Statistical process control using optimized neural networks: A case study

The most common statistical process control (SPC) tools employed for monitoring process changes are control charts. A control chart demonstrates that the process has altered by generating an out-of-control signal. This study investigates the design of an accurate system for the control chart patterns (CCPs) recognition in two aspects. First, an efficient system is introduced that includes two main modules: feature extraction module and classifier module. In the feature extraction module, a proper set of shape features and statistical feature are proposed as the efficient characteristics of the patterns. In the classifier module, several neural networks, such as multilayer perceptron, probabilistic neural network and radial basis function are investigated. Based on an experimental study, the best classifier is chosen in order to recognize the CCPs. Second, a hybrid heuristic recognition system is introduced based on cuckoo optimization algorithm (COA) algorithm to improve the generalization performance of the classifier. The simulation results show that the proposed algorithm has high recognition accuracy.     

Tolerance allocation using neural networks

The purpose of tolerance allocation is to find a combination of tolerances to individual components such that the assembly tolerance constraint is met with minimum production cost. There are several methods available to allocate or apportion the assembly tolerance to individual parts. Some of the most common methods use linear programming, Lagrange multipliers, exhaustive search and statistical distributions. However, all the methods have some limitations. Moreover, most of these methods cannot account for the frequently observed mean shift phenomena that occur owing to tool wear, chatter, bad coolant, etc. This paper presents a neural networks-based approach for the tolerance allocation problem considering machines' capabilities, and mean shifts. The network is trained using the backpropagation learning method and used to predict individual part tolerances.     

A vision system for surface roughness assessment using neural networks

In this study we use machine vision to assess surface roughness of machined parts produced by the shaping and milling processes. Machine vision allows for the assessment of surface roughness without touching or scratching the surface, and provides the flexibility for inspecting parts without fixing them in a precise position. The quantitative measures of surface roughness are extracted in the spatial frequency domain using a two-dimensional Fourier transform. Two artificial neural networks, which take roughness features as the input, are developed to determine the surface roughness. The first network is for test parts placed in a fixed orientation, which minimises the deviation of roughness measures. The second network is for test parts placed in random orientations, which gives maximum flexibility for inspection tasks. Experimental results have shown that the proposed roughness features and neural networks are efficient and effective for automated classification of surface roughness.     

Diagnosis of roller bearing defects using neural networks

A new approach for the diagnosis of bearing defects has been utilised. Artificial neural networks (ANN) were employed for the diagnosis of various kinds of bearing defects. The features selected for this purpose were: the average of the top five values of amplitude in the high-frequency region (5 kHz-22 kHz), the peak value of the amplitude in the high-frequency region, the average of the top five values in the prime spike region (340 Hz-3262 Hz), the autocorrelation function in the prime spike region, the autocorrelation function in the high-frequency region, and the cepstrum function in the high-frequency region.Data were collected using a data acquisition system. The data collected for the five different defective roller bearings as well as for a normal bearing were used to train neural networks. The trained neural networks were used for the diagnosis of roller bearings. Various neural network sizes were used. It was found that neural networks were able to distinguish a normal bearing from defective bearings with 100% reliability. Furthermore, roller bearings can be classified into six different states with a success rate of up to 94%.     

Control of chemical processes using neural networks: implementation in a plant for xylose production

This work demonstrates the use of artificial intelligence for control of xylose reactor performance in a paper factory. Two types of neural networks are used, a perceptron for the temperature controller and an adaptive formulation for the noise filter. The results show an improvement in the temperature stabilization time with respect to a classic PID control.     

Development of inferential measurements using neural networks

In many industrial processes, the most desirable variables to control are measured infrequently off-line in a quality control laboratory. In these situations, use of advanced control or optimization techniques requires use of inferred measurements generated from correlations. For well-understood processes, the structure of the correlation as well as the choice of inputs may be known a priori. However, many industrial processes are too complex and the appropriate form of the correlation and choice of input measurements are not obvious. Here, process knowledge, operating experience, and statistical methods play an important role in development of correlations. This paper describes a systematic approach to the development of nonlinear correlations for inferential measurements using neural networks. A three-step procedure is proposed. The first step consists of data collection and preprocessing. Next, the process variables are subjected to simple statistical analyses to identify a subset of measurements to be used in the inferential scheme. The third step involves generation of the inferential scheme. We demonstrate the methodology by inferring the ASTM 95% endpoint of a petroleum product using data from a domestic US refinery.     

