基于神经网络的激光熔覆高度预测

激光成形过程中,对熔覆高度进行实时检测,从而实现熔覆高度闭环控制是成形高质量零件的保证.激光成形过程是一个多参数耦合的非线性过程,大量激光参数对成形熔覆表面质量具有重要影响.在分析激光参数对熔覆高度影响的基础上,建立利用激光工艺参数预测熔覆高度的误差反向传播(Back propagation,BP)神经网络模型,完成了网络算法设计.通过激光成形试验采集样本,利用训练样本对所建立的网络进行训练,完成网络输入输出高度映射关系,并利用测试样本对所训练的网络进行检验.仿真试验表明,神经网络熔覆高度预测模型具有很高的精度,验证了该预测模型在理论和实践上的可行性与有效性.神经网络熔覆高度预测模型为实现激光加工过程熔覆高度实时预测与闭环控制打下基础,对提高成形产品质量具有重要意义.     

混沌算法在成型机节能中的应用

电火花成型机加工工艺复杂,加工过程中受影响的因素众多,其中电参数的选择对加工结果有很大的影响,但影响规律很难用精确的数学模型来表达.采用基于混沌神经网络算法的Bp网络对电火花成型机加工工艺效果进行了预测,结果表明:该预测模型能有效地避免BP算法在能量的最小化过程中陷入局部极小化的问题而得到最优解,最终能够很好地映射出电参数与该加工网络电火花加工工艺效果之间的关系.     

电弧铣削加工间隙电压预测模型的建立与研究

针对电弧铣削加工过程中极间间隙难以直接测量的问题,通过极间间隙电压的变化判断极间间隙的变化,采用系统辨识理论确定间隙电压预测模型的结构和模型参数,对该预测模型的建立方法进行了详细阐述,对预测模型的拟合程度进行了实验验证。实验结果表明,预测模型的拟合精度随着拟合时间的延长而降低,因此,采用递推最小二乘方法进行间隙电压的在线预测,在线预测的平均误差为6.82%,结果表明所建模型能够稳定、有效地超前一步预测间隙电压,并且模型在线辨识的参数少,模型预测的精度高。     

薄壁件残余应力变形仿真预测与切削参数优化

针对薄壁件铣削残余应力变形难以准确预测的问题,提出了一种仿真预测方法,并在此基础上研究了薄壁件铣削切削参数优化方法。首先,提出了基于工况映射与薄壳应力贴合的残余应力变形仿真预测方法,实验结果表明该方法能够有效预测薄壁件的加工残余应力变形。在此基础上,利用支持向量回归机建立了基于切削参数的残余应力变形响应预测模型;然后,根据所建立的预测模型,采用遗传算法,以残余应力变形为约束、最大加工效率为目标对工艺参数进行优化。结果分析表明,该优化方法获得了最优的加工参数。     

基于支持向量机的电火花加工8418钢表面粗糙度预测模型

由于电火花加工过程的复杂性,单纯通过电火花加工实验方法研究各种放电参数及非电参数对工件表面粗糙度Ra的影响不但耗费大量时间,而且实验成本较高,为此基于支持向量机提出了一种适用于电火花加工表面粗糙度预测的模型。利用遗传算法对该模型中的各参数进行优化,预测不同电火花加工参数组合下的表面粗糙度;以电火花加工8418模具钢为例,将预测值与实验值进行对比,并且通过实验验证了电火花加工8418钢表面粗糙度预测模型参数的准确性;最后进行了误差分析,模型的最大误差值为2.27%。     

数字孪生驱动的数控铣削刀具磨损在线监测方法

针对数控铣床不断老化导致刀具磨损预测模型误差较大,加工过程中动态数据难以在线采集等问题,提出一种数字孪生驱动的刀具磨损在线监测方法。采用神经网络对加工过程中的多源数据进行特征提取,建立考虑机床老化的刀具磨损时变偏差量化模型,并在此基础上提出数控铣削刀具磨损的在线预测方法;开发了面向刀具磨损的数控铣削数字孪生系统,在线感知加工过程中的动态数据并实时仿真刀具磨损过程;最后,将该方法应用于实际加工中并与其他的预测方法进行了对比,结果表明该方法有效降低了机床老化带来的误差,实现了刀具磨损的精确预测。     

