基于多模态集成卷积神经网络的数控机床齿轮箱故障诊断

针对数控机床齿轮箱在实际工作环境中负载多变且噪声干扰大、传统神经网络难以充分提取信号中的故障特征等问题,提出一种多模态集成卷积神经网络（MECNN）用于数控机床齿轮箱故障诊断。该方法将多模态融合技术与多个卷积神经网络结合,利用快速傅里叶变换方法将时域信号转换成频域信号;利用时域信号和频域信号对2个卷积神经网络进行训练,使模型能够分别从时域和频域2个角度提取特征,再将浅层特征融合;最后,将融合后的特征输入到卷积神经网络中进行故障特征的深度挖掘,并进行故障诊断。使用东南大学的齿轮箱数据集进行验证,设计了2种特征融合的方法并进行了对比。实验结果表明：在噪声下,MECNN模型用于故障诊断的准确性和鲁棒性均优于单一的时域CNN和频域CNN。     

Recognition of chatter with neural networks

Chatter deteriorates surface finish, reduces tool life, and damages machine tools. A chatter development prediction procedure is proposed for the cylindrical turning of long slender bars. The procedure uses two synthetically trained neural networks to recognize the harmonic acceleration signals and their frequency, and based on these observations, the future vibration characteristics of the system are estimated. The developed neural networks are capable of identifying 98% of the harmonic signals with over 90% certainty and estimate their frequencies with less than ±5% error from very short data sequences (only 11 sampled points). The accuracy of the neural networks is equivalent to time domain time series method based approaches; however, the proposed procedure can be implemented very quickly by using commercially available neural network hardware and software, and can use the new neural network chips to make the estimations very quickly by using parallel processors. The validity of the chatter prediction procedure is also demonstrated on the experimental data.     

Using neural network for tool condition monitoring based on wavelet decomposition

This paper presents a neural network application for on-line tool condition monitoring in a turning operation. A wavelet technique was used to decompose dynamic cutting force signal into different frequency bands in time domain. Two features were extracted from the decomposed signal for each frequency band. The two extracted features were mean values and variances of the local maxima of the absolute value of the composed signal. In addition, coherence coefficient in low frequency band was also selected as a signal feature. After scaling, these features were fed to a back-propagation neural network for the diagnostic purposes. The effect on tool condition monitoring due to the presence of chip breaking was studied. The different numbers of training samples were used to train the neural network and the results were discussed. The experimental results show that the features extracted by wavelet technique had a low sensitivity to changes of the cutting conditions and the neural network has high diagnosis success rate in a wide range of cutting conditions.     

三维椭圆振动切削系统模糊自适应滑模控制

三维椭圆振动切削被认为是目前最具潜力的机械加工方式,但是加工过程中的控制问题还未被很好的解决,特别是加工过程中对于外界干扰的自适应问题,为了在三维椭圆振动切削过程中实现控制方法的鲁棒性,根据所研制的一种三维椭圆振动切削装置独特的结构方式,首先分析了各个运动之间的串扰情况,并根据装置柔性铰链的特征建立了三维椭圆振动切削装置的动力学模型。提出了三维椭圆振动切削模糊自适应滑模控制的滑模函数与滑模控制律,并通过李雅普诺夫稳定性条件证明了所设计的滑模控制器系统稳定性,采用正弦信号数字实现模糊自适应滑模控制,位置跟踪在3 s内误差控制在0.005范围内,速度跟踪在0.4 s内控制在±0.05之内,能够达到满意的精度,系统模糊自适应滑模控制的模糊逼近也能在0.2 s内收敛,证明了三维椭圆振动切削系统采用模糊自适应滑模控制可以实现较强的鲁棒性。     

基于GRA-PLS的滚动轴承故障诊断北大核心

针对仅用时域和频域指标无法准确诊断滚动轴承故障的问题,提出一种基于灰色关联度(GRA)与偏最小二乘(PLS)的故障诊断算法。首先,对原始振动信号进行灰色关联度分析,提取关联度较高的振动信号作为样本信号;其次,通过时域分析和频域分析获得故障特征集,利用基于遗传算法(GA)和Elman神经网络的组合算法(GA-ENN)对故障特征进行提取;最后,利用PLS算法对滚动轴承的故障类别进行识别。实验结果表明,所提方法能有效剔除原始振动信号中无信息变量,并且实现时、频域指标下滚动轴承故障的准确诊断。     