A formulation for collision identification and distance calculation in motion planning using neural networks

The collision identification and object-to-object distance calculation play an important role in the motion planning for robots and manufacturing facilities. A formulation for collision identification and distance calculation in motion planning, using neural networks, is presented. The method calculates the distances between the vertices of an object and the given polyhedral obstacles using the modified Hamming net. This formulation is derived from the homogeneous geometric transformations. The method can be used to identify collision between the vertices of a moving object and the obstacles, to calculate the distance and interference between the moving object and the obstracle, and to find the optimal direction for collision removal. The parallel computation formulation is simple in form, and can be extended to line-to-object and object-to-object collision identification and distance calculation. The method can considerably decrease required computation time, and has the potential for being applied to on-line trajectory planning.     

A study on fatigue damage modeling using neural networks

Fatigue crack growth and life have been estimated based on established empirical equations. In this paper, an alternative method using artificial neural network (ANN) -based model developed to predict fatigue damages simultaneously. To learn and generalize the ANN, fatigue crack growth rate and life data were built up using in-plane bending fatigue test results. Single fracture mechanical parameter or nondestructive parameter can’t predict fatigue damage accurately but multiple fracture mechanical parameters or nondestructive parameters can. Existing fatigue damage modeling used this merit but limited real-time damage monitoring. Therefore, this study shows fatigue damage model using backpropagation neural networks on the basis of X-ray half breadth ratioB/B ₀ , fractal dimensionD _f and fracture mechanical parameters can estimate fatigue crack growth rateda/dN and cycle ratioN/N _f at the same time withinengineering limit error (5%).     

Adaptive nonlinear control using input normalized neural networks

An adaptive feedback linearization technique combined with the neural network is addressed to control uncertain nonlinear systems. The neural network-based adaptive control theory has been widely studied. However, the stability analysis of the closed-loop system with the neural network is rather complicated and difficult to understand, and sometimes unnecessary assumptions are involved. As a result, unnecessary assumptions for stability analysis are avoided by using the neural network with input normalization technique. The ultimate boundedness of the tracking error is simply proved by the Lyapunov stability theory. A new simple update law as an adaptive nonlinear control is derived by the simplification of the input normalized neural network assuming the variation of the uncertain term is sufficiently small.     

Adaptive control of nonlinear system using online error minimum neural networks

In this paper, a new learning algorithm named OEM-ELM (Online Error Minimized-ELM) is proposed based on ELM (Extreme Learning Machine) neural network algorithm and the spreading of its main structure. The core idea of this OEM-ELM algorithm is: online learning, evaluation of network performance, and increasing of the number of hidden nodes. It combines the advantages of OS-ELM and EM-ELM, which can improve the capability of identification and avoid the redundancy of networks. The adaptive control based on the proposed algorithm OEM-ELM is set up which has stronger adaptive capability to the change of environment. The adaptive control of chemical process Continuous Stirred Tank Reactor (CSTR) is also given for application. The simulation results show that the proposed algorithm with respect to the traditional ELM algorithm can avoid network redundancy and improve the control performance greatly.     

Identifying three-phase induction motor faults using artificial neural networks

This paper presents an artificial neural network (ANN) based technique to identify faults in a three-phase induction motor. The main types of faults considered are overload, single phasing, unbalanced supply voltage, locked rotor, ground fault, over-voltage and under-voltage. Three-phase currents and voltages from the induction motor are used in the proposed approach. A feedforward layered neural network structure is used. The network is trained using the backpropagation algorithm. The trained network is tested with simulated fault current and voltage data. Fault detection is attempted in the no fault to fault transition period. Off-line testing results on a 3 HP induction motor model show that the proposed ANN based method is effective in identifying various types of faults.     

