Feature evaluation and extraction based on neural network in analog circuit fault diagnosis

Choosing the right characteristic parameter is the key to fault diagnosis in analog circuit. The feature evaluation and extraction methods based on neural network are presented. Parameter evaluation of circuit features is realized by training results from neural network; the superior nonlinear mapping capability is competent for extracting fault features which are normalized and compressed subsequently. The complex classification problem on fault pattern recognition in analog circuit is transferred into feature processing stage by feature extraction based on neural network effectively, which improves the diagnosis efficiency. A fault diagnosis illustration validated this method.

RESEARCHES ON THE STABILITIES OF ANALOG ELECTRONIC NEURAL NETWORKS

A new method for analyzing the stabilities of analog electronic neural networks ispresented.The energy functions with clear physical meaning are derived by introducing the staticequivalent circuit models,which has expanded the Tellegen Theorem for application on circuitanalysis.The method used to derive the energy functions of nets from first order differentialequations is valid for all first order continuous autonomous systems.The stability analysis ofcellular neural networks is made by the use of the stationary cocontent theorem.Some resultsare instructive for the network implementation on circuits.

Fault Diagnosis of Analog Circuit Based on PSO and BP Neural Network

In order to improve the speed and accuracy of analog circuit fault diagnosis,using Back Propagation Neural Network(BPNN),a new method is proposed based on Particle Swarm Optimization(PSO)to adjust weights of BP neural network.The model can not only overcome the limitations of the slow convergence and the local extreme values by basic BP algorithm,but also improve the learning ability and generalization ability with a higher precision.The response signals of analog circuit is preprocessed by Wavelet Packet Transform(WPT)as the fault feature.The simulation result shows that the proposed method has higher diagnostic accuracy and faster convergence speed,which is effective for fault location.

Analog-Circuit Model of FGH96 Superalloy Hot Deformation Behaviors Based on Artificial Neural Network

At the present time, numerical models （such as, numerical simulation based on FEM） adopted broadly in technological design and process control in forging field can not implement the realtime control of material forming process. It is thus necessary to establish a dynamic model fitting for the real-time control of material deformation processing in order to increase production efficiency, improve forging qualities and increase yields. In this paper, hot deformation behaviors of FGH96 superalloy are characterized by using hot compressive simulation experiments. The artificial neural network （ANN） model of FGH96 superalloy during hot deformation is established by using back propagation （BP） network. Then according to electrical analogy theory, its analog-circuit （AC） model is obtained through mapping the ANN model into analog circuit. Testing results show that the ANN model and the AC model of FGH96 superalloy hot deformation behaviors possess high predictive precisions and can well describe the superalloy＇s dynamic flow behaviors. The ideas proposed in this paper can be applied in the real-time control of material deformation processing.

Combinatorial Optimization Based Analog Circuit Fault Diagnosis with Back Propagation Neural Network

Electronic components' reliability has become the key of the complex system mission execution. Analog circuit is an important part of electronic components. Its fault diagnosis is far more challenging than that of digital circuit. Simulations and applications have shown that the methods based on BP neural network are effective in analog circuit fault diagnosis. Aiming at the tolerance of analog circuit,a combinatorial optimization diagnosis scheme was proposed with back propagation( BP) neural network( BPNN).The main contributions of this scheme included two parts:( 1) the random tolerance samples were added into the nominal training samples to establish new training samples,which were used to train the BP neural network based diagnosis model;( 2) the initial weights of the BP neural network were optimized by genetic algorithm( GA) to avoid local minima,and the BP neural network was tuned with Levenberg-Marquardt algorithm( LMA) in the local solution space to look for the optimum solution or approximate optimal solutions. The experimental results show preliminarily that the scheme substantially improves the whole learning process approximation and generalization ability,and effectively promotes analog circuit fault diagnosis performance based on BPNN.

STUDIES ON PROGRAMMING FEATURES AND METHODS OF FLOATING GATE MOSFET AS ANALOG MEMORY FOR SYNAPTIC WEIGHTS IN NEURAL NETWORKS

The features of the floating gate devices as analog memory have been investigatedexperimentally.Programming properties of the devices,compatibility and endurance of program-ming,and programming methods are presented in this paper.The results illustrate that thedevice can be used to store the analog weights for the neural networks,and the method that thestored value is adjusted continuously to approach to a given analog values is a rather practicalmethod for storing weights of neural networks.