Generalized modeling of drilling vibrations. Part I: Time domain model of drilling kinematics, dynamics and hole formation

This two part paper presents a comprehensive exercise in modeling dynamics, kinematics and stability in drilling operations. While Part II focuses on the chatter stability of drilling in frequency domain, Part I presents a three-dimensional (3D) dynamic model of drilling which considers rigid body motion, and torsional–axial and lateral vibrations in drilling, and resulting hole formation. The model is used to investigate: (a) the mechanism of whirling vibrations, which occur due to lateral drill deflections; (b) lateral chatter vibrations; and (c) combined lateral and torsional–axial vibrations. Mechanistic cutting force models are used to accurately predict lateral forces, torque and thrust as functions of feedrate, radial depth of cut, drill geometry and vibrations. Grinding errors reflected on the drill geometry are considered in the model. A 3D workpiece, consisting of a cylindrical hole wall and a hole bottom surface, is fed to the rotating drill while the structural vibrations are excited by the cutting forces. The mechanism of whirling vibrations is explained, and the hole wall formation during whirling vibrations is investigated by imposing commonly observed whirling motion on the drill. The time domain model is used to predict the cutting forces and frequency content as well as the shape of the hole wall, and how it depends on the amplitude and frequency of the whirling vibration. The model is also used to predict regenerative, lateral chatter vibrations. The influence of pilot hole size, spindle speed and torsional–axial chatter on lateral vibrations is observed from experimental cutting forces, frequency spectra and shows good similarity with simulation results. The effect of the drill–hole surface contact during drilling is discussed by observing the discrepancies between the numerical model of the drilling process and experimental measurements.     

Using detrended fluctuation analysis to monitor chattering in cutter tool machines

This paper considers accelerometer signals in order to detect chatter instabilities under different spindle speed and depth of cut ratio conditions. Detrended fluctuation analysis (DFA), adapted for time–frequency domain, was used to monitor the evolution of cutter tool dynamics. The DFA offers the advantage over traditional spectral analysis that can be deal with nonstationary, nonlinear data and, in contrast to wavelet approaches, its application does not rely on the selection of basis functions. The underlying idea behind the application is to use the Hurst exponent, an index of the signal fractal roughness, to detect dominance of unstable oscillatory components in the complex, presumably stochastic, dynamics of machine acceleration. Several experiments with a lab-scale cutting machine were performed to illustrate the ability of the DFA to detect unstable cutting behavior. The results, presented in time–frequency domain, show that instabilities are detected in a certain frequency range as the Hurst exponent decreases to reflect anti-persistency of the chatter dynamics.     

Design and analysis of a milling cutter with the improved dynamic characteristics

In this paper, a face-milling cutter is proposed and manufactured to improve a cutter’s dynamic characteristics. The proposed cutter, comprises double step blades, and is designed to give better stability in terms of vibration and to suppress the possibility of abrupt wear and chipping. Vibration experiments with the conventional type and the proposed cutter were performed, and the results showed that the vibration amplitude in the time domain and the peak value in the vibration spectrum in the frequency domain are considerably lower for the proposed cutter than for the conventional cutter. In addition, the validity of a cutting dynamics model, which can effectively predict the cutting dynamics on the machine–tool–workpiece (M–T–W), was examined by the vibration experiments.     

Dynamics of boring processes: Part III-time domain modeling

This article presents a mathematical model and a computational algorithm for the time domain solution of boring process dynamics. The model is developed in a modular form; it includes a workpiece geometry and surface topography module, a kinamatics and tool position module, a dynamic chip load module, a dynamic cutting force prediction module and a structural dynamics module. The time domain model takes cutting process parameters, tool and workpiece geometries and modal parameters of the structure as inputs. It predicts instantanous cutting forces and vibrations along the machining time, and machined workpiece topography as outputs. Some of the simulated and experimental results for various cutting conditions are presented and compared for validation purposes.     

Adaptive control optimization in end milling using neural networks

In this paper, we propose an architecture with two different kinds of neural networks for on-line determination of optimal cutting conditions. A back-propagation network with three inputs and four outputs is used to model the cutting process. A second network, which parallelizes the augmented Lagrange multiplier algorithm, determines the corresponding optimal cutting parameters by maximizing the material removal rate according to appropriate operating constraints. Due to its parallelism, this architecture can greatly reduce processing time and make real-time control possible. Numerical simulations and a series of experiments are conducted on end milling to confirm the feasibility of this architecture.     