在VC＋＋开发环境下的神经网络程序设计

目前神经网络程序大多是在C环境下编写的,可操作性、人机交互性与通用性较差,针对这一现象,这里提出了采用VC＋＋实现的神经网络程序。     

Solving multiple response optimisation problems using adaptive neural networks

This paper describes an integrated methodology using experimental designs and neural networks technologies for solving multiple response problems. This new methodology consists of an experiment reference template for designing and collecting training data samples and a parallel distributed computational adaptive neural network system to provide a powerful tool for data modelling, guiding experimentation and empirical investigations. While the experiment reference template is for determining the measurements to adopt in order to extract maximum information within minimum experimental efforts, the adaptive neural network provides a nonlinear multivariate data-fitting algorithm for analysing the results of the experimental design and providing decision support. This integrated methodology is used to model and optimise a multiple response metal inert gas (MIG) welding process. The neural network is trained with optimum welding experimental data, tested and compared in an actual welding environment in terms of weld quality. The relevant data is established using experimental design methods and is highlighted in the case study. The implementation for this case study was carried out using a semi-automatic welding facility, to mass weld a 20 in.×0.438 in. pin/box onto a 20 in.×0.5 in.×37 ft pipe (tubular drilling products), in an actual workshop which makes oilfield equipment. The entire range of welding combination that the process might be subject to during actual welding operations is included to study the weld quality.     

Dimensional and angular measurements using least squares and neural networks

In this study a machine vision approach is developed for dimensional and angular measurements of manufactured components comprising straight line segments. We aim at the measurements of distance between two parallel lines and angle between two intersecting lines using both least mean square (LMS) and artificial neural network (ANN) techniques. LMS models estimate the line parameters based on the sum of squared perpendicular distances, rather than the vertical distances, between the observed data points and the line. A set of 23 gauge blocks of varying sizes is used to evaluate the performance of the LMS line estimators. Experimental results show that the measurement errors of the LMS models are affected by the line length and orientation of digital images. ANN techniques are, therefore, used to adjust the measurement errors resulting from the LMS models. Two back-propagation neural networks are developed, one for measuring the distance between two parallel lines, and the other for measuring the angle between two intersecting lines. Experimental results show that the ANNs are very effective for correcting the measurement errors regardless of line lengths and orientations of digital images. A 90% improvement in measurement accuracy for the ANN compared to the LMS was achieved. By using the ANNs, the measurement accuracy and flexibility in manufacturing applications can be significantly improved.     

Modeling of hysteresis in piezoelectric actuators using neural networks

For the application of neural networks to the approximation of hysteresis which is characterized of multi-valued mapping and non-smooth nonlinearities, a novel modeling technique based on a transformation of one-to-one mapping is proposed in this paper. In this method, a special hysteretic operator is introduced to describe the change tendency of the hysteresis with regard to its input. Then an expanded input space is constructed for hysteresis with the introduction of such hysteretic operator, on which the multi-valued hysteresis is decomposed into a one-to-one mapping. Thus, neural networks model for hysteresis is derived, avoiding the calculation of the gradient of hysteresis. Subsequently, for approximation of rate-dependent hysteresis in piezoelectric actuators which is caused by the dynamic voltage excitations, a hybrid model, i.e. the dynamic extension of the proposed neural hysteresis submodel is developed. In the model, a linear dynamic block is introduced in series with the proposed neural model to allow for rate-dependent dynamics of the piezoelectric actuator simultaneously. Also the corresponding optimization algorithm by use of the modified Levenberg–Marquarqt (MLM) method is given. Finally, the experimental validation results of applying both the proposed neural hysteresis model and hybrid model to a piezoelectric actuator are presented.     

