Integration of artificial neural network and simulated annealing algorithm to optimize deep drawing process

Deep drawing is characterized by very complicated deformation affected by the process parameter values including die geometry, blank holder force, material properties, and frictional conditions. The aim of this study is to model and optimize the deep drawing process for stainless steel 304 (SUS304). To achieve the purpose, die radius, punch radius, blank holder force, and frictional conditions are designated as input parameters. Thinning, as one of the major failure modes in deep drawn parts, is considered as the process output parameter. Based on the results of finite element (FE) analysis, an artificial neural network (ANN) has been developed, as a predictor, to relate important process parameters to process output characteristics. The proposed feed forward back propagation ANN is trained and tested with pairs of input/output data obtained from FE analysis. To verify the FE model, the results obtained from the FE model were compared with those of several experimental tests. Afterward, the ANN is integrated into a simulated annealing algorithm to optimize the process parameters. Optimization results indicate that by selecting the proper process parameter settings, uniform wall thickness with minimum thinning can be achieved.     

Condition monitoring of planetary gearbox by hardware implementation of artificial neural networks

The hazards of planetary gearboxes’ failures are the most crucial in the machinery which directly influence human safety like aircrafts. But also in an industry their damages can cause the large economic losses. Planetary gearboxes are used in wind turbines which operate in non-stationary conditions and are exposed to extreme events. Also bucket-wheel excavators are equipped with high-power gearboxes that are exposed to shocks. Continuous monitoring of their condition is crucial in view of early failures, and to ensure safety of exploitation. Artificial neural networks allow for a quick and effective association of the symptoms with the condition of the machine. Extensive research shows that neural networks can be successfully used to recognize gearboxes’ failures; they allow for detection of new failures which were not known at the time of training and can be applied for identification of failures in variable-speed applications. In a majority of the studies conducted so far neural networks were implemented in the software, but for dedicated engineering applications the hardware implementation is being used increasingly, due to high efficiency, flexibility and resistant to harsh environmental conditions. In this paper, a hardware implementation of an artificial neural network designed for condition monitoring of a planetary gearbox is presented. The implementation was done on a Field Programmable Gate Array (FPGA). It is characterized by much higher efficiency and stability than the software one. To assess condition of a gearbox working in non-stationary conditions and for chosen failure modes, a signal pre-processing algorithm based on filtration and estimation of statistics from the vibration signal was used. Additionally, the rewards-punishments training process was improved for a selected neural network, which is based on a Learning Vector Quantization (LVQ) algorithm. Presented classifier can be used as an independent diagnostic system or can be combined with traditional data acquisition systems using FPGAs.     

Indirect model for roughness in rough honing processes based on artificial neural networks

In the present paper an indirect model based on neural networks is presented for modelling the rough honing process. It allows obtaining values to be set for different process variables (linear speed, tangential speed, pressure of abrasive stones, grain size of abrasive and density of abrasive) as a function of required average roughness Ra. A multilayer perceptron (feedforward) with a backpropagation (BP) training system was used for defining neural networks. Several configurations were tested with different number of layers, number of neurons and type of transfer function. Best configuration for the network was searched by means of two different methods, trial and error and Taguchi design of experiments (DOE). Once best configuration was found, a network was defined by means of trial and error method for roughness parameters related to Abbott–Firestone curve, Rk, Rpk and Rvk.     

A smart force platform using artificial neural networks

The human body may interact with structures and these interactions are developed through the application of contact forces, for instance when walking. The aim of this paper is to propose a new methodology using Artificial Neural Network (ANN) for calibrating a force platform in order to reduce the uncertainties in the values of estimated vertical Ground Reaction Force and the positioning of the applied force in the human gait. Force platforms have been used to evaluate the pattern of human applied forces and to fit models for the interaction between pedestrians and structures. Linear relation assumptions between input and output are common in traditional Least Mean Square methods used in calibration. Some discrepancies due to nonlinearities in the experimental setup (looseness, wear, support settlements, electromagnetic noise, etc.) may harm the overall fitting. Literature has shown that nonlinear models, like ANN, can better handle this. During the calibration, the input data to the ANN were the reference voltages applied to the Wheatstone bridge, while the output data were the values of the standard weights applied in the force platform in defined sites. Supervised training based on k-fold cross validation was used to check the ANN generalization. The use of ANN shows significant improvements for the measured variables, leading to better results for predicted values with low uncertainty when compared to the results of a simple traditional calibration using Least Mean Squares.     

