基于人工神经网络的柴油机失火故障诊断

车载诊断系统在诊断失火故障时,采用基于曲轴段角加速度和阈值规则相结合的方法,该方法在内燃机高速轻载运行时诊断单缸完全失火工况存在一定的局限性。通过对比分析失火和正常工况下曲轴瞬时转速的幅频和相频特征,提取不同谐次的幅值和相位信息,结合人工神经网络作为故障模式识别工具,得到了一种改善方法。通过台架实验,对此改善方法进行了单缸完全失火、两缸完全失火和单缸一定程度失火的故障诊断测试。结果表明,在实验条件下该方法可以有效识别不同的失火模式,并可在单缸失火模式下实现失火程度判别。同时,该方法通过少量工况数据训练神经网络,即可实现一定转速范围内的失火诊断,可行性强,可用于发动机失火故障在线诊断。     

基于神经网络-遗传算法优化柴油机油台架试验

根据15W/40 CD柴油机油的模拟实验与台架试验的基础数据,用人工神经网络(ANN)的反向传播算法建立了模拟实验与台架试验神经网络预测模型,该神经网络模型合适的拓扑结构为5-7-1,学习速率为0.2,动量因子为0.9。探讨了用模拟实验数据预测台架试验结果的可能性,检验证明用人工神经网络方法建立的模型能准确预报15W/40CD柴油机油的台架试验结果。该神经网络预测模型用遗传算法优化,得到了15W/40 CD级柴油机油能通过台架试验的最优模拟实验结果。     

基于神经网络的多维力传感器静态解耦的研究

提出了基于人工神经网络进行多维力传感器静态解耦的方法。维间耦合是制约多维力传感器测量精度的主要因素，为克服传统线性解耦方法的局限性，利用BP神经网络的强非线性逼近能力研究了多维力传感器的非线性静态解耦。以研制的微型5维指尖力／力矩传感器为对象进行了解耦实验，结果表明，与基于最小二乘的线性解耦方法相比，提高了解耦精度。     

基于多压力变温实验的缸套-活塞环摩擦因数预测研究

为探究内燃机中缸套-活塞环摩擦副在复杂工况下的摩擦因数变化规律,根据实际工况下的温度和压力,使用SRV高温摩擦试验机模拟内燃机系统缸套-活塞环之间的往复式摩擦运动,获得不同温度和压力下的摩擦因数,并测试不同温度下润滑油的黏度。以摩擦接触面温度、载荷和润滑油黏度为输入,摩擦因数为输出,构建BP神经网络预测模型。结果表明,BP神经网络具有较高的预测精度,测试集误差值稳定在1%以内,满足工程精度要求。     

Sparse Autoencoder-based Multi-head Deep Neural Networks for Machinery Fault Diagnostics with Detection of Novelties

Supervised fault diagnosis typically assumes that all the types of machinery failures are known.However,in practice unknown types of defect,i.e.,novelties,may occur,whose detection is a challenging task.In this paper,a novel fault diagnostic method is developed for both diagnostics and detection of novelties.To this end,a sparse autoencoder-based multi-head Deep Neural Network(DNN)is presented to jointly learn a shared encoding representation for both unsupervised reconstruction and supervised classification of the monitoring data.The detection of novelties is based on the reconstruction error.Moreover,the computational burden is reduced by directly training the multi-head DNN with rectified linear unit activation function,instead of performing the pre-training and fine-tuning phases required for classical DNNs.The addressed method is applied to a benchmark bearing case study and to experimental data acquired from a delta 3D printer.The results show that its performance is satisfactory both in detection of novelties and fault diagnosis,outperforming other state-of-the-art methods.This research proposes a novel fault diagnostics method which can not only diagnose the known type of defect,but also detect unknown types of defects.     

Prediction of friction coefficient of treated betelnut fibre reinforced polyester (T-BFRP) composite using artificial neural networks

The current work is an attempt of using artificial neural network configuration to predict frictional performance of treated betelnut fibre reinforced polyester (T-BFRP) composite. Experimental dataset at different applied loads (5-30 N) and sliding distances (0-6.72 km) was used to train the ANN configuration with a large volume of experimental data (492 sets) where three different fibre mat orientations were considered (anti parallel, parallel and normal orientations). Results obtained from the developed ANN model were compared with experimental results. It is found that the experimental and numerical results showed good accuracy when the developed ANN model was trained with Levenberg-Marqurdt training function.     

