Modelling process mean and variation with MLP neural networks

Most industrial processes are intrinsically noisy and non-deterministic. To date, most multilayer perceptron (MLP)-based process models were established for process mean only. This paper proposes an approach to modelling the mean and variation of a non-deterministic process simultaneously using a MLP network. The input neurons consist of process variables and one additional neuron for the Z value. The corresponding output responses are calculated based on , s_p/√k.

The process variance s_p² is determined by pooling the individual sample variances for k experimental conditions. Each sample variance is calculated from the replicated data. The effects of a number of hidden neurons and learning algorithms are studied. Two learning algorithms are applied. They are the back-propagation with momentum (BPM) and Fletcher-Reeves (FR) algorithms. The effectiveness of the proposed approach is tested with a fictitious process and an actual manufacturing process. The test results are provided and discussed.     

A chaotic cutting process and determining optimal cutting parameter values using neural networks

A model of an orthogonal cutting system is described as an elastic structure deformable in two directions. In the system, a cutting force is generated by material flow against the tool. Nonlinear dependency of the cutting force on the cutting velocity can cause chaotic vibrations of the cutting tool which influence the quality of a manufactured surface. The intensity and the characteristics of vibrations are determined by the values of the cutting parameters. The influence of cutting depth on system dynamics is described by bifurcation diagrams. The properties of oscillations are illustrated by the time dependence of tool displacement, the corresponding frequency spectra and phase portraits. The corresponding strange attractors are characterized by correlation dimension. The vibrations are characterized by the maximum Lyapunov exponent. The manufactured surface at the first cut is taken as the incoming surface in the second cut, thus incorporating the influence of the rough surface in the model. Again, bifurcation diagrams, the correlation dimension and the maximum Lyapunov exponent are employed to describe the effects of parametrical excitation on the cutting dynamics. A cost function is defined which describes the dependence of the cutting performance on cutting depth. The cost function is empirically modeled using a self-organizing neural network. A conditional average estimator is applied to determine the optimal value of the cutting depth applicable as a control variable of the cutting process.     

基于CMAC神经网络的激光切割加工工艺参数的选取

提出了一种用神经网络技术构造激光切割加工工艺参数模型和建立加工参数自动选取系统的方法。利用CMAC神经网络算法，产生一个工艺参数自动选择器，将它嵌入到激光切割加工控制系统中，尤其适用于实时控制选取和改变加工工艺参数，进一步提高控制系统的智能和加工的柔性。     

Context-free and context-sensitive dynamics in recurrent neural networks

Continuous-valued recurrent neural networks can learn mechanisms for processing context-free languages. The dynamics of such networks is usually based on damped oscillation around fixed points in state space and requires that the dynamical components are arranged in certain ways. It is shown that qualitatively similar dynamics with similar constraints hold for aⁿbⁿcⁿ , a context-sensitive language. The additional difficulty with aⁿbⁿcⁿ , compared with the context-free language aⁿbⁿ , consists of 'counting up' and 'counting down' letters simultaneously. The network solution is to oscillate in two principal dimensions, one for counting up and one for counting down. This study focuses on the dynamics employed by the sequential cascaded network, in contrast to the simple recurrent network, and the use of backpropagation through time. Found solutions generalize well beyond training data, however, learning is not reliable. The contribution of this study lies in demonstrating how the dynamics in recurrent neural networks that process context-free languages can also be employed in processing some context-sensitive languages (traditionally thought of as requiring additional computation resources). This continuity of mechanism between language classes contributes to our understanding of neural networks in modelling language learning and processing.     

Stack-like and queue-like dynamics in recurrent neural networks

What dynamics do simple recurrent networks (SRNs) develop to represent stack-like and queue-like memories? SRNs have been widely used as models in cognitive science. However, they are interesting in their own right as non-symbolic computing devices from the viewpoints of analogue computing and dynamical systems theory. In this paper, SRNs are trained on two prototypical formal languages with recursive structures that need stack-like or queue-like memories for processing, respectively. The evolved dynamics are analysed, then interpreted in terms of simple dynamical systems, and the different ease with which SRNs aquire them is related to the properties of these simple dynamical systems. Within the dynamical systems framework, it is concluded that the stack-like language is simpler than the queue-like language, without making use of arguments from symbolic computation theory.     

基于分子动力学结合神经网络的Au表面能计算方法

利用嵌入原子模型,采用分子动力学方法计算了贵金属Au低指数晶面及部分简单高指数晶面的表面能.同时,采用Levenberg-Marquardt算法,建立了Au表面能的BP神经网络模型;结合分子动力学模型的计算数据,通过大量数据的自学习训练,完成神经网络模型对Au高指数晶面表面能的预测.计算结果表明:该方法具有较高的预测精度,能正确预言低指数晶面表面能的排序,Au各晶面的表面能随其晶面与(111)密排面夹角的增大呈现先增大后减小的特点.     