Process modelling and optimisation using artificial neural networks and gradient search method

Process modelling refers to the development of a process model that serves to provide the input-output relationship of a process, while process optimisation provides the optimum operating conditions of a process for a high-yield, low cost and robust operation. Normally, process modelling is a starting point of process optimisation. In this paper, a method of integrating artificial neural networks with a gradient search method for process modelling and optimisation is presented. Artificial neural networks are used to develop process models while a gradient search method is used in process optimisation. Application of the method to the modelling and optimisation of epoxy dispensing for microchip encapsulation is described. Results of the validation tests indicate that good quality of encapsulation can be obtained based on the proposed method.     

Feedrate optimization based on hybrid forward-reverse mappings of artificial neural networks for five-axis milling

With recent advances in five-axis milling technology, feedrate optimization methods have shown significant effects in regard to enhancing milling productivity, especially when machining complex surface parts. The existing study is aimed at calculating the optimal feedrate values through modeling milling processes. However, due to the complexity of five-axis milling processes, optimization efficiency is the bottleneck of applying them in practice. This paper proposes a novel milling process optimization method based on hybrid forward-reverse mappings (HFRM) of artificial neural networks. The feedrate values are directly used as the outputs of network mappings. Three kinds of artificial neural networks are compared to determine the one with the highest accuracy and the best training efficiency. The study shows that with the collected datasets, the trained Levenberg-Marquardt back-propagation network (LMBPN) could predict feedrate values more precisely than other alternatives. Compared with previous methods, this HFRM-based optimization method is more adept in the area of parameter adjustment because as it has the advantages of high precision and much less calculation time. Combining other multiple milling constraints, an optimization system is developed for five-axis milling processes. The optimized results could be directly used to modify a cutter location (CL) file. A typical milling case was provided to verify the optimization performance of this method, which was found to be effective and reliable.     

New correlations for better monitoring the all-oil mud rheology by employing artificial neural networks

The rheological properties of the drilling fluid are crucial to the success of the drilling project. The traditional mud experiments normally performed by the mud engineers provide rheological data with a small resolution. Monitoring higher-resolution rheological properties is particularly important for all-oil mud because it is widely used with problematic drilled formations. The design and monitoring of the drilling fluid rheology is a critical issue for drilling, and therefore, this paper is a contribution to the effort to completely automate the process of highly accurate and real-time recording of the rheological mud properties. This paper aims to develop intelligent predictive models for the mud rheological properties using artificial neural networks [ANN] by linking the high-frequency mud parameters such as fluid density or mud weight [MWT] and Marsh funnel viscosity [MFV] with the rheological measurements of low frequency for drilling mud such as plastic viscosity [PV], yield point [YP], behavior indicator [n] and viscosity appearance [AV]. New empirical correlations have additionally been established to assess the rheological properties of water. In order to construct ANN models, data was obtained from 56 different wells during drilling operations of different drilling sections with various sizes. The data was fairly enough for building and testing the models as 369 data points were obtained. The models were optimized by trainlm which was the best training function and tansig was the best transfer function. 42 neurons in the hidden layer optimized AV and PV models where 34 neurons optimized all other rheological models [YP, n, R300, and R600]. ANN models presented good results as correlation coefficient [R] was 0.9 and an average absolute [AAPE] error of less than 8% for training and testing data sets. The new models were used to derive the empirical correlations for the estimation of rheological parameters. The empirical correlations were extracted to easily monitor the rheological properties of an all-oil mud system in real-time, which enables better control of the drilling activity by maintaining rheological properties at optimal values as well as early detection of other problems that might require immediate interactions, including well control and stuck pipe.     

基于人工神经网络的轧机轧制力矩在线监测方法研究

轧制力矩过大是轧机主传动系统传动轴断裂的主要原因,传统的电机电流过载保护方式不能真实反映轧机主传动系统扭矩的动力学特性。利用人工神经网络的优点,建立基于人工神经网络的轧制力矩在线监测的数学模型。     

A novel algorithm based on hybridization of artificial immune system and simulated annealing for clustering problem

A hybrid clustering method is proposed in this paper based on artificial immune system and simulated annealing. An integration of simulated annealing and immunity-based algorithm, combining the merits of both these approaches, is used for developing an efficient clustering method. Tuning the parameters of method is investigated using Taguchi method in order to select the optimum levels of parameters. Proposed method is implemented and tested on three real datasets. In addition, its performance is compared with other well-known meta-heuristics methods, such as ant colony optimization, genetic algorithm, simulated annealing, Tabu search, honey-bee mating optimization, and artificial immune system. Computational simulations show very encouraging results in terms of the quality of solution found, the average number of function evaluations and the processing time required, comparing with mentioned methods.     