模糊神经网络在工程陶瓷电火花加工中的应用

在工程陶瓷电火花加工过程中,电参数与表面粗糙度之间具有高度的非线性关系,很难在两者之间建立精确的数学模型来预测其加工效果。针对这一问题,提出了将模糊控制理论和神经网络技术相融合的智能控制方法,采用多个输入单输出的模糊控制神经网络结构,建立了工程陶瓷碳化硼电火花加工的数学模型来预测其工艺效果。该模型能够很好地反映出电参数与表面粗糙度之间的非线性关系。通过实验值与网络模型的预测值之间的比较,验证了该模型具有较高的预测精度,证实了模型的可靠性和有效性。     

铣削力预测方法和影响因素综述

为减少航空发动机薄壁件铣削加工过程中的加工变形,提高加工质量,需对铣削加工过程中的切削力进行预测。因此,综述了多远回归分析预测模型、微元铣削力预测模型、有限元预测模型和人工神经网络预测模型,并对切削用量、刀具几何参数、工件材料、冷却作用、刀具材料和刀具磨损对铣削力的影响进行了分析。     

加工表面残余应力的有限元—应力松弛联合建模及预测分析

针对切削加工表面残余应力有限元模型预测效率低、解析预测模型互换性不强的问题,提出切削有限元模型和应力松弛解析模型的联合建模预测。基于有限元仿真模拟切削加工过程获取已加工表面的应力、应变及温度等基础物理变量,通过应力松弛解析模型计算残余应力,通过H13热作模具钢直角切削实验和有限元仿真验证预测模型,基于联合模型分析了不同刀具几何参数对残余应力分布的影响规律和显著性。结果表明,该模型能高效、低成本、高精度地预测沿工件表面深度的残余应力分布。本研究对促进残余应力预测方法的发展有一定的参考价值。     

基于赋值型误差传递网络的多工序加工质量预测

加工质量实时预测是工件多工序加工质量控制的关键。航空制造领域关键零部件的异形空间大尺寸、材料难加工与小批量加工等特性,导致加工样本数据不足与加工误差监测困难。针对上述问题,提出一种基于赋值型误差传递网络的多工序加工质量预测建模方法。通过将质量特征引入多工序误差传递网络来描述加工过程中节点间的影响关系,形成赋值型的误差传递网络。并以关键质量特征节点为基础,采用基于粒子群算法优化的支持矢量回归机方法,构建单工序质量预测模型。在此基础上,基于赋值型误差传递网络的拓扑结构,合并单工序加工质量预测模型,以构建多工序加工质量预测模型。最后,开发了一个面向多工序加工质量预测的软件平台并以起落架零件的加工为例验证上述模型,结果表明该方法能够有效地预测加工误差,并从多工序的角度为异形零件的加工过程控制提供依据。     

刀具磨损估计的多信号人工神经网络方法研究

本文研究了采用多种传感信号经人工神经网络估计刀具磨损量的方法，提出了有监督线性特征映射算法，研究了网络参数对学习速度和网络精度的影响，并与多层前向网络（ＢＰ算法）进行对比。研究表明，有监督线性特征映射网络具有学习快、精度高的优点，具有广阔的应用前景     

涂层刀具高速铣削模具钢SKD11的表面粗糙度模型预测

为有效控制和预测高硬度模具钢加工的表面质量和加工效率,通过设计正交切削试验,研究了在不同切削参数组合(主轴转速、进给速度、轴向切削深度和径向切削深度)及冷却润滑方式条件下、Ti Si N涂层刀具对模具钢SKD11(62HRC)的高速铣削。应用BP神经网络原理建立表面粗糙度预测模型,并进行试验验证其准确性。研究表明,在不同加工条件下,基于BP神经网络模型建立的涂层刀具铣削模具钢SKD11表面粗糙度模型有较好的预测精度,其预测误差在3.45%-6.25%之间,对于模具制造企业选择加工工艺参数、控制加工质量和降低加工成本有重要意义。     

