Vibration-based damage detection for composite structures using wavelet transform and neural network identification

This paper presents an integrated method for damage detection of composite structures using their vibration responses, wavelet transform and artificial neural networks (ANN) identification. Structural damage feature proxy vectors are constructed and calculated based on energy variation of the structural vibration responses decomposed using wavelet package before and after the occurrence of structural damage. The ANN are applied to establish the mapping relationship between structural damage feature proxy and damage status (location and severity). The results of crack damage detection for PVC sandwich plates show that the method developed in this paper can be applied to online structural damage detection and health monitoring for various industrial structures.     

A neural network approach for identification and fault diagnosis ondynamic systems

The possibilities offered by neural networks for system identification and fault diagnosis problems in dynamic systems are investigated. In particular, an original “neural” procedure is illustrated: its sensitivity and response time enable it to be used in on-line fault diagnosis applications. Some examples are also reported. Even though they pertain to a simple linear dynamic system, these examples highlight the general applicability and advantages of a neural approach     

A two-stage neural network approach for process variance change detection and classification

Statistical process control charts (SPCC) have become one of the most commonly used tools for monitoring process variability in today's manufacturing environment. Meanwhile, neural networks have been gradually recommended as alternatives to SPCC due to their superior performances, especially in the case of monitoring process mean and unnatural patterns. Little attention has been given to the use of neural networks for monitoring the process variance. This paper describes a neural network approach to monitor process variance changes and to predict change-magnitudes. The performances of the proposed neural network monitoring scheme are compared to those of SPCC for a sample size of five and for individual observations. Simulation results show that the performance of the proposed method is comparable to that of SPCC in terms of average run lengths. In addition, the proposed neural network scheme has the capability to estimate the magnitude of the variance change by combining with a bootstrap resampling scheme. A robustness test is also applied to examine the performance of the proposed scheme for observations from a non-normal distribution.     

A compressive MUSIC spectral approach for identification of closely-spaced structural natural frequencies and post-earthquake damage detection

Motivated by practical needs to reduce data transmission payloads in wireless sensors for vibration-based monitoring of engineering structures, this paper proposes a novel approach for identifying resonant frequencies of white-noise excited structures using acceleration measurements acquired at rates significantly below the Nyquist rate. The approach adopts the deterministic co-prime sub-Nyquist sampling scheme, originally developed to facilitate telecommunication applications, to estimate the autocorrelation function of response acceleration time-histories of low-amplitude white-noise excited structures treated as realizations of a stationary stochastic process. Next, the standard multiple signal classification (MUSIC) spectral estimator is applied to the estimated autocorrelation function enabling the identification of structural natural frequencies with high resolution by simple peak picking in the frequency domain without posing any sparsity conditions to the signals. This is achieved by processing autocorrelation estimates without undertaking any (typically computationally expensive) signal reconstruction step in the time-domain, as required by various recently proposed in the literature sub-Nyquist compressive sensing-based approaches for structural health monitoring, while filtering out any broadband noise added during data acquisition. The accuracy and applicability of the proposed approach is first numerically assessed using computer-generated noise-corrupted acceleration time–history data obtained by a simulation-based framework examining white-noise excited structural systems with two closely-spaced modes of vibration carrying the same amount of energy, and a third isolated weakly excited vibrating mode. Further, damage detection potential of the developed method is numerically illustrated using a white-noise excited reinforced concrete 3-storey frame in a healthy and two damaged states caused by ground motions of increased intensity. The damage assessment relies on shifts in natural frequencies between the pre-earthquake and post-earthquake state. Overall, numerical results demonstrate that the considered approach can accurately identify structural resonances and detect structural damage associated with changes to natural frequencies as minor as 1% by sampling up to 78% below Nyquist rate for signal to noise ratio as low as 10dB. These results suggest that the adopted approach is robust and noise-immune while it can reduce data transmission requirements in acceleration wireless sensors for natural frequency identification and damage detection in engineering structures.     