Adaptive neural network based sliding mode altitude control for a quadrotor UAV

Reasons and realities such as being non-linear of dynamical equations,being lightweight and unstable nature of quadrotor,along with internal and external disturbances and parametric uncertainties,have caused that the controller design for these quadrotors is considered the challenging issue of the day.In this work,an adaptive sliding mode controller based on neural network is proposed to control the altitude of a quadrotor.The error and error derivative of the altitude of a quadrotor are the inputs of neural network and altitude sliding surface variable is its output.Neural network estimates the sliding surface variable adaptively according to the conditions of quadrotor and sets the altitude of a quadrotor equal to the desired value.The proposed controller stability has been proven by Lyapunov theory and it is shown that all system states reach to sliding surface and are remaining in it.The superiority of the proposed control method has been proven by comparison and simulation results.

基于小波神经网络的模拟集成电路短路故障诊断方法

针对传统方法短路故障诊断效果较差的问题,本文提出基于小波神经网络的模拟集成电路短路故障诊断方法。首先重构模拟集成电路原始信号,提取短路故障信号特征,然后利用确定的故障位置通过小波神经网络识别故障类型,最后判别短路故障严重程度,实现模拟集成电路短路故障诊断。实验结果表明:所提设计方法判别的短路故障程度与目标值十分接近,具有实际应用价值。

RBF神经网络在船舶模拟电路故障诊断中的应用

针对船舶模拟电路元件复杂交互,故障信号在大量的正常信号中难以凸显,故障特征提取识别难度较大的问题,提出基于RBF神经网络的船舶模拟电路故障诊断方法。由基于小波包的船舶模拟电路故障特征提取方法,以小波分解重构的方式,捕捉电路频带能量变化特征;使用基于状态转移算法优化RBF神经网络的故障诊断模型,由状态转移算法优化RBF神经网络参数,构建用于诊断电路故障的RBF神经网络模型后,学习所提取故障特征与类型之间关系,诊断新输入的船舶模拟电路输出信号故障类型。实验测试结果显示,此方法在有效捕捉船舶模拟电路故障频带能量变化特征后,对多种船舶模拟电路故障的诊断结果均未出现明显错误。

基于多尺度卷积神经网络和注意力机制的模拟电路早期故障诊断方法

模拟电路具有非线性、元件容差等特性,导致不同故障模式之间存在混叠现象,特别是模拟电路早期故障,这大大增加了故障诊断的难度。因此,提出了一种基于小波变换和多尺度特征注意力卷积神经网络(MS-FACNN)的模拟电路早期故障诊断方法,采用小波变换得到脉冲响应信号的多尺度分量,利用设计好的MS-FACNN网络自动提取更加全面且高可分性故障特征,并实现故障模式识别。此外,采用高效通道注意力(ECA)聚焦故障高相关性特征,过滤低相关性的冗余信息,进一步提升模型特征提取能力。实验结果表明,相比传统方法,所提方法具有更强的故障特征提取能力,对四运放双二阶高通滤波器早期故障诊断的准确率达到99.18%。

数值仿真生成的汽车后空调气动噪声预测及评价

为解决需要项目开发后期样车完成后才能进行整车空调气动噪声性能测试及评价的滞后性问题,针对某SUV整车后空调高档范围工作产生的气动噪声,提出了一种基于空调整车计算流体动力学和FW-H声类比方程的气动噪声仿真计算分析和神经网络的主观评价预测方法。首先采用计算流体动力学和FW-H声类比相结合的方法仿真计算和验证后空调高档运行时车内产生的气动噪声特性;然后将仿真得到的时域气动噪声样本转化成声频格式,并开发GUI程序界面进行噪声样本主观评价和客观参数计算;最后建立基于遗传算法优化的主客观映射神经网络预测模型以实现车内后空调气动噪声性能的预测评价。仿真计算及预测评价结果表明:该方法计算的气动噪声仿真误差在10%以内,主观预测误差在0.5分以内,可有效指导汽车空调气动噪声性能的前瞻性预测开发。

