Investigation of deep neural network with drop out for ultrasonic flaw classification in weldments

Ultrasonic signal classification of defects in weldment, in automatic fashion, is an active area of research and many pattern recognition approaches have been developed to classify ultrasonic signals correctly. However, most of the developed algorithms depend on some statistical or signal processing techniques to extract the suitable features for them. In this work, data driven approaches are used to train the neural network for defect classification without extracting any feature from ultrasonic signals. Firstly, the performance of single hidden layer neural network was evaluated as almost all the prior works have applied it for classification then its performance was compared with deep neural network with drop out regularization. The results demonstrate that given deep neural network architecture is more robust and the network can classify defects with high accuracy without extracting any feature from ultrasonic signals.     

Identification of faults through wavelet transform vis-à-vis fast Fourier transform of noisy vibration signals emanated from defective rolling element bearings

Fault diagnosis of rolling element bearings requires efficient signal processing techniques. For this purpose, the performances of envelope detection with fast Fourier transform (FFT) and continuous wavelet transform (CWT) of vibration signals produced from a bearing with defects on inner race and rolling element, have been examined at low signal to noise ratio. Both simulated and experimental signals from identical bearings have been considered for the purpose of analysis. The bearings have been modeled as spring-mass-dashpot systems and the simulated signals have been obtained considering transfer functions for the bearing systems subjected to impulsive loads due to the defects. Frequency B spline wavelets have been applied for CWT and a discussion on wavelet selection has been presented for better effectiveness. Results show that use of CWT with the proposed wavelets overcomes the short coming of FFT while processing a noisy vibration signals for defect detection of bearings.     

涡流检测的数值模拟与缺陷定量分析

将数值模拟与信号分析技术用于涡流检测缺陷的定量分析中。针对差动线圈,检测管子上不同埋藏深度缺陷的模型进行数值计算,得到对应的阻抗平面图。为了识别不同阻抗平面图所对应的缺陷,利用傅立叶变换方法进行了特征量提取,得到了不同的描述符;设计人工神经网络对不同的特征量进行分析识别,变换的傅立叶描述符可用作缺陷定量的有效特征量。     

Maximum kurtosis principle for the parameter selection of Gabor wavelet and its application to ultrasonic signal processing

The noise suppression techniques with wavelet transform (WT) are widely used in nondestructive testing and evaluation (NDT&E), especially in ultrasonics. But the wavelet based filter has the property of equal Q-factor, so, it is impossible to choose the central frequency and the bandwidth arbitrarily at the same time. This paper develops a new technique using WT to eliminate this drawback. In this paper, a weak ultrasonic signals identification method by using the optimal parameter Gabor wavelet transform is proposed. We address the choice of the optimal central frequency and bandwidth of the Gabor wavelet using the kurtosis maximization algorithm. The central frequency and bandwidth of the optimal parameter Gabor wavelet matched that of the ultrasonic signal very well. Numerical and experimental results have been presented to evaluate the effectiveness of the optimal parameter Gabor wavelet transform on ultrasonic flaw detection. This technique is a simpler and effective technique for processing heavy noised ultrasonic signals.     

HHT在Lamb波检测信号分析中的应用

将一种新的超声信号处理技术用于Lamb波波形中多个模式到达时间的提取。通过将希尔伯特-黄变换(Hilbert-Huang transform,简称HHT)与快速傅里叶变换(fast Fourier transform,简称FFT)、小波变换(wavelettransform,简称WT)在时频分辨率方面的比较,表明HHT能够精确识别信号中两种频率分量突变的时刻,显示了HHT方法的优越性。将HHT方法的特性用于Lamb波模式到达时间的提取,从HHT的能量-时间图上可以看出,能量峰值时刻对应着各Lamb波模式的到达时间。试验结果与理论值具有较好的一致性。     

脉冲涡流信号检测与分析

为了对较厚钢板进行缺陷检测,采用对低压大功率的电源进行斩波的方式产生大电流脉冲,设计圆柱形探头,采用激励线圈缓慢放电的方式检测钢板表面缺陷;设计矩形探头,采用激励线圈瞬间放电的方式检测钢板亚表面;对脉冲涡流瞬态感应电压信号进行小波滤噪和幅值归一化处理。实验结果表明,小波滤噪可以提高检测精度,对信号幅值进行归一化处理可以更直接地反映不同深度缺陷产生的曲线间的相互关系,大电流激励的脉冲涡流信号幅值能反映较厚钢板表面和亚表面缺陷深度。     