A simulated annealing for hybrid flow shop scheduling with multiprocessor tasks to minimize makespan

This paper studies a hybrid flow shop scheduling problem (hybrid FSSP) with multiprocessor tasks, in which a set of independent jobs with distinct processor requirements and processing times must be processed in a k-stage flow shop to minimize the makespan criterion. This problem is known to be strongly nondeterministic polynomial time (NP)-hard, thus providing a challenging area for meta-heuristic approaches. This paper develops a simulated annealing (SA) algorithm in which three decode methods (list scheduling, permutation scheduling, and first-fit method) are used to obtain the objective function value for the problem. Additionally, a new neighborhood mechanism is combined with the proposed SA for generating neighbor solutions. The proposed SA is tested on two benchmark problems from the literature. The results show that the proposed SA is an efficient approach in solving hybrid FSSP with multiprocessor tasks, especially for large problems.     

A study on three dimensional layout design by the simulated annealing method

Modern engineered products are becoming increasingly complicated and most consumers prefer compact designs. Layout design plays an important role in many engineered products. The objective of this study is to suggest a method to apply the simulated annealing method to the arbitrarily shaped three-dimensional component layout design problem. The suggested method not only optimizes the packing density but also satisfies constraint conditions among the components. The algorithm and its implementation as suggested in this paper are extendable to other research objectives. This paper was recommended for publication in revised form by Associate Editor Joo Ho Choi Seung-Ho Jang received his B.S. in mechanical engineering and M.S. in precision mechanical engineering from University of Han Yang in 1986 and 1988, respectively. He then went on to receive his Ph.D. in mechanical engineering from University of Tokyo in 1991. His main research areas are CAD/CAE, mechanical vibration, and optimal design.     

Due-date assignment in wafer fabrication using artificial neural networks

Due-date assignment (DDA) is the first important task of shop floor control in wafer fabrication. Due-date related performance is impacted by the quality of the DDA rules. Assigning order due dates and timely delivering the goods to the customer will enhance customer service and competitive advantage. A new methodology for lead-time prediction, artificial neural network (ANN) prediction is considered in this work. An ANN-based DDA rule combined with simulation technology and statistical analysis is developed. Besides, regression-based DDA rules for wafer fabrication are modelled as benchmarking. Whether neural networks can outperform conventional and regression-based DDA rules taken from the literature is examined. From the simulation and statistical results, ANN-based DDA rules perform a better job in due-date prediction. ANN-based DDA rules have a lower tardiness rate than the other rules. ANN-based DDA rules have better sensitivity and variance than the other rules. Therefore, if the wafer fab information is not difficult to obtain, the ANN-based DDA rule can perform better due-date prediction. The SFM_sep and JIQ in regression-based and conventional rules are better than the others.     

A neural network-based control scheme for monitoring start-up processes and short runs

Traditional control charts are commonly used as a monitoring tool in long-run processes. However, such control charts, due to the need for phase I analysis, are not suitable for start-up processes or short runs. Q control charts have been developed to help monitor start-up processes and short runs. In this article, a back propagation network is proposed for detecting a mean shift in start-up processes and short runs. In-control run length distribution of the control scheme is estimated using simulation study to provide information about the possibility of a false alarm within a specified number of observations. Performance of the proposed control scheme is assessed using different performance measures. It is shown numerically that the proposed control scheme outperforms the CUSUM of Q charts in detecting small to moderate mean shifts.     

Practical applications of artificial neural networks in chemical process development

Artificial neural networks (ANN) have the ability to map non-linear relationships without a-priori information about process or system models. This significant feature allows the network to “learn” the behavior of a system by example when it may be difficult or impractical to complete a rigorous mathematical solution. Recently ANN technology has been leaving the academic arena and placed in user-friendly software packages. This paper will offer an introduction to artificial neural networks and present a case history of two problems in chemical process development that were approached with ANN. Both optimal PID control tuning parameters and product particle size predictions were constructed from process information using neural networks. The ANN provides a rapid solution to many applications with little physical insight into the underlying system function. The amount of data preparation and performance limitations using a neural network will be discussed. However, the properly applied ANN will generally provide insight to which variables are most influential to the model and evolve dynamically to the minimum performance surface squared error. Neural networks have been used successfully with non-linear dynamic systems and can be applied to chemical process development for system identification and multivariate optimization problems.     

人工神经网络在智能机械设计中的应用

介绍了智能ＣＡＤ的概念和发展,分析了人工神经网络（ＡＮＮ）的特点,针对目前机械设计专家系统存在的问题,提出将ＡＮＮ应用