一种面向基于闪存的脉冲卷积神经网络的模拟神经元电路

该文面向基于闪存(Flash)的脉冲卷积神经网络(SCNN)提出一种积分发放(IF)型模拟神经元电路,该电路实现了位线电压箝位、电流读出减法和积分发放功能。为解决低电流读出速度较慢的问题,该文设计一种通过增加旁路电流大幅提高电流读出范围和读出速度的方法;针对传统模拟神经元复位方案造成的阵列信息丢失问题,提出一种固定泄放阈值电压的脉冲神经元复位方案,提高了阵列电流信息的完整性和神经网络的精度。基于55 nm互补金属氧化物半导体(CMOS)工艺对电路进行设计并流片。后仿结果表明,在20 μA电流输出时,读出速度提高了100%,在0 μA电流输出时,读出速度提升了263.6%,神经元电路工作状态良好。测试结果表明,在0~20 μA电流输出范围内,箝位电压误差小于0.2 mV,波动范围小于0.4 mV,电流读出减法线性度可达到99.9%。为了研究所提模拟神经元电路的性能,分别通过LeNet和AlexNet对MNIST和CIFAR-10数据集进行识别准确率测试,结果表明,神经网络识别准确率分别提升了1.4%和38.8%。

基于二维卷积神经网络的模拟电路故障诊断方法

传统的基于机器学习的模拟电路故障诊断方法依赖复杂的信号处理技术和专业知识来进行故障特征提取,其故障诊断过程复杂。针对上述问题,提出一种基于二维卷积神经网络(2D-CNN)的模拟电路故障诊断方法,将被测电路的原始输出电压转换成故障灰度图(faultgrayimage,简称FGI),作为2D-CNN模型的输入,使用模型的卷积层自动提取故障的深层特征,并在模型中通过添加批量归一化(BatchNormalization,简称BN)层对数据分布进行正则化,以减小数据分布偏移带来的影响。该方法在Sallen-Key带通滤波器电路和四阶二运放高通滤波器电路的故障诊断实验中分别实现了100%和99.46%的故障诊断率。该方法不仅简化了故障诊断流程,还保证了故障诊断精度,并具有较强的泛化能力。

基于改进蝶形反馈型神经网络的海关风险布控方法

针对现阶段我国海关风险布控方法存在效率、准确率较低、人力资源占用过多的问题和智能化分类算法小型化部署需求,提出一种基于改进蝶形反馈型神经网络(BFNet-V2)的海关风险布控方法。首先,运用编码填充(FC)算法实现海关表格数据到模拟图像的语义替换;其次,运用BFNet-V2训练模拟图像数据,由左右两条链路、不同卷积核和块、小块的设计组成规则的神经网络结构,并添加残差短路径干预改善过拟合和梯度消失;最后,提出历史动量自适应矩估计算法(H-Adam)优化梯度下降过程,取得更优的自适应学习率调整方式,并分类海关数据。选取Xception(eXtreme inception)、移动网络(MobileNet)、残差网络(ResNet)和蝶形反馈型神经网络(BF-Net)为基线网络结构进行对比。BFNet-V2的接受者工作特征曲线(ROC)和查准率-查全率曲线(PR)包含了基线网络结构的曲线,与4种基线网络结构相比,基于迁移学习(TL)的BFNet-V2分类准确率分别提高了4.30%、4.34%、4.10%和0.37%。在真实标签数据分类过程中,BFNet-V2的查获误判率分别降低了70.09%、57.98%、58.36%和10.70%。比较所提方法与包含浅层和深度学习方法在内的8种分类方法,在3个数据集上的准确率均提升1.33%以上,可见所提方法能够实现表格数据自动分类,提升海关风险布控的效率和准确度。

基于压缩子波的碳酸盐岩薄储层孔隙度反演方法应用——以M油田为例

由于油田开发程度的不断加深及地震资料分辨率偏低的限制,常规波阻抗反演方法很难满足油田精细开发对储层描述的要求。M油田为碳酸盐岩礁滩相储层,上覆巨厚膏盐层广泛发育,深层目的层地震主频衰减较快、频带较窄,部分储层段较薄也低于地震分辨率,且井壁垮塌现象普遍存在。针对上述难题,应用压缩地震子波方法提高地震分辨率,引入神经网络方法修正井垮塌段密度曲线。最终,综合井点及地震波阻抗信息,采用“井-震”波形类比方法实现了对薄储层孔隙度的预测。相比常规方法,该方法首次应用于礁滩相碳酸盐岩薄储层的预测,显著改善薄储层段的地震分辨率,并充分结合了地震和测井的波形信息,提高了孔隙度反演精度,揭示了储层纵横向展布规律,指导了M油田储量复算及薄储层段的新井部署实施。