An adaptive Morlet wavelet filter for time-of-flight estimation in ultrasonic damage assessment

Ultrasonic non-destructive testing has been widely used in assessing the integrity of engineering materials such as high-temperature alloys and structures such as pipelines, bridges, and other load-bearing structures. The ultrasonic signals received from these structures are often noisy. Effective noise-reduction techniques are needed in order to accurately assess their condition. This paper presents a new digital signal processing method for estimating ultrasonic time-of-flight diffraction (TOFD). This method is based on wavelet analysis using the Morlet wavelet and the least mean squares (LMS) adaptive filter. The adaptive line enhancer (ALE) structure is used for the adaptive filter. The filter is designed to remove noise and identify the point at which the ultrasonic signal starts to reflect an echo from the tip of a crack. Both simulated and experimental data obtained from a steel plate with a crack produced by electrical-discharge-machining (EDM) are used to demonstrate the performance of the proposed method. This method is especially useful when the properties of the crack signal are unknown and the signal-to-noise ratio is low.     

基于PCI总线的超声检测虚拟仪器系统设计

为了实现超声检测的数字化、智能化、图像化以及自动化,开发了一种基于PCI总线的超声检测虚拟仪器系统.系统以计算机为核心,利用FPGA强大逻辑处理能力,PCI总线高速数据传输功能,实现了超声检测信号的发射接收,采集处理,数据存储、显示和输出,并运用小波变换对超声信号进行了降噪处理.实验结果表明,小波变换对超声信号具有良好的滤波效果,该系统具有数据传输速度快,信噪比高、性价比高等优点,为缺陷信号的准确定量和定性分析奠定了良好的基础.     

基于经验与变分混合分解的超声回波信号噪声消除方法

超声缺陷检测结果易受超声回波信号中复杂噪声的干扰,为了提高超声缺陷检测的准确度,提出一种基于混合分解的 超声回波信号噪声消除方法。 采用经验模态分解算法结合相关系数指标对超声回波信号进行预处理,得到消除低频噪声分量 的超声回波预处理信号。 基于变分模态分解将该预处理信号分解为一系列窄带本征模态函数,引入互信息指标估计变分模态 分解的最优模态数量,并根据窄带本征模态函数与预处理信号的相关系数提取有用的模态分量,实现对超声回波信号去噪结果 的重构。 通过仿真和实测超声回波信号验证了本文方法的去噪性能,并与现有方法进行了对比。 结果表明,本文方法可同时消 除超声回波信号中的高频和低频噪声,在不同信噪比条件下 EMD、VMD 和本文方法去噪结果的 SNR 均值分别为 10. 01、9. 48 和 16. 09 dB,验证了本文方法对于超声回波信号噪声消除的优越性。     

小波变换在机械系统非线性信号消噪中的应用

噪声的去除一直是信号处理中较为关键的技术之一。小波变换在时、频两域都具有表征信号局部特征的能力，突破了传统Fourier分析的局限性，很适合检测信号的奇异现像。用Daubechies小波和Fourier变换分别对洛仑兹混沌信号以及撞击流反应器压力波动非线性信号进行去噪分析，结果表明，二者的去噪效果有较大的不同，突出了小波变换用于非线性信号去噪的性能。因而，将小波用于信号消噪具有重要的意义。     

Effects of the number of pulse repetitions and noise on the velocity data from the ultrasonic pulsed Doppler method with different algorithms

The ultrasonic pulsed Doppler technique known as the ultrasonic velocity profile (UVP) method has been widely used in many engineering fields. The analysis algorithms of the UVP, the number of pulse repetitions (N_pulse), noise and reflector conditions, etc. all affect the measurement accuracy. N_pulse is related to the temporal resolution, thus to improve this resolution it must be set as low as possible. However, it is known that the measurement accuracy of the instantaneous velocity becomes worse with decreasing values of N_pulse. In this study, UVP analysis algorithms including the fast Fourier transform (FFT), autocorrelation, and the wavelet transform (WT) were compared via simulations and experiments using varying values of N_pulse and the signal-to-noise ratio (SNR). We show that there is an appropriate N_pulse for each algorithm that depends on the SNR; specifically, the value of N_pulse increases with decreasing SNR. The difference between the algorithms for the velocity data was small under low noise conditions. However, a FFT with a Gaussian interpolation produced the best result under noisy conditions. In contrast the WT was relatively unaffected by noise. Therefore, a WT is the preferred choice for measuring velocity distributions if high sampling measurement is not required.     