概率神经网络PNN在发动机故障诊断中的应用

介绍了概率神经网络的模型，分析了其特点，并探讨了基于PNN的发动机故障诊断方法。通过MATLAB进行仿真试验，结果表明基于概率神经网络的故障诊断方法可以最大程度地利用故障先验知识，提高发动机故障诊断的准确率。     

制造过程多目标优化的集成计算智能方法

针对制造过程因动态多变而难以定量控制的问题,提出了用集成计算智能方法进行多目标优化。利用人工神经网络进行系统建模,并为遗传算法找到适应度函数及求得目标函数值的方法,进而利用遗传算法进行多目标优化。通过实例验证了方法的有效性与实用性,实现了制造过程的定量分析,为复杂制造系统的建模和优化提出了一种新的方法。     

人工神经网络及其在机械加工中的应用实例

介绍了人工神经网络知识及其在磨削表面粗糙度研究中的应用和建模过程。试验表明神经网络能自适应各种加工条件,具有较高的柔性和智能,能更准确地反映加工因素之间的变化关系,并可推广到其他领域中处理模糊的、非线性的问题。     

A novel approach for the prediction of bend allowance in air bending and comparison with other methods

Air-bending is a major sheet-metal forming operation, where precise prediction of the developed shape is a key factor for the accuracy assessment of the final shape for the part after bending. To predict the blank shape, accurate estimation of the bend-allowance (BA) is necessary, which can be defined as the length of the un-stretched fiber at the bent state of shape. There are several different approaches to find the BA values depending on either experience-based or knowledge-based techniques. In this paper, a brief summary is provided for different approaches to find the BA values by comparing their advantages as well as, their drawbacks. They are evaluated in terms of accuracy, efficiency and ease of implementation for integrated CAD/CAM environment. Then, a novel approach; by using higher order response surface (RS) fitting for the prediction of BA during air-bending is demonstrated. This technique is in general found very promising as an integrated tool for both CAD interfaces, as well as CNC machine tools. The RS predictions, which are generated from over 1,000 bending experiments using combinations of bending radius, bending angle and material thickness, are built for different orders and are compared to Artificial Neural Network (ANN) models that are also trained by using the same experimental data.     

Prediction of wear behaviour of A356/SiCp MMCs using neural networks

In the present investigation, artificial neural network (ANN) approach was used to predict the wear behaviour of A356/SiC metal matrix composites (MMCs) prepared using rheocasting route. The ANN model was obtained to aid in prediction and optimization of the wear rates of the composites. The effect of the SiC particles size, SiC weight percent, applied pressure and test temperature on the wear resistance was evaluated using the ANN model. The results have shown that ANN is an effective tool in the prediction of the properties of MMCs, and quite useful instead of time-consuming experimental processes.     

A study on separate learning algorithm using support vector machine for defect diagnostics of gas turbine engine

A separate learning algorithm with support vector machine (SVM) has been studied for the development of a defect-diagnostic algorithm applied to the gas turbine engine. The system using only an artificial neural network (ANN) falls in a local minima and its classification accuracy rate becomes low in case it is learning nonlinear data. To make up for this risk, a separate learning algorithm combining ANN with SVM has been proposed. In the separate learning algorithm, a sequential ANN learns selectively after classification of defect patterns and discrimination of defect position using SVM, resulting in higher classification accuracy rate as well as the rapid convergence by decreasing the nonlinearity of the input data. The results have shown this suggested method has reliable and suitable estimation accuracy of the defect cases of the turbo-shaft engine. This paper was recommended for publication in revised form by Associate Editor Dongsik Kim Tae-Seong Roh received his B.S. and M.S. degrees in Aeronautical Engineering from Seoul National University in 1984 and 1986. He then went on to receive his Ph.D. degree from Pennsylvania State University in 1995. Dr. Roh is currently a Professor at the department of Aerospace Engineering at Inha University in Incheon, Korea. His research interests are in the area of combustion instabilities, rocket and jet propulsions, interior ballistics, and gasturbine engine defect diagnostics. Dong-Whan Choi received his B.S. degree in Aeronautical Engineering from Seoul National University in 1974. He then went on to receive his M.S. and Ph.D. degrees from University of Washington in 1978 and 1983. Dr. Choi had served three years as a President of Korea Aerospace Research Institute since 1999. He is currently a Professor at the department of Aerospace Engineering at Inha University in Incheon, Korea. His research interests are in the area of turbulence, jet propulsions, and gasturbine defect diagnostics.     

