法能够有效消除复杂光照变化对人脸识别的影响,即在光照极差情况下,单样本图像作为训练图像也可以有很好的识别效果,且识别率显著高于韦伯脸、多尺度韦伯脸和梯度脸。     

基于欧氏距离方法的缺陷识别技术研究

通过分析常见的印刷缺陷特征,提出了基于缺陷几何形状和墨量变化共同判断缺陷类型,即通过欧氏距离方法检测缺陷的形状、位置等特征,并设计了一种色度和密度转化算法,来判断印刷缺陷的墨量变化,成功实现了对印刷品缺陷的提取、分类及反馈。     

焊接过程声信号在线检测技术现状与展望

基于焊接过程声信号的声学检测技术是一种实时采集和检测缺陷的有效方法,对材料内部结构变化反映明显。概述了常规焊接缺陷无损检测方法的优缺点,对焊接过程声信号的分类和声信号采集系统进行了论述,分析总结了声发射信号和可听声信号的发声机理、检测原理和信号处理方法,并阐述其在焊接质量在线检测领域的研究现状及应用,特别是焊接缺陷、熔滴过渡形式和熔透状态的识别与预测。为了满足在线检测要求,包含声学检测在内的多传感信息融合焊接系统研究是关键,智能传感、信号处理、自动控制与人工智能技术的进一步研究能促进焊接产业的长足发展。     

Frequency invariant classification of ultrasonic weld inspection signals

Automated signal classification systems are finding increasing use in many applications for the analysis and interpretation of large volumes of signals. Such systems show consistency of response and help reduce the effect of variabilities associated with human interpretation. This paper deals with the analysis of ultrasonic NDE signals obtained during weld inspection of piping in boiling water reactors. The overall approach consists of three major steps, namely, frequency invariance, multiresolution analysis, and neural network classification. The data are first preprocessed whereby signals obtained using different transducer center frequencies are transformed to an equivalent reference frequency signal. Discriminatory features are then extracted using a multiresolution analysis technique, namely, the discrete wavelet transform (DWT). The compact feature vector obtained using wavelet analysis is classified using a multilayer perceptron neural network. Two different databases containing weld inspection signals have been used to test the performance of the neural network. Initial results obtained using this approach demonstrate the effectiveness of the frequency invariance processing technique and the DWT analysis method employed for feature extraction.     

高噪下的钢丝绳断丝定量检测

研究高噪情况下的钢丝绳断丝定量检测信号处理。利用时域与频域相结合的处理算法实现断丝的特征提取，采用这些方法对高噪下的钢丝绳磁检测信号进行处理，能够清楚地区分噪声和损伤信号。该方法运算量小，适合于实时在线的钢丝绳检测。     

Signal detection and noise suppression using a wavelet transform signal processor: application to ultrasonic flaw detection

The utilization of signal processing techniques in nondestructive testing, especially in ultrasonics, is widespread. Signal averaging, matched filtering, frequency spectrum analysis, neural nets, and autoregressive analysis have all been used to analyze ultrasonic signals. The Wavelet Transform (WT) is the most recent technique for processing signals with time-varying spectra. Interest in wavelets and their potential applications has resulted in an explosion of papers; some have called the wavelets the most significant mathematical event of the past decade. In this work, the Wavelet Transform is utilized to improve ultrasonic flaw detection in noisy signals as an alternative to the Split-Spectrum Processing (SSP) technique. In SSP, the frequency spectrum of the signal is split using overlapping Gaussian passband filters with different central frequencies and fixed absolute bandwidth. A similar approach is utilized in the WT, but in this case the relative bandwidth is constant, resulting in a filter bank with a self-adjusting window structure that can display the temporal variation of the signal's spectral components with varying resolutions. This property of the WT is extremely useful for detecting flaw echoes embedded in background noise. The detection of ultrasonic pulses using the wavelet transform is described and numerical results show good detection even for signal-to-noise ratios (SNR) of -15 dB. The improvement in detection was experimentally verified using steel samples with simulated flaws.     

Statistical pattern recognition techniques: A sample problem of ultrasonic determination of interfacial weakness in adhesive joints

A pattern recognition approach was applied to the analysis of ultrasonic echo signals from two classes of aluminum-to-aluminum adhesive bonds. The two classes differed in the surface preparation of the adherends prior to bonding, resulting in different interfacial properties of the joints. These properties have a crucial effect on the long-term adhesive properties of the specimens. Application of advanced signal processing and pattern recognition techniques enabled the classification of the joints according to the surface preparation of the adherends, based on features extracted from the ultrasonic signals. The statistics yielded an upper bound for the probability of mis-classification of the specimens. The sensitivity of certain features, extracted from the ultrasonic signal, to the interfacial characteristics of the specimens is explained by means of the natural frequencies of a joint's components and surface condition of the adherends. This leads to a method for selecting the optimal probe frequency for carrying out the ultrasonic inspection.     