Pseudospectra,MUSIC, and dynamic wavelet neural network for damage detection of highrise buildings

A non-parametric system identification-based model is presented for damage detection of highrise building structures subjected to seismic excitations using the dynamic fuzzy wavelet neural network (WNN) model developed by the authors. The model does not require complete measurements of the dynamic responses of the whole structure. A large structure is divided into a series of sub-structures around a few pre-selected floors where sensors are placed and measurements are made. The new model balances the global and local influences of the training data and incorporates the imprecision existing in the sensor data effectively, thus resulting in fast training convergence and high accuracy. A new damage evaluation method is proposed based on a power density spectrum method, called pseudospectrum. The multiple signal classification (MUSIC) method is employed to compute the pseudospectrum from the structural response time series. The methodology is validated using the data obtained for a 38-storey concrete test model. The results demonstrate the effectiveness of the WNN model together with the pseudospectrum method for damage detection of highrise buildings based on a small amount of sensed data. Copyright © 2007 John Wiley & Sons, Ltd.     

应用层组播路由的一种改进神经网络解法

研究了使用双层递归神经网络(DRNN)模型求解应用层组播路由的问题.对原有模型的神经元矩阵及能量函数进行改变,并引入了新的线性编程神经元,解决了原模型不能求解组播路由的缺陷.与启发式组播路由算法相比,该解决方案的计算复杂度低,速度较快,而且与其它的神经网络相比,由于引入了基尔霍夫限制条件,保障了解的质量,且所使用的神经元数量少,而在解的精确度上则与其它算法相当.     

A neural network for positron identification by transition radiation detector

A neural network algorithm has been applied in order to distinguish positrons from protons by a transition radiation detector (TRD). New variables are introduced, that simultaneously take into account spatial and energy TRD information. This method is found to be better than the one based on classical analysis: the results improve the detector performance in particle identification for efficiency higher than 90%. The high accuracy achieved with this method is used to identify positrons versus protons with 3 × 10⁻³ contamination, as required by TRAMP-SI cosmic ray space experiment on the NASA Balloon-Borne Magnet Facility.     

A neural network approach to microwave imaging

We present a neural network approach to microwave imaging for medical diagnosis. The problem is to reconstruct the complex permittivity of the biological tissues illuminated by the transverse magnetic (TM) incident waves. In order to avoid the inherent ill‐posedness of the inverse scattering problem, we introduce a stochastic process based on Markov random field and a priori knowledge. A coupled gradient neural network is proposed to deal with the mixed‐variable problem because the reconstructed dielectric permittivities are continuous complex variables and the line processes, which can preserve the edges of the reconstructed image, are binary variables. We report the numerical results of a simple human forearm model. We also point out the advantages and the limitations of this method. © 2001 John Wiley & Sons, Inc. Int J Imaging Syst Technol 11, 159–163, 2000     

机器人关节面参数的行波与神经网络混合辨识方法

针对柔性关节机器人,提出了运动状态下关节面参数辨识的新方法。将应用于结构中的行波分析方法与机器人关节旋转变换矩阵相结合,建立机构系统的波导方程。通过各结点力平衡及位移边界条件,得到系统振动激励,从而建立系统振动模型。利用神经网络对振动模型进行求解,将关节刚度和阻尼作为网络权值,辨识出关节动态刚度和阻尼。对3自由度机械臂进行辨识实验研究,实验结果表明该辨识方法是可行的、有效的。     

A neural network approach for batching decisions in wafer fabrication

This paper considers a stochastic batching problem for a batch process in wafer fabrication, where various numbers of wafer lots are allowed to process together in each batch, and wafer lots arrive randomly but not in a specified pattern. The objective is to determine the optimal size (number of wafer lots) of each batch with respect to the measure of minimizing the mean queueing time of wafer lots. For the problem, a multi-layer perceptron neural network model is proposed to make real-time batching control, and its effectiveness is investigated in comparison with that of the well-known minimum batch size policy.     