GABP算法构建高精度CMOS电压自举采样开关性能预测模型

模拟集成电路中器件的设计参数与性能指标具有非线性映射关系,同时繁复的设计参数相互制约,使模拟IC满足应用约束下的折中设计极为复杂,电路研发耗时费力。基于自适应学习的神经网络算法能够建立具有非线性映射关系的预测模型,同时具有宽解空间和易获取全局最优解的遗传算法可进一步弥补建模和求解的精度。采用BP神经网络结合遗传算法(GABP)的复合优化框架对CMOS电压自举采样开关的设计参数和性能指标进行精准建模并优化整体电路性能,建模结果与单BP神经网络模型进行了对比,结果表明,GABP复合建模精度高于BP神经网络,拟合相关度从0.73007有效提升到0.94596,模型可靠性有大幅提高,证明了GABP复合优化可有效应用于电路性能的高效预测和辅助优化设计。

Soft Fault Diagnosis for Analog Circuits Based on Slope Fault Feature and BP Neural Networks

Fault diagnosis is very important for development and maintenance of safe and reliable electronic circuits and systems. This paper describes an approach of soft fault diagnosis for analog circuits based on slope fault feature and back propagation neural networks (BPNN). The reported approach uses the voltage relation function between two nodes as fault features; and for linear analog circuits, the voltage relation function is a linear function, thus the slope is invariant as fault feature. Therefore, a unified fault feature for both hard fault (open or short fault) and soft fault (parametric fault) is extracted. Unlike other NN-based diagnosis methods which utilize node voltages or frequency response as fault features, the reported BPNN is trained by the extracted feature vectors, the slope features are calculated by just simulating once for each component, and the trained BPNN can achieve all the soft faults diagnosis of the component. Experiments show that our approach is promising.

A Neural Network Appraoch to Fault Diagnosis in Analog Circuits

This paper presents a neural network based fault diagnosis approach for analog circuits, taking the tolerances of circuit elements into account. Specifi-cally, a normalization rule of input information, a pseudo-fault domain border (PFDB) pattern selection method and a new output error function are proposed for training the backpropagation (BP) network to be a fault diagnoser. Experi-mental results demonstrate that the diagnoser performs as well as or better than any classical approaches in terms of accuracy, and provides at Ieast an order-of magnitude improvement in post-fault diagnostic speed.

基于神经网络自学习的模拟电路故障诊断方法

由于传统方法在模拟电路故障诊断中应用效果不佳,不仅错误诊断数量较多,而且故障诊断用时较长,无法达到预期的故障诊断效果,提出基于神经网络自学习的模拟电路故障诊断方法。将模拟电路等效为一个双端口网络,利用便携式模拟仪器采集双端口网络运行数据,通过小波变换信号提取模拟电路信号时域特征和频域特征,利用神经网络自学习模型卷积运算信号特征,诊断并识别电路故障。经实验证明,设计方法错误诊断数量占总样本数量的比例较小,诊断用时较短,在模拟电路故障诊断方面具有良好的应用前景。

Hierarchical Neural Networks Method for Fault Diagnosis of Large-Scale Analog Circuits

A novel hierarchical neural networks (HNNs) method for fault diagnosis of large-scale circuits is proposed. The presented techniques using neural networks(NNs) approaches require a large amount of computation for simulating various faulty component possibilities. For large scale circuits, the number of possible faults, and hence the simulations, grow rapidly and become tedious and sometimes even impractical. Some NNs are distributed to the torn sub-blocks according to the proposed torn principles of large scale circuits. And the NNs are trained in batches by different patterns in the light of the presented rules of various patterns when the DC, AC and transient responses of the circuit are available. The method is characterized by decreasing the over-lapped feasible domains of responses of circuits with tolerance and leads to better performance and higher correct classification. The methodology is illustrated by means of diagnosis examples.