Statistical process control with intelligence using fuzzy ART neural networks

With the automation development of manufacturing processes, artificial intelligence technology has been gradually employed to increase the automation and intelligence degree in quality control using statistical process control (SPC) method. In this paper, an SPC method based on a fuzzy adaptive resonance theory (ART) neural network is presented. The fuzzy ART neural network is applied to recognize the special disturbance of the manufacturing processes based on the classification on the histograms, which shows that the fuzzy ART neural network can adaptively learn the features of the histograms of the quality parameters in manufacturing processes. As a result, the special disturbance can be automatically detected when a feature of the special disturbance starts to appear in the histograms. At the same time, combined with spectrum analysis of the autoregressive model of quality parameters, the fuzzy ART neural network can also be utilized to adaptively detect the abnormal patterns in the control chart.     

Statistical process control with intelligence using fuzzy ART neural networks

With the automation development of manufacturing processes, artificial intelligence technology has been gradually employed to increase the automation and intelligence degree in quality control using statistical process control (SPC) method. In this paper, an SPC method based on a fuzzy adaptive resonance theory (ART) neural network is presented. The fuzzy ART neural network is applied to recognize the special disturbance of the manufacturing processes based on the classification on the histograms, which shows that the fuzzy ART neural network can adaptively learn the features of the histograms of the quality parameters in manufacturing processes. As a result, the special disturbance can be automatically detected when a feature of the special disturbance starts to appear in the histograms. At the same time, combined with spectrum analysis of the autoregressive model of quality parameters, the fuzzy ART neural network can also be utilized to adaptively detect the abnormal patterns in the control chart.     

基于深度置信网络的大数据制造过程实时智能监控

针对基于浅层学习模型的过程监控方法难以对大数据制造过程运行状态进行实时智能监控的问题,提出了基于深度置信网络的大数据制造过程实时智能监控方法。利用灰度图建立大数据制造过程质量图谱,以精准表达其过程的运行状态;构建用于识别大数据制造过程质量图谱的深度置信网络;应用离线训练好的深度置信网络模型对当前监控窗口内的过程质量图谱进行识别,实现大数据制造过程实时智能监控。最后,应用该方法对某注塑件大数据制造过程进行实时质量智能监控,结果表明：所提方法的识别性能明显优于基于主成分分析与BP神经网络、支持向量机的识别模型,能有效应用于大数据制造过程实时质量智能监控。     

MapReduce Based Parallel Bayesian Network for Manufacturing Quality Control

Increasing complexity of industrial products and manufacturing processes have challenged conventional statistics based quality management approaches in the circumstances of dynamic production. A Bayesian network and big data analytics integrated approach for manufacturing process quality analysis and control is proposed. Based on Hadoop distributed architecture and MapReduce parallel computing model, big volume and variety quality related data generated during the manufacturing process could be dealt with. Artificial intelligent algorithms, including Bayesian network learning, classification and reasoning, are embedded into the Reduce process. Relying on the ability of the Bayesian network in dealing with dynamic and uncertain problem and the parallel computing power of MapReduce, Bayesian network of impact factors on quality are built based on prior probability distribution and modified with posterior probability distribution. A case study on hull segment manufacturing precision management for ship and offshore platform building shows that computing speed accelerates almost directly proportionally to the increase of computing nodes. It is also proved that the proposed model is feasible for locating and reasoning of root causes, forecasting of manufacturing outcome, and intelligent decision for precision problem solving. The integration of bigdata analytics and BN method offers a whole new perspective in manufacturing quality control.     