基于小波包分解和奇异谱分析方法的立铣加工颤振监测

在立铣加工过程中,颤振是加工过程失稳的一个最重要的原因。颤振将会严重影响工件表面质量和材料去除率,加剧刀具磨损和恶化工作环境。虽然大部分颤振监测系统可以监测到颤振发生,但颤振发生时已经对工件和刀具产生了严重的损伤,因此,需要提前监测到颤振特征。在颤振发生过程中,振动信号具有在时域中不断增大,在频域中能量频移的特性。考虑这两个振动信号特征,提出了一种颤振特征提取方法。提取颤振发生频带中振动信号的能量比和奇异谱熵系数作为两个颤振特征,并通过人工神经网络模型实现切削颤振的识别。文中提出的颤振监测系统包括特征提取和分类,能够精确辨识立铣加工中的稳定、过渡和颤振状态。     

The influence of cryogenic cooling on milling stability

Cryogenic cooling is emerging as an effective process for high performance machining. However, the influence of cryogenic cooling on milling stability is seldom reported. This paper involves experimental study on the effect of cryogenic cooling on milling stability, using a dedicated cryogenic cooling system to applying liquid nitrogen (LN2) jet to the cutting zone. We observe that cryogenic cooling leads to higher stability limit compared with conventional milling operations, which indicates that the cutting efficiency can be improved greatly in LN2 environment as opposed to the conventional one. The stability improvement is explained from the perspective of machining dynamics parameters variation between the two conditions. Cutting force coefficients and modal parameters of spindle-tool system are identified during cryogenic machining, then milling stability lobe diagrams are predicted by time domain and frequency domain methods. On the basis of milling stability analysis, the enhancement of stability boundary is attributed to the significant reduction of cutting force coefficients during cryogenic cooling. Additionally, the experiment result indicates that cryogenic cooling decreases the dominant modal frequency of the spindle-tool system, which shifts the milling stability boundary slightly to lower spindle speed range. The explanations are verified by a plenty of cutting tests.     

Milling force convolution modeling for identification of cutter axis offset

This paper discusses the application of a convolution integral force model to the identification of the geometry of cutter axis offset in milling operations. This analysis builds upon the basis of linear decomposition of elemental local cutting forces into a nominal component and an offset-induced component. The convolution of each elemental local cutting force component with the chip width density in the context of cutter angular position provides an integral expression for the total cutting forces. By virtue of the convolution integration property, the total cutting forces in the frequency domain can be derived as closed-form functions of the cutting pressure constants, various cutting conditions, as well as the cutter offset geometry. Subsequently, the magnitude and phase angle of cutter axis offset are shown to be algebraic and explicit functions of the Fourier series coefficients of cutting forces at the spindle frequency. Following the theoretical analysis, experimental study is discussed to illustrate the implementation procedure for offset identification, and frequency domain data are presented to verify the analytical results. Potential industrial applications of this work include the real-time monitoring of dynamic cutter runout and the in-process compensation for the loss of tolerance or finish using automatic controls based on the feedback information of offset magnitude and phase angle.     

基于DE—BP神经网络的刀具寿命预测研究

引入了BP神经网络技术对刀具寿命进行预测,建立了刀具寿命预测模型.并针对BP神经网络所存在的缺陷,结合差异演化算法,提出了实数编码的DE-BP神经网络预测模型.实验表明,该模型对刀具寿命预测精度高,为刀具需求计划制定、成本核算、切削参数制定提供了理论依据,节约了制造执行系统成本.     

Cutting state monitoring in milling by a neural network

The application of a neural network to cutting state monitoring in face milling was introduced and evaluated on multiple sensor data such as cutting forces and vibrations. This monitoring system consists of a statistically based adaptive preprocessor (autoregressive (AR) time series modeling) for generating features from each sensor, followed by a highly parallel neural network for associating the preprocessor outputs (sensor fusion) with the appropriate decisions. AR model parameters were used as features, and the cutting states (normal, unstable and tool life end) were successfully detected by monitoring the evolution of model parameters during face milling. The proposed system offers fast operation through recursive preprocessing and highly parallel association, and a data-driven training scheme without explicit rules or a priori statistics. It appears proven on limited experimental data.     