运用DE-RBFNN的直线伺服系统定位精度研究

提出一种基于差分进化算法(DE)的径向基函数神经网络(RBFNN)模型,用于预测直线伺服系统的定位误差.该模型用差分进化算法训练径向基函数(RBF)网络隐层中心位置、宽度和输出层连接权重.为了评价优化后RBF网络预测的精度,运用部分误差样本进行训练和仿真.构建了以数字信号处理器(DSP)为核心的直线电动机定位误差实验平台,根据误差校正值进行误差实时补偿实验.仿真和实验结果表明:经过DE算法训练的神经网络模型对工作台的误差具有良好的学习能力和泛化能力,与单纯RBF网络、基于遗传优化的RBF神经网络相比,该建模方法具有更高的定位精度.     

基于神经网络与遗传算法的振动钻削仿真

将人工神经网络与遗传算法相融合,利用遗传算法优化多层前馈神经网络的结构和初始参数,利用多层前馈网络的映射能力进行振动钻削仿真,最后对训练好的神经网络稍加改造,利用BP网的逆映射进行参数优化。试验表明神经网络与遗传算法应用于振动钻削过程仿真是可行的,为振动钻削的研究提供了新的思路。     

内燃机缸盖系统模式识别方法研究

根据往复式内燃机结构和工作特点 ,对内燃机缸盖系统模式的识别方法进行了研究。根据激励与内燃机相位的对应关系 ,提出了基于相位相关的模式分解方法 ,用子模式集合描述缸盖系统模式。对人工神经网络技术用于缸盖系统模式识别进行的研究表明 ,人工神经网络技术对内燃机缸盖系统模式识别具有较强的分类表达和诊断能力     

神经网络矢量控制在永磁同步电动机中的应用

在永磁同步电动机(PMSM)矢量控制基本原理分析的基础上,提出了一种基于矢量控制的新型人工神经网络(ANN).人工神经网络用于速度控制和空间矢量脉宽调制(SVM).神经网络速度控制器不依赖于系统精确数学模型,具有动态响应快和稳态精度高的特点;同时基于SVM的人工神经网络算法(ANN-SVM)较易实现,它的计算量小,而且能有效降低谐波干扰.在Matllab/Simulink环境下,建立了一个基于人工神经网络矢量控制的PMSM仿真模型,并对其进行研究.仿真结果证明所提出的基于矢量控制的人工神经网络的可行性和有效性.     

Analysis and modeling of exhaust gas temperature in an ethanol fuelled HCCI engine

Low exhaust temperature in homogeneous charge compression ignition (HCCI) significantly limits efficiency of an exhaust aftertreatment system to mitigate high HC and CO emissions in HCCI engines. This article aims to understand the effect of varying input parameters on HCCI exhaust gas temperature (T_exh) for an ethanol fuelled engine. A single cylinder engine is used to collect experimental data at 100 different HCCI conditions. The results indicate that variation in combustion parameters such as start of combustion (SOC), burn duration (BD) and maximum in-cylinder pressure (P_max) are not effectively correlated with variations of Texh, but the indicated mean effective pressure (IMEP) and constant-volume adiabatic flame temperature (T_ad) are strongly related to T_exh. These experimental findings were then used to design an artificial neural network (ANN) model to predict T_exh. The model was validated with the experimental data, indicating an average error less than 4.5°C between predicted and measured T_exh.     

基于BP神经网络的TC4钛合金超塑性变形后组织及性能预测研究

分别用Visual Fortran语言和MATLAB软件建立了TC4钛合金超塑性变形时变形参数与其力学性能和晶粒尺寸之间的BP神经N络模型，通过用较少的力学性能和晶粒尺寸的试验数据进行训练，进而对其性能进行预测。结果表明，BP神经网络用于材料超塑性变形后的力学性能及晶粒尺寸预测是可行的，其预测误差小于7%。