基于知识深度置信网络的加工粗糙度预测

深度神经网络是一种具有复杂结构和多个非线性处理单元的模型,目前也已逐步被应用在工业生产过程中。但由于神经网络不可解释,不可控制的"黑箱"问题,以及海量的数据需求问题,使得深度学习在工业领域的应用仍有巨大的障碍。提出一种新的深度神经网络模型：知识深度置信网络（Knowledge-based deep belief network,KBDBN）。这种逻辑符号语言与深度神经网络的结合,不仅使得模型具有良好的模式识别性能,还可自适应地确定网络模型并具有可解释和可视化特性。进一步提出基于KBDBN的工件表面粗糙度加工过程的预测模型,实现了精确预测且有效地提取了制造过程的关键知识。试验结果证明：相较于传统机器学习器,KBDBN的网络性能更加优越,具有可解释性,可应用性更强。创新性的将符号规则与深度学习相结合并建立加工粗糙度预测模型,可以在精准预测的前提下提取工艺知识,指导加工工艺优化。     

Process Monitoring Strategies for Surface Mount Manufacturing Processes

Establishing reliable surface mount assemblies requires robust design and assembly practices, including stringent process control schemes for achieving high yield processes and high quality solder interconnects. Conventional Shewhart-based process control charts prevalent in today's complex surface mount manufacturing processes are found to be inadequate as a result of autocorrelation, high false alarm probability, and inability to detect process deterioration. Hence, new strategies are needed to circumvent the shortcomings of traditional process control techniques. In this article, the adequacy of Shewhart models in a surface mount manufacturing environment is examined and some alternative solutions and strategies for process monitoring are discussed. For modeling solder paste deposition process data, a time series analysis based on neural network models is highly desirable for both controllability and predictability. In particular, neural networks can be trained to model the autocorrelated time series, learn historical process behavior, and forecast future process performance with low prediction errors. This forecasting ability is especially useful for early detection of solder paste deterioration, so that timely remedial actions can be taken, minimizing the impact on subsequent yields of downstream processes. As for the automated component placement process where very low fraction nonconforming frequently occurs, control-charting schemes based on cumulative counts of conforming items produced prior to detection of nonconforming items is more sensitive in flagging process deterioration. For the reflow soldering and wave-soldering processes, the use of demerit control charts is appealing as it provides not only better control when various defects with a different degree of severity are encountered, but also leads to an improved ARL performance. Illustrative examples of actual process data are presented to demonstrate these approaches.     

一种新型的神经网络及其在智能质量诊断分析中的应用

提出了一种适用于模式识别的新型神经网络模型———局部有监督特征映射 (RegionalSupervisedFeatureMapping, RSFM)网络,将其应用到质量控制图的模式识别中,为基于统计过程控制(SPC)的智能工序质量诊断分析系统提供了技术支持。文中研究了网络的基本性能并对其参数进行优化,提出了采用欧氏距离判别法作为混合型多特征异常模式的识别方法。实验证明,所提出的模型对控制图的基本模式和混合型多特征异常模式都能够有效识别,网络收敛速度快、识别精度高,可进行大样本训练,适用于控制图的在线实时模式识别。     

On modeling of tool wear using sensor fusion and polynomial classifiers

With increased global competition, the manufacturing sector is vigorously working on enhancing the efficiency of manufacturing processes in terms of cost, quality, and environmental impact. This work presents a novel approach to model and predict cutting tool wear using statistical signal analysis, pattern recognition, and sensor fusion. The data are acquired from two sources: an acoustic emission sensor (AE) and a tool post dynamometer. The pattern recognition used here is based on two methods: Artificial Neural Networks (ANN) and Polynomial Classifiers (PC). Cutting tool wear values predicted by neural network (ANN) and polynomial classifiers (PC) are compared. For the case study presented, PC proved to significantly reduce the required training time compared to that required by an ANN without compromising the prediction accuracy. The predicted results compared well with the measured tool wear values.