Modeling for Flaw Sizing Potential with Guided Waves

Guided waves have demonstrated their value in flaw detection in a variety of different and unusual circumstances, including inspection over long distances with just 1 or 2 probes and inspection under coatings, fluids, and insulation. Via mode control and phase velocity and frequency tuning, defect detection sensitivity can be superb. The problem of going beyond detection to defect classification and sizing, however, is extremely difficult. By way of boundary element modeling, some new approaches to classification and sizing are introduced. A theoretical presentation illustrates some trends and features that might be useful in sizing analysis. Parametric studies and analysis showing amplitude versus frequency profiles for various mode input and mode conversion output via through transmission are presented. A few basic flaw shapes of different size are studied in an attempt to shed some light onto the difficult classification and sizing process.     

Spectral histogram using the minimization algorithm-theory and applications to flaw detection

In ultrasonic flaw detection in large grained materials, backscattered grain noise often masks the flaw signal. To enhance the flaw visibility, a frequency diverse statistical filtering technique known as split-spectrum processing has been developed. This technique splits the received wideband signal into an ensemble of narrowband signals exhibiting different signal-to-noise ratios (SNR). Using a minimization algorithm, SNR enhancement can be obtained at the output. The nonlinear properties of the frequency diverse statistic filter are characterized based on the spectral histogram, which is the statistical distribution of the spectral windows selected by the minimization algorithm. The theoretical analysis indicates that the spectral histogram is similar in nature to the Wiener filter transfer function. Therefore, the optimal filter frequency region can be determined adaptively based on the spectral histogram without prior knowledge of the signal and noise spectra.     

应用Hilbert变换和ZFFT提取变速器齿轮故障特征

变速器故障齿轮振动信号,调幅现象和调频现象同时存在,其频谱中包括啮合频率及其谐波、调制产生的耦合频率。Hilbert变换无法提供足够高的频率分辨率解调低频调制信号,为此提出复调制细化谱分析方法。通过变速器齿轮故障模拟实验,采集齿轮正常、轻微磨损和严重磨损时的稳态振动信号,对其进行Hilbert变换得到信号的包络,对包络信号进行复调制细化谱分析,得到齿轮轴转频基波及其谐波幅值。随着齿轮磨损程度的增加,齿轮轴转频基波及其谐波幅值明显增大,可作为齿轮磨损故障特征参数。     

Thermal aging effects on the microstructure of Nb-bearing nickel based superalloy weld overlays using ultrasound techniques

Secondary phases such as Laves and carbides are formed during the final stages of solidification of the nickel based superalloy Inconel 625 coatings deposited during the gas tungsten arc welding (GTAW) cold wire process. However, when aged at high temperatures, other phases can precipitate in the microstructure, like the γ″ and δ phases. The aim of this work was to evaluate the different phases formed during thermal aging of the as-welded material through ultrasound inspection, as well as the influence of background echo and backscattered ultrasound signals on the computational classification of the microstructures involved. The experimental conditions employed an aging temperature of 650 °C for 10, 100 and 200 h. The ultrasound signals were acquired using transducers with frequencies of 4 and 5 MHz and then processed to determine the ultrasound velocity and attenuation, as well as to study the background echo and backscattered signals produced by wave propagation. Both signal types were used to study the effectiveness and speed for classifying the secondary phases, using detrended fluctuation analysis and the Hurst method in the signal pre-processing and the Karhunen–Loève Transform in the classification of the microstructures. The ultrasound signals and the signal processing tools used were considered sufficiently sensitive, fast and accurate in the detection and classification of the microstructures in the as-welded and aged Inconel 625 alloy using this nondestructive technique.     

电视信号中标准时间频率的应用

本文介绍了电视信号中标准时间、频率的应用 ,其中包括彩色副载波标准频率综合器、电视行同步标准频率综合器、采样保持鉴相器、电视标准时间数字钟等设备的研制。藉助这些设备 ,只要直接接收中央电视台 1、2或 4套节目的全电视信号 ,便可输出常用的标准频率和时间 ,可用于精密时间测量、计量和校频 ,也可作为标准频率信号源使用 ,其校频精度达 5× 10 - 12 / 30min ,时间同步精度在 1μs以下。     

基于声发射信号EMD-WPD特征融合的航天器在轨泄漏辨识方法

长期运行在空间环境中的航天器可能由于撞击、振动、老化等因素而发生气体泄漏,在轨泄漏辨识对航天器安全保障具有重要意义。提出了一种基于声发射信号经验模态分解(empirical mode decomposition,EMD)和小波包分解(wavelet packet decomposition,WPD)特征融合的航天器泄漏辨识方法,首先将声发射信号分别通过EMD和WPD分解成为不同频率范围内的子带信号,考虑能量特征误差与不稳定性,提取信号无量纲因子和频率特征参数并应用Relief F算法选取特征。最后,构建支持向量机(support vector machines,SVM)机器学习数据库,训练泄漏分类模型并利用测试集交叉验证模型分类精度。结果表明,EMD和WPD分解特征并行融合分类模型可显著提高辨识精度,最高可达96.9%,且输入特征数量少,是一种具有应用前景的航天器在轨气体泄漏辨识方法。