On-line surface roughness recognition system using artificial neural networks system in turning operations

In modern manufacturing environments, the quality assurance of machined parts has attracted great attention from manufacturers. The surface roughness of a workpiece is one of the most important factors to consider. The need for developing a surface recognition system that is able to replace stylus-style surface measuring systems has increased to improve the efficiency of production. In this research an on-line surface recognition system was developed based on artificial neural networks (OSRR-ANN) using a sensing technique to monitor the effect of vibration produced by the motions of the cutting tool and workpiece during the cutting process. Different combinations of cutting conditions were conducted to develop an OSRR system for a lathe. In order to determine the direction of the vibration which most significantly affects surface roughness, a triaxial accelerometer was employed. Three directional vibrations which were detected simultaneously by the accelerometer were analyzed using a statistical method. The radial direction vibration was found to be the most significant vibration in turning operations. The accuracy of the developed systems showed that the developed system could predict surface roughness efficiently. The developed system not only proposes a surface recognition system which is alternative to that using a traditional measurement instrument, but also provides an on-line surface recognition system for turning operations.     

神经元网络在甘蔗收获机智能设计系统评价体系中的应用

针对甘蔗收获机智能设计系统中评价体系的缺乏智能性、对设计专家经验知识要求较高以及评价准确度较低的缺陷,阐述了将神经元网络嵌入智能设计系统评价体系,以提高该体系的可靠性,并探讨了神经元网络在评价体系中的应用形式和模式转换,建立了对甘蔗收获机中重要部件性能进行评价的神经元网络。     

Improved delay-dependent stability result for neural networks with time-varying delays

This paper is concerned with a new Lyapunov-Krasovskii functional (LKF) approach to the stability for neural networks with time-varying delays. The LKF has two features: First, it can make full use of the information of the activation function. Second, it employs the information of the maximal delayed state as well as the instant state and the delayed state. When estimating the derivative of the LKF we employ a new technique that has two characteristics: One is that Wirtinger-based integral inequality and an extended reciprocally convex inequality are jointly employed; the other is that the information of the activation function is used as much as we can. Based on Lyapunov stability theory, a new stability result is obtained. Finally, three examples are given to illustrate the stability result is less conservative than some recently reported ones.     

Neural networks for process control and optimization: two industrial applications

The two most widely used neural models, multilayer perceptron (MLP) and radial basis function network (RBFN), are presented in the framework of system identification and control. The main steps for building such nonlinear black box models are regressor choice, selection of internal architecture, and parameter estimation. The advantages of neural network models are summarized: universal approximation capabilities, flexibility, and parsimony. Two applications are described in steel industry and water treatment, respectively, the control of alloying process in a hot dipped galvanizing line and the control of a coagulation process in a drinking water treatment plant. These examples highlight the interest of neural techniques, when complex nonlinear phenomena are involved, but the empirical knowledge of control operators can be learned.     

Wear volume prediction with artificial neural networks

In this study, the potential of artificial neural network techniques to predict and analyze the wear behaviour of short fibre reinforced polymeric bearing materials was investigated. Artificial neural networks have been recently introduced into tribology by Jones et al. (1997) [Jones, Jansen and Fusaro, Preliminary investigation of neural network techniques to predict tribological properties. Tribology Transactions 1997;40(2):312]. Their work is extended here in three directions: 1. A higher number of input variables characterizing the materials and the experimental conditions is used (10 instead of 6). Based on a principal component analysis, correlations in the input vector are identified and used to reduce its dimensionality. 2. A statistical technique is used to improve the predictive capabilities of the artificial neural network (Bayesian regularization instead of early stopping used by Jones et al.). This technique also automatically identifies the optimal size of the artificial neural network, which was determined experimentally by the other group of authors. 3. The quality of the predictions is evaluated. While Jones et al. found values well above 0.9 for the coefficient of determination on a fixed set of test data, it is shown here based on a dataset which was obtained at the IVW in fretting experiments for different composites that the average results are not so good if a large number of randomly chosen test data sets is considered. However, it is argued that the results are still reasonable when compared with the substantial wear volume measurement error. Also, improved results can be expected in the future from a further optimization of the network construction as well as from an increasing availability of measurement data that can be used to train the network.