Adaptive noise filtering using an error-backpropagation neuralnetwork

A neural network of the feedforward-error backpropagation type proposed by D.E. Rumelhart et al. (1986) was applied to filter noise from spectral data commonly encountered in infrared absorption of molecular transitions. The purpose was to gain insight into the way a neural network can be trained to remove noise from a noise-corrupted signal with implications for signal processing in general. The neural network simulation was implemented in Fortran and run on a VAX 8800. Training of the neural network occurred on a set of spectral data with random transitions and line shape parameters. Preliminary results of the performance of the adopted neural network are reported and discussed along with observed limitations. Future improvements on noise filtering using a neural network are proposed     

测量与信息技术

测量是获取信息的重要手段。科学研究、生产、高科技发展都离不开测量。测量技术发展的重要特点是它与信息技术的融合,这是提高测量精度、在线和动态测量、复杂参数和复杂环境下测量、智能测量的需要。这一融合贯彻始终,它包括信号调制与解调、采样与重构、信号融合、数据压缩、滤波、信号变换、时间序列分析、谱分析、数据拟合与建模、模式识别、神经网络、仿真与虚拟、误差分离、误差补偿、冗余技术、决策与智能技术等,它们互相支持。章通过三坐标测量机误差补偿、圆度和轴系误差测量,以及大型工程的柔性坐标测量系统,介绍其在提高测量精度、优化、自标定、丢失信息自恢复、系统重组等方面的作用及其关键技术。     

Analysis of one-dimensional stochastic finite elements using neural networks

This paper explores the applicability of neural networks for analyzing the uncertainty spread of structural responses under the presence of one-dimensional random fields. Specifically, the neural network is intended to be a partial surrogate of the structural model needed in a Monte Carlo simulation, due to its associative memory properties. The network is trained with some pairs of input and output data obtained by some Monte Carlo simulations and then used in substitution of the finite element solver. In order to minimize the size of the networks, and hence the number of training pairs, the Karhunen–Loéve decomposition is applied as an optimal feature extraction tool. The Monte Carlo samples for training and validation are also generated using this decomposition. The Nyström technique is employed for the numerical solution of the Fredholm integral equation. The radial basis function (RBF) network was selected as the neural device for learning the input/output relationship due to its high accuracy and fast training speed. The analysis shows that this approach constitutes a promising method for stochastic finite element analysis inasmuch as the error with respect to the Monte Carlo simulation is negligible.     

Errors analysis of transponder-based radiometric calibration for high-resolution imaging radars

Calibration is often required for precision measurement instruments like synthetic aperture radar (SAR) such that an image pixel intensity is directly expressed in terms of the surface backscatter coefficient. An active reflector named as transponder is an effective SAR calibration device which yields a larger radar cross-section than a corner reflector. Likewise, encoding the retransmitted signal can decouple the transponder from the background scatterers, hence, it can be used to calibrate SAR for radiometry, polarimetry and interferometry. However, as this transponder acts as an amplitude modulator, any signal errors in the transponder will result in degraded performance of radiation calibration. Literature search reveals that little work has been reported. As such, various kinds of signal errors and their effects on radiometric calibration are investigated. The system configuration and corresponding signal processing algorithms are detailed. Starting with an investigation of different errors, closed analytic expressions for calibration performance are derived. Furthermore, the most essential element of this transponder, one X-band wideband voltage-controlled attenuator, is developed. Experimental results show that, because of its low cost and wideband modulation technique, this transponder has extensive applications, not just for radiometric calibration.     

基于多输入层卷积神经网络的滚动轴承故障诊断模型

针对滚动轴承信号易受噪声干扰和智能诊断模型鲁棒性差的问题,在一维卷积网络的基础上,提出基于多输入层卷积神经网络的滚动轴承故障诊断模型。相比传统卷积神经网络诊断模型,该模型具有多个输入层,初始输入层为原始信号,以最大化地发挥卷积网络自动学习原始信号特征的优势;同时可将谱分析数据在模型任意位置输入模型,以提升模型的识别精度和抗干扰能力。通过滚动轴承模拟试验,进行可行性和有效性验证,同时与人工神经网络(Artificial Neural Network,ANN)、支持向量机(Support Vector Machine,SVM)和典型的卷积神经模型进行对比,证明了所提出模型的优势;向测试集中加入噪声来检验模型的鲁棒性,并且运用增量学习方法提升模型在强噪声环境下的识别性能;通过滚动轴承故障实例,验证模型的识别性能和泛化能力。试验结果表明,所提出的模型提升了传统卷积模型的识别率和收敛性能,并具有较好的鲁棒性和泛化能力。     