一种基于压电晶片阵列的板结构损伤识别方法

提出一种通过单个压电晶片对板结构进行激励,使用十字形压电晶片阵列接收响应信号,然后基于相控阵原理对信号进行处理,从而对板结构损伤进行识别的方法。首先用谱元法分析了在单点激励下铝板结构有损伤和无损伤时Lamb波的传播情况,得到压电晶片阵列中各阵元所在位置的响应信号。根据压电晶片阵列布置形式及各阵元之间的相互位置,对各阵元接收到的时域信号进行处理,从而对结构的损伤进行识别。通过对板结构中含有一个和两个损伤的情况进行了研究,验证了该方法的有效性。     

A point-by-point impact vibration method for damage detection in beam-type structures

This paper develops the point-by-point impact vibration method, a damage diagnosis method for beam structures. The theoretical development and experimental verification were carried out on aluminum beams with single and double cracks in the vertical direction and on pre-stressed concrete beams with local breaks. The processing signals were the time histories of the acceleration responses of the beams and the impact force. The point-by-point impact vibration method is combined with the Dynamic Reciprocal Theorem, the Frequency Domain Decomposition method, and the Damage Index idea, all of which cover the modal curvature change (MCC) and the change of strain energy, as well as other measures. One of the advantages of this method is the ability to use fewer sensors to obtain higher order eigenmodes, which can be applied in the damage index. Experimental results indicate that higher order vibration modes exhibit great potential for damage identification.     

A damage identification method for truss structures using a flexibility-based damage probability index and differential evolution algorithm

An efficient method for the multiple damage detection of truss systems using a flexibility-based damage probability index (FBDPI) and differential evolution algorithm (DEA) is proposed. In the first step, a new FBDPI is introduced to find the potentially damaged elements of truss systems. The proposed FBDPI is based on the changes of elemental strain, due to damage, computed by the flexibility matrix of the structure. The flexibility matrix of the structure is dynamically estimated using modal analysis data. In the second step, the reduced damage problem is transformed into a standard optimization problem having few damage variables. Then, the DEA is employed to solve the optimization problem for determining the actual location and severity of damaged elements. Simulation results considering measurement noise demonstrate the high efficiency of the proposed method for the damage detection of truss structures.     

A neural network approach for safety and collision avoidance in robotic systems

A major factor which has limited the application of robots in industrial and human service applications has been the lack of robust sensing and control algorithms for detection and prevention of collision conditions. This paper discusses an approach to the collision avoidance control of robots using a neural network methodology for the integration of sensory input data from the robot's environment. The paper presents a formulation of the collision avoidance problem using the occupancy grid formulation, and discusses the use of a combination of Dempster-Shafer inference and neural networks in fusing the sensory information and making robot movement decisions. Initial studies have shown this approach to be both robust and computationally tractable in providing enhanced safety capabilities.     

引入信息熵理论的砼结构损伤动力识别新思路

对基于振动特性的混凝土结构损伤识别方法进行综述,提炼出损伤识别指标的敏感性不足、非线性分析困难、适用范围有限、实验验证方法单一、数值模拟不合理等不足。进而在理论和技术的可行性分析基础上,将耗散结构和信息熵理论系统引入到混凝土结构损伤识别领域中,通过构造信息熵指标来反映系统损伤的无序程度,在传感器的布点优化、损伤位置的确定、损伤程度的量化和处理结构损伤上显示出较大的潜力,有望利用信息熵理论来部分解决目前存在的混凝土结构损伤领域的难点,为混凝土结构健康监测系统的建设和运行提供科学依据和技术支撑     

A technique for rapid impact damage detection with implication for composite aircraft structures

A technique using a simple optical arrangement (called D-Sight) involving a source of light and a retroreflective screen was patented by Diffracto Ltd. The method was used to locate indentations associated with low energy impact damage. In graphite/epoxy specimens good correlation was observed between internal impact damage as shown on ultrasonic C-scan images and indentations detected with the D-Sight method. Test specimens are currently mounted on an aircraft to observe the influence of in-service surface degradation on technique resolution. The method has the potential for inexpensive, rapid and objective detection of low energy impact damage over large areas of composite aircraft structure. Application of this method, because of its inherent reliability, could result in the increasing of the design allowable strain levels for some composite components.     

面向目标检测的卷积神经网络优化方法

针对星载等功耗受限平台下遥感影像目标检测存在的高准确率、低功耗以及高吞吐量等要求,本文提出了一种面向目标检测的现场可编程门阵列(FPGA)卷积神经网络(CNN)优化方法.采用数据流调度技术以及基于乘法矩阵与前向加法链的卷积计算阵列设计对浮点卷积神经网络模型进行加速.利用该方法在FPGA开发板上实现了浮点卷积目标检测网络...