薄镀锌钢板点焊超声成像分析

采用水浸超声聚焦直入射法对镀锌薄板点焊焊核进行超声C扫描成像检测,分析C扫描图像中各特征部分的超声A扫描信号。超声波C扫描成像方法能够测量点焊焊核直径,检测焊核中常见的几种缺陷如气孔、裂纹以及飞溅,此外采用该方法还能对焊核的外观进行检测。为了获得更为精确的焊核内部结构信息,进一步分析了焊核内部不同形状缺陷的C扫描图像特征及A扫描信号特征,通过这些图像特征和信号特征能够判断焊核内部缺陷的形状及类型。因此,超声C扫描成像法能够在不破坏点焊焊核的前提下全面直观地显示焊核的内外部结构。该方法不仅能够用于评价点焊质量,而且还可以作为辅助手段用于点焊的其他研究中。     

Surface defect detection in tiling Industries using digital image processing methods: Analysis and evaluation

Ceramic and tile industries should indispensably include a grading stage to quantify the quality of products. Actually, human control systems are often used for grading purposes. An automatic grading system is essential to enhance the quality control and marketing of the products. Since there generally exist six different types of defects originating from various stages of tile manufacturing lines with distinct textures and morphologies, many image processing techniques have been proposed for defect detection. In this paper, a survey has been made on the pattern recognition and image processing algorithms which have been used to detect surface defects. Each method appears to be limited for detecting some subgroup of defects. The detection techniques may be divided into three main groups: statistical pattern recognition, feature vector extraction and texture/image classification. The methods such as wavelet transform, filtering, morphology and contourlet transform are more effective for pre-processing tasks. Others including statistical methods, neural networks and model-based algorithms can be applied to extract the surface defects. Although, statistical methods are often appropriate for identification of large defects such as Spots, but techniques such as wavelet processing provide an acceptable response for detection of small defects such as Pinhole. A thorough survey is made in this paper on the existing algorithms in each subgroup. Also, the evaluation parameters are discussed including supervised and unsupervised parameters. Using various performance parameters, different defect detection algorithms are compared and evaluated.     

Wire bonding quality monitoring via refining process of electrical signal from ultrasonic generator

In this paper, a technique for on-line quality detection of ultrasonic wire bonding is developed. The electrical signals from the ultrasonic generator supply, namely, voltage and current, are picked up by a measuring circuit and transformed into digital signals by a data acquisition system. A new feature extraction method is presented to characterize the transient property of the electrical signals and further evaluate the bond quality. The method includes three steps. First, the captured voltage and current are filtered by digital bandpass filter banks to obtain the corresponding subband signals such as fundamental signal, second harmonic, and third harmonic. Second, each subband envelope is obtained using the Hilbert transform for further feature extraction. Third, the subband envelopes are, respectively, separated into three phases, namely, envelope rising, stable, and damping phases, to extract the tiny waveform changes. The different waveform features are extracted from each phase of these subband envelopes. The principal components analysis (PCA) method is used for the feature selection in order to remove the relevant information and reduce the dimension of original feature variables. Using the selected features as inputs, an artificial neural network (ANN) is constructed to identify the complex bond fault pattern. By analyzing experimental data with the proposed feature extraction method and neural network, the results demonstrate the advantages of the proposed feature extraction method and the constructed artificial neural network in detecting and identifying bond quality.     

粗集理论在污水参数软测量中的应用研究

用粗糙集理论约简属性,消除冗余信息后建立了污泥体积指数的神经网络软测量模型.用某城市污水厂实际水质参数进行仿真实验.仿真结果表明,与未采用粗糙集进行预处理的模型相比,应用该模型不仅测量值的误差更小,而且输入参数从9个降至4个,大大降低了输入数据的维数,减少了神经网络的训练时间及训练步数,有利于软测量模型的实用化.     

超声检测信号时频邻域自适应消噪技术

分析超声检测回波信号中噪声的组成和特性.考虑缺陷信号和噪声分量的分布差异,提出一种基于小波包变换时频邻域统计特征的自适应消噪方法.该方法根据噪声水平和邻域数据的方差,自动调节邻域数据对中心点值的平滑处理强度,从而达到自适应消噪目的.由于不存在参考信号和参数选择的问题,该方法稳健性好.仿真和实测信号的试验结果表明:该方法能有效提高信号的信噪比以及不同类型缺陷信号之间的可区分性,并抑制波形失真和信号的能量衰减.     