Prediction of Impact Contact Forces of Composite Plates Using Fiber Optic Sensors and Neural Networks

Real-time determination of contact forces due to impact on composite plates is necessary for on-line impact damage detection and identification. We demonstrate the use of fiber optic strain sensor data as inputs to a neural network to obtain contact force history. An experimental and theoretical study is conducted to determine the in-plane strains of a clamped graphite/epoxy composite plate upon low-velocity impacts using surface-mounted extrinsic Fabry-Perot interferometric (EFPI) strain sensors. The plate is impacted with a semispherical impactor with various impact energies using the drop-weight technique. The significant features of the strain and contact force response are contact duration, peak strain, and strain rise time. We have designed and built an instrumented drop-weight impact tower to facilitate the measurement of contact force during an impact event. The impact head assembly incorporates a load cell to measure the contact forces experimentally. An in-house finite-element program is used to establish the validity of the EFPI fiber optic sensor contact force response. The finite-element model is based on a higher-order shear deformation theory and accounts for geometric nonlinearity. Experimental load cell data and finite-element impact-induced contact force responses are in close agreement. The load cell data is used to train a three-layer feed-forward neural network which utilizes the Delta Bar Delta back-propagation algorithm. The output of the neural network simulation is the contact force history and the inputs are fiber optic sensor data in two different locations and time in 10-ms intervals. The efficiency and accuracy of the neural network method is discussed. The neural network scheme recovers the impact contact forces without using any complex signal processing techniques.     

Neural-network-based method of calibration and measurandreconstruction for a high-pressure measuring system

The problem of applying the neural networks for static calibration of measuring systems and for measurand reconstruction is addressed. A multilayered neural network based method for the static calibration of this system is proposed. The functioning of the calibrated measuring system is based on three fiber-optic transducers whose static characteristics are nonmonotonic and significantly influenced by temperature. The applicability of the proposed calibration method is demonstrated in the case under consideration using synthetic and real data. The neural network is designed and implemented in a general purpose microcontroller. In comparison with the spline-based method of calibration, for the same reference data, the proposed method allows obtention of a better quality of calibration and, most important, when calibrated, the multilayered neural network does not require the measurement of temperature for pressure reconstruction     

用神经网络提高SO2传感器的检测速度

制作了新型IDC结构聚苯胺膜SO2传感器,测得不同SO2浓度的电导响应与恢复时间曲线。从这些气敏特性曲线出发,以电导响应曲线的斜率作为网络的输入,对应的SO2浓度作为输出,建立了BP网络预测推理模式,对四组数据的预测结果表明精度较高（误差小于3％）,具有很好的预测能力。这种方法不同于传统的标定方法,后者需要4min才能稳定的响应,神经网络样本检测不需要达到稳定的响应就可以预测SO2的浓度,从而大大缩短了结果的响应时间（缩短了75％）。     

Neural network prediction in a system for optimizing simulations

Neural networks have been widely used for both prediction and classification. Back-propagation is commonly used for training neural networks, although the limitations associated with this technique are well documented. Global search techniques such as simulated annealing, genetic algorithms and tabu search have also been used for this purpose. The developers of these training methods, however, have focused on accuracy rather than training speed in order to assess the merit of new proposals. While speed is not important in settings where training can be done off-line, the situation changes when the neural network must be trained and used on-line. This is the situation when a neural network is used in the context of optimizing a simulation. In this paper, we describe a training procedure capable of achieving a sufficient accuracy level within a limited training time. The procedure is first compared with results from the literature. We then use data from the simulation of a jobshop to compare the performance of the proposed method with several training variants from a commercial package.     