An experimental analysis of damped coupled vibrations in broaching

The paper reports for the first time aspects related to dynamics of broaching when features with complex geometries are generated. It describes an experimental analysis of causes and outcomes of damped-coupled vibrations when broaching semi-closed profiles, i.e. dovetails of gas turbine engines disks. Singular indentations or groups of tilted chatter marks were found in particular zones on broached surfaces. Analysis of force and acceleration signals revealed that damped-coupled vibrations that result in tilted chatter marks mainly occur due to specific geometry of cutting edges that enable coupling of three-dimensional (3D) vibrations. A new method to detect the appearance of tilted marks as a result of damped-coupled vibrations with particular frequency has been proposed by monitoring the elliptical movement of cutting edge via time and frequency domain analysis of two acceleration signals.     

Identification of dynamic instabilities in machining process using the approximate entropy method

Chatter instability is a persistent problem in machining process that produces vibrations characterized by nonlinear and nonstationary dynamics. Although traditional fast Fourier transform approaches are typically used for the monitoring of chattering in industry, the method is suitable only for linear and stationary signals. In this paper, methods based on approximate entropy (ApEn) are proposed to identify chatter instabilities in milling process. as entropy is an index of the irregularity and complexity related to randomness from data series. The attractiveness of the ApEn approach is that it can deal with nonlinear and nonstationary data, requires a relatively small number of observations and can be used for noisy signals. For a lab-scale milling experimental setup, the ApEn was implemented under a time-frequency monitoring method, showing that instable chattering is associated with entropy increment for a frequency range. In contrast, stable milling led to an entropy pattern where high-entropy values are concentrated at high frequencies, which are related to the natural dynamics of the cutting tool.     

Inspection of micro-tools at high rotational speeds

Micro-tools have been widely used in industry, primarily by biomedical and electronic equipment manufacturers. The life of these cutting tools is extremely unrpedictable and much shorter than conventional tools. Also, these miniature tools, with a diameter of less than 1 mm, cannot be inspected by an operator without the aid of magnifying glass.

In this paper, evaluation of the intensity variation of a reflected laser light beam from the cutting tool surfaces is proposed as a method of estimating cutting tool surface conditions. Various encoding methods, including wavelet transformations, were proposed to obtain a small and meaningful set of data from the intensity variation readings of one tool rotation. The encoded data are classified using a simple threshold method, Restricted Coulomb Energy (RCE), and Adaptive Resonance Theory (ART2)-type neural networks. The proposed encoding and classification approaches were tested with over one hundred sets of data.

The threshold method detects only severe tool damage. The RCE neural networks and graphical presentation of the encoded sets demonstrated the feasibility of the proposed monitoring technique and encoding methods. The ART2-type neural networks were found to be the best candidate for tool condition monitoring because of their self learning capability. Wavelet transformation-based encoding and ART2-type neural networks were found to be sensitive enough to recognize wear at the cutting edge.     

基于BAS-BP模型的轴承剩余使用寿命预测

为有效评估轴承退化趋势,提高设备健康管理的智能化,提出一种基于BAS-BP模型的轴承剩余使用寿命预测方法。提取轴承全生命周期振动信号的时域和频域特征,构建18维退化特征;为提高神经网络的预测精度,采用天牛须搜索算法对初始权重和阈值进行优化,建立BAS-BP预测模型;通过在公开数据集上验证该模型的有效性。结果表明：所提模型可对轴承剩余寿命进行有效预测且精度较高。     

Drilling wear detection and classification using vibration signals and artificial neural network

In automated flexible manufacturing systems the detection of tool wear during the cutting process is one of the most important considerations. This study presents a comparison between several architectures of the multi-layer feed-forward neural network with a back propagation training algorithm for tool condition monitoring (TCM) of twist drill wear. The algorithm utilizes vibration signature analysis as the main and only source of information from the machining process. The objective of the proposed study is to produce a TCM system that will lead to a more efficient and economical drilling tool usage. Five different drill wear conditions were artificially introduced to the neural network for prediction and classification. The experimental procedure for acquiring vibration data and extracting features in both the time and frequency domains to train and test the neural network models is detailed. It was found that the frequency domain features, such as the averaged harmonic wavelet coefficients and the maximum entropy spectrum peaks, are more efficient in training the neural network than the time domain statistical moments. The results demonstrate the effectiveness and robustness of using the vibration signals in a supervised neural network for drill wear detection and classification.