Development of a system for monitoring tool condition using acousto-optic emission signal in face turning—an experimental approach

In automated manufacturing systems, one of the most important issues is accurate detection of the tool conditions under given cutting conditions so that worn tools can be identified and replaced in time. In metal cutting as a result of the cutting motion, the surface of workpiece will be influenced by cutting parameters, cutting force, and vibrations, etc. But the effects of vibrations have been paid less attention. In the present paper, an investigation is presented of a tool condition monitoring system, which consists of a fast Fourier transform preprocessor for generating features from an online acousto-optic emission (AOE) signals to develop a database for appropriate decisions. A fast Fourier transform (FFT) can decompose AOE signals into different frequency bands in the time domain. Present work uses a laser Doppler vibrometer for online data acquisition and a high-speed FFT analyser used to process the AOE signals. The generation of the AOE signals directly in the cutting zone makes them very sensitive to changes in the cutting process due to vibrations. AOE techniques is a relatively recent entry into the field of tool condition monitoring. This method has also been widely used in the field of metal cutting to detect process changes like displacement due to vibration and tool wear, etc. In this research work the results obtained from the analysis of acousto-optic emission sensor employs to predict flank wear in turning of AISI 1040 steel of 150 BHN hardness using Carbide insert and HSS tools. The correlation between the tool wear and AOE parameters is analyzed using the experimental study conducted in 16 H.P. all geared lathe. The encouraging results of the work pave the way for the development of a real-time, low-cost, and reliable tool condition monitoring system. A high degree of correlation is established between the results of the AOE signal and experimental results in identification of tool wear state.     

滤波截止频率对粗糙表面接触机理的影响

使用基于快速傅里叶变换和共轭梯度法的扩展干接触程序,以及基于Greenwood和Williamson接触模型(G-W模型)的接触求解方法,分析滤波截止频率对粗糙表面接触机理的影响。结果表明,粗糙表面形貌数据再处理时使用的滤波器滤波截止频率,对力－接近量关系有着明显的影响;G-W 接触模型仅适用于波长较大的粗糙表面。因此,在G-W模型的各种应用中,应注意滤波截止频率对粗糙表面接触机理的影响。     

PARAMETER ESTIMATION OF MULTICOMPONENT TRANSIENT SIGNALS USING DECONVOLUTION AND ARMA MODELLING TECHNIQUES

Parameter estimation of transient signals, having real decaying exponential constants, is a difficult but important problem that often arises in many areas of scientific disciplines. The frequency domain method of analysis that involves Gardner transformation and conventional inverse filtering often degrades the quality of the deconvolved data, leading to inaccurate results, especially for noisy data. An improved method that is based on the combination of Gardner transformation, optimal compensation deconvolution, and signal modelling techniques is suggested in this paper. In this method of analysis the exponential signal is converted to a convolution model whose input is a train of weighted delta function that contains the signal parameters to be determined. The resolution of the estimated decay rates is poor if the conventional fast Fourier transform (FFT) algorithm is used to analyse the resulting deconvolved data. Using an autoregressive moving (ARMA) model whose AR parameters are determined by solving high-order Yule–Walker equations (HOYWE) via the singular value decomposition (SVD) algorithm can alleviate this shortcoming. The effect of sampling conditions, noise level, number of components and relative sizes of the signal parameters on the performance of this modified method of analysis is examined in this paper. Simulation results show that high-resolution estimates of decay constants can be obtained when the above signal processing techniques are used to analyse multiexponential signals with varied signal-to-noise ratio (SNR). This approach also provides a graphical procedure for detecting and validating the number of exponential signals present in the data. Some computer simulation results are presented to justify the need for this modified method of analysis.     

电触头钎焊接头超声检测信号的集成神经网络分类

研究电触头钎焊接头超声无损检测中的缺陷分类问题,提出了一种新的集成神经网络分类方法。该方法分四步：频率不变性预处理,多分辩分析,特征量预处理,集成 Ｂ Ｐ神经网络分类。使用不同中心频率探头检测得到的缺陷信号首先通过预处理变换到一个等效的参考频率上,然后利用离散小波变换提取特征量。特征量被预处理后,输入到集成 Ｂ Ｐ神经网络分类器中分类。本文用２１３ 个超声检测信号测试了集成神经网络的性能。实验结果表明了频率不变性技术和集成 Ｂ Ｐ神经网络分类技术的有效性。