重载并联六维力传感器及静态标定

针对传感器重载小尺寸需求,提出一种具有混合分支的重载并联六维力传感器,分析了其结构特点和测量原理。搭建了重载并联六维力传感器标定系统,为改善维间耦合及制造误差等对测量精度产生的影响,从标定算法及模型优化方面对其进行了研究。分别利用最小二乘法和BP神经网络算法对加载实验数据进行了处理,分析结果表明BP神经网络算法要明显优于最小二乘法,并通过数据随机分组测试验证了结果的正确性。基于BP神经网络,提出了一种基于人工鱼群算法的BP神经网络算法,并采用优化后的BP神经网络标定算法对实验数据进行了计算分析,结果表明优化后的BP神经网络计算结果较好且稳定,不易陷入局部极值。     

Estimation of composite damage model parameters using spectral finite element and neural network

A multi-layer perceptron (MLP) network using error back propagation algorithm is employed in this paper to estimate the damage parameters from broad-band spectral data as diagnostic signal. Various existing models of damage in laminated composite and the resulting stiffness degradation are discussed from comparative view-point. Degradation of ply properties can be considered to be one of the damage model parameters while monitoring transverse matrix cracks in cross-ply, splitting in longitudinal ply, and evolution of consecutive stages of damage, such as delaminations and fiber fracture. The stiffness degradation factor, the location and size of the damaged zone in laminated composite beam are considered as damage model parameters in the present paper. Fourier spectral data, which is typical to most of the diagnostic wave measurements, are used as input to the neural network. Since, training the neural network in such case involves many data sets and all of these data are difficult to generate using experiments, a spectral finite element model (SFEM) with embedded degraded zone in laminated composite beam is developed. Numerical simulation using this element is carried out, which shows the nature of temporal signal that are likely to be measured. Analytical studies on the performance of the neural network are presented based on numerically simulated data. Effect of measurement noise on the network performance is also reported.     

几种超声NDE信号降噪方法的性能比较

为提高超声无损检测的准确性,需要对超声NDE信号中因随机分布于媒质中的大量散射微粒所引起的结构噪声进行降噪。由于信号和噪声的频谱范围基本重叠,传统的线性滤波方法不能提供理想的降噪结果。介绍了几种对超声NDE信号进行降噪的新方法:Wigner-Ville分布法、小波变换法和基于时间延迟的神经网络法,并从信噪比(SNR)、检测概率(PD)和估测深度(ED)等三个重要参数对它们的降噪性能进行计算机仿真实验的比较。结果表明:小波变换法和神经网络法的降噪效果较Wigner-Ville分布法要好。对实际信号的测试还表明,小波变换由于不像神经网络那样需要训练,是一种更为理想的超声NDE信号降噪方法。     

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.     

Development of a real-time multi-objective scheduler for a semiconductor fabrication system

Semiconductor wafer fabrication involves possibly one of the most complex manufacturing processes ever used. This causes a number of decision problems. A successful system control strategy would assign appropriate decision rules for decision variables. Therefore, the goal of this study is to develop a scheduler for the selection of decision rules for decision variables in order to obtain the desired performance measures given by a user at the end of a certain production interval. In this proposed methodology, a system control strategy based on a simulation technique and a competitive neural network is suggested. A simulation experiment was conducted to collect the data containing the relationship between the change of decision rule set and current system status and the performance measures in the dynamic nature of semiconductor manufacturing fabrication. Then, a competitive neural network was applied to obtain the scheduling knowledge from the collected data. The results of the study indicate that applying this methodology to obtaining a control strategy is an effective method considering the complexity of semiconductor wafer fabrication systems.     

Practical Risk Assessment for Project Management

We study the phase I analysis of data when the quality of a process or product is characterized by a linear function. We assume that simple linear regression data are available for a fixed number of samples collected over time, a situation common in calibration applications. Using a simulation study, we compare the performance of some of the recommended approaches used to assess the stability of the process. We also propose a method based on using indicator variables in a multiple regression model. We show that this method has competitive performance relative to other methods in terms of the probability of a signal for certain types of parameter shifts within the set of historical process data. We also show that two methods proposed in the literature are ineffective in detecting shifts in the process parameters. Finally, we apply some of the proposed methods to a calibration example.     