Accurate 3-D cutting force prediction using cutting condition independent coefficients in end milling

The instantaneous uncut chip thickness and specific cutting forces have a significant effect on predictions of cutting force. This paper presents a systematic method for determining the coefficients in a three-dimensional mechanistic cutting force model—the cutting force coefficients (two specific cutting forces, chip flow angle) and runout parameters. Some existing models have taken the approach that the cutting force coefficients vary as a function of cutting conditions or cutter rotation angle. This paper, however, considers that the coefficients are affected only by the uncut chip thickness. The instantaneous uncut chip thickness is estimated by following the movement of the position of the center of a cutter. To consider the size effect, the present method derives the relationship between the re-scaled uncut chip thickness and the normal specific cutting force, K_n with respect to the cutter rotation angle, while the other two coefficients—frictional specific cutting force, K_f and chip flow angle, θ_c—remain constant. Subsequently, all the coefficients can be obtained, irrespective of cutting conditions. The proposed method was verified experimentally for a wide range of cutting conditions, and gave significantly better predictions of cutting forces.     

Prediction of cutting force for self-propelled rotary tool using artificial neural networks

In this paper, a cutting force model for self-propelled rotary tool (SPRT) cutting force prediction using artificial neural networks (ANN) has been introduced. The basis of this approach is to train and test the ANN model with cutting force samples of SPRT, from which their neurons relations are gradually extracted out. Then, ANN cutting force model is achieved by obtaining all weights for each layer. The inputs to the model consist of cutting velocity V, feed rate f, depth of cut a_p and tool inclination angle λ, while the outputs are composed of thrust force F_x, radial force F_y and main cutting force F_z. It significantly reduces the complexity of modeling for SPRT cutting force, and employs non-structure operator parameters more conveniently. Considering the disadvantages of back propagation (BP) such as the convergence to local minima in the error space, developments have been achieved by applying hybrid of genetic algorithm (GA) and BP algorithm hence improve the performance of the ANN model. Validity and efficiency of the model were verified through a variety of SPRT cutting samples from our experiment tested in the cutting force model. The performance of the hybrid of GA–BP cutting force model is fairly satisfactory.     

Modelling of dynamic cutting coefficients in three-dimensional cutting

A new theory to determine the dynamic cutting coefficients from steady state cutting data for three dimensional cutting has been developed. It is based on direct measurements of cutting forces, without any hypothesis relating to the steady state cutting. The experimental results show fairly good coincidence with the theoretical prediction of the stability limit.     

Modelling and prediction of cut shape for plasma arc cutting based on artificial neural network

Abstract

An artificial neural network approach for the modelling of plasma arc cutting processes is introduced. Neural network models have been proposed for predicting the cut shape and estimating the special cutting variables. The implementation of artificial neural networks in the modelling of cutting processes is discussed in detail. The performance of the neural networks in modelling is presented and evaluated using actual cutting data. Moreover, prediction applications of the above neural network models are described for various cutting conditions. It is shown that estimated results based on the proposed models agree well with experimental data; the neural network models yield good prediction results over the entire range of cutting process parameters spanned by the training data. The testing and prediction results show the effectiveness and satisfactory prediction accuracy of the artificial neural network modelling. The developed models are applicable to carbon steel.     

Modelling weld bead geometry using neural networks for GTAW of austenitic stainless steel

Abstract

The authors analyse the importance of different weld control parameters on the weld pool geometry of gas tungsten arc welding using an online feature selection technique that suggests weld voltage and vertex–angle pair as more important than the weld voltage and torch speed pair. Using the selected features multi layer perception and radial basis function networks are developed for prediction of bead width, penetration depth, and bead area. With cross-validation the authors have extensively studied the performance of composite models (one model for all outputs) and individual models (one model for each output). The individual models are found to work better than composite models. Usually, radial basis function networks are found to work better than the multi layer perception networks. To assess the influence of weld control parameters the authors have studied the performance of both networks using different combination of inputs. Overall, the performance of the proposed models is found to be quite satisfactory.     

An in-process surface recognition system based on neural networks in end milling cutting operations

An in-process based surface recognition system to predict the surface roughness of machined parts in the end milling process was developed in this research to assure product quality and increase production rate by predicting the surface finish parameters in real time. In this system, an accelerometer and a proximity sensor are employed as in-process surface recognition sensors during cutting to collect the vibration and rotation data, respectively. Using spindle speed, feed rate, depth of cut, and the vibration average per revolution (VAPR) as four input neurons, an artificial neural networks (ANN) model based on backpropagation was developed to predict the output neuron-surface roughness R_a values. The experimental results show that the proposed ANN surface recognition model has a high accuracy rate (96–99%) for predicting surface roughness under a variety of combinations of cutting conditions. This system is also economical, efficient, and able to be implemented to achieve the goal of in-process surface recognition by retrieving the weightings (which were generated from training and testing by the artificial neural networks), predicting the surface roughness R_a values while the part is being machined, and giving feedback to the operators when the necessary action has to be taken.     

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.     