基于RBF神经网络的色空间转换模型

研究了RBF神经网络的结构及算法,应用RBF神经网络建立了打印机的色空间转换模型.根据实验数据,对网络结构进行了优化,通过比较不同参数时网络的性能,确定最优网络参数.最后对所建模型进行了仿真验证,验证结果表明,预测数据与实测数据的色差较小,说明该模型具有实用价值.     

New absolute vector error correction technique for a transmitter/receiver module

The authors demonstrate how a transmitter (Tx), a reciprocal transmitter/receiver (Tx/Rx) signal path and two unidirectional receiver (Rx) paths can be used together with short, open, and load standards for the absolute vector error correction (AVEC) of a Tx/Rx module. Once calibrated, this Tx/Rx module can then provide accurate vector measurements of the signals that are flowing into and/or out of the test port. This novel AVEC technique is one of the key concepts in the design of a wideband absolute vector signal measurement system, which overcomes the limitations of traditional measurement instruments by combining the features of vector signal analysers, spectrum analysers, and vector network analysers. The AVEC method is validated using numerical simulation data for a simplified baseband test circuit. The AVEC technique is then extended to the calibration of wideband, high-frequency Tx/Rx modules that involve frequency up/down conversion mixers in a follow-on paper     

Calibration modeling for mobile vision based laser imaging and approximation networks

A mobile calibration technique for three-dimensional vision is presented. In this technique, the parameters of mobile vision are computed automatically by approximation networks and image processing of a laser line. Also, the three-dimensional vision is performed by the networks. The proposed vision system provides online geometric modifications from the initial configuration. Here, an online re-calibration is performed via network data to determine the setup modifications. Thus, vision limitations caused by the geometric modifications are overcome. The network is built based on the behavior of the laser line and the camera position. In this manner, the network provides the data for the automatic re-calibration. This system avoids calibrated references and external procedures for re-calibration. Therefore, the calibration model improves the accuracy and performance of the mobile vision. It is because the external data are not passed to the procedure of the three-dimensional vision. This kind of modeling represents a contribution for calibration of mobile vision. To elucidate this contribution an evaluation is carried out based on the calibration via perspective projection model. The time processing is also described.     

A learning algorithm for self-calibration of a voltage calibrator

An algorithm either to extend the calibration period or to reduce the measurement uncertainty of a DC voltage reference module is presented. This module is used either as a transfer, independent, or working standard, or as a reference module incorporated into a larger measuring system. The basic idea is that the deviation history of measured voltage differences of reference elements of a group reference module during the calibration period can be used as a learning period for a neural network. This neural network, when created, can numerically correct particular reference elements later in the working period. Results were obtained by simulation and evaluated on the basis of empirical data and simulated input functions. Hardware solutions to model this algorithm are discussed     

Calibration of multiplexed fiber-optic spectroscopy

Large-scale commercial bioprocesses that manufacture biopharmaceutical products such as monoclonal antibodies generally involve multiple bioreactors operated in parallel. Spectra recorded during in situ monitoring of multiple bioreactors by multiplexed fiber-optic spectroscopies contain not only spectral information of the chemical constituents but also contributions resulting from differences in the optical properties of the probes. Spectra with variations induced by probe differences cannot be efficiently modeled by the commonly used multivariate linear calibration models or effectively removed by popular empirical preprocessing methods. In this study, for the first time, a calibration model is proposed for the analysis of complex spectral data sets arising from multiplexed probes. In the proposed calibration model, the spectral variations introduced by probe differences are explicitly modeled by introducing a multiplicative parameter for each optical probe, and then their detrimental effects are effectively mitigated through a "dual calibration" strategy. The performance of the proposed multiplex calibration model has been tested on two multiplexed spectral data sets (i.e., MIR data of ternary mixtures and NIR data of bioprocesses). Experimental results suggest that the proposed calibration model can effectively mitigate the detrimental effects of probe differences and hence provide much more accurate predictions than commonly used multivariate linear calibration models (such as PLS) with and without empirical data preprocessing methods such as orthogonal signal correction, standard normal variate, or multiplicative signal correction.