Modelling and simulation of micro-milling cutting forces

The paper presents a new approach for predicting micro-milling cutting forces using the finite element method (FEM). The trajectory of the tool and the uncut chip thickness for different micro-milling parameters (cutting tool radius, feed rate, spindle angular velocity and number of flutes) are determined and used for predicting the cutting forces in micro-milling. The run-out effect is also taken into account. An orthogonal FE model is developed. A number of FE analyses (FEA) are performed at different uncut chip thicknesses (0–20 μm) and velocities (104.7–4723 mm/s) for AISI 4340 steel. Based on the FE results, the relationship between the cutting forces, uncut chip thickness and cutting velocity has been described by a non-linear equation proposed by the authors. The suggested equation describes the ploughing and shearing dominant cutting forces. The micro-milling cutting forces have been determined by using the predicted forces from the orthogonal cutting FE model and the calculated uncut chip thickness. Different feed rates and spindle angular velocities have been investigated and compared with experimentally obtained results. The predicted and the measured forces are in very good agreement.     

金刚石探针划切石墨烯的分子动力学模拟

石墨烯被认为是当前最有发展前景的二维纳米材料,拥有优越的物化特性和广泛的应用前景,但石墨烯没有带隙,极大限制了其在电子领域的应用,精密切割能为石墨烯打开一定的带隙。本文采用分子动力学模拟方法对石墨烯进行划切,分析金刚石探针沿不同方向划切石墨烯的微观形貌,研究有无基底、不同切割方向等参数对石墨烯划切边缘能量及划切力等的影响规律。模拟结果表明:金刚石划切石墨烯具有各向异性特征,切割边缘粗糙,没有明显armchair型边缘或zigzag型边缘特征。      

Extension of basic geometric analysis of 3-D chip forms in metal cutting to chips with obstacle-induced deformation

In machining, chips are known to break mainly because of obstacle-induced deformation. Recently, the present authors had reported on a new and basic geometric analysis of 3-D chips in the absence of deformation after separation from the tool rake face. This paper continues the analysis to cover the full lifecycle of chips subjected to obstacle-induced deformation. The main contributions of this paper are the identification of the geometric properties that are likely to be preserved during obstacle-induced chip deformation and their implications, and the utilisation of these properties to obtain insights concerning the geometry of the tool–chip contact area. The new theoretical findings are verified against empirical data obtained through manual deformation of chips, video recordings of the development of obstructed chips, and the use of a specially prepared grooved tool that imposes a predetermined side-curl on the chip.     

基于神经网络的7055铝合金流变应力模型和加工图

在Geeble-1500热模拟机上对7055铝合金进行热压缩试验,基于热压缩试验数据,建立流变应力的反向传播(BP)神经网络预测模型和加工图。结果表明:用人工神经网络能更精确地预测热压缩过程中的流变应力,预测精度明显高于线性经验公式的;通过预测模型可以获得样本数据值范围内的非样本数据变形条件下的流变应力,其预测结果充分反映该合金的高温变形特征;在本实验条件下,7055铝合金在高温变形时存在一个失稳区,即变形温度在实验温度范围内应变速率为0.025s-1以上的区域;在375～425℃的范围内,应变速率小于0.001s-1的区域,最大功率耗散系数为0.45;EBSD技术分析表明在安全区发生部分动态再结晶。利用加工图确定了热变形时的流变失稳区,并且获得了试验参数范围内热变形的最佳工艺参数,其热加工温度为350-430℃低应变速率区。     

Modelling of orthogonal cutting processes with the method of smoothed particle hydrodynamics

In the last decades, mesh-free methods for simulating various cutting processes have been used very widely as they can eliminate numerical problems in the simulation of material failure and large plastic deformations. This paper deals with the results from modelling the orthogonal cutting of AISI 1045 steel using smoothed particle hydrodynamics (SPH) method. Moreover, it is determined how the parameters of the SPH solver such as initial smoothing length, initial particle density and coefficient for the timestep increase affect the prediction error for the values of cutting force and chip compression ratio as well as computing time. The optimum values of the SPH solver parameters are determined by minimising an objective function. The best balance between the prediction error of machining variables and computing time is achieved for an initial particle density of 40 μm and a coefficient for the timestep increase of 0.4.     

3-D or not 3-D

When developing mathematical models of physical phenomena, physical scientists have been limited to two-dimensional solutions. In some cases, those solutions may be qualitatively correct, but in many others, the solutions are far from reality. This paper presents three examples of the limitations of two-dimensional modeling :diffusion-limited electrolytic reduction of a liquidproduct from a liquid electrolyte, morphological development of a solid dendrite, and behavior of a two-phase fluid under shear.     

On-line metal cutting tool condition monitoring.: II: tool-state classification using multi-layer perceptron neural networks

Excessive wear on cutting tools give rise to distortions in dimension of manufactured components, sometimes increasing scrapped levels thereby incurring additional costs. Methods for detecting and monitoring the wear on a cutting tool is therefore crucial in most metal cutting processes and several research efforts have striven to develop on-line tool condition monitoring systems. This paper describes an experimental and analytical method for one such technique involving the use of three mutually perpendicular components of the cutting forces (static and dynamic) and vibration signature measurements. The ensuing analyses in time and frequency domains showed some components of the measured signals to correlate well to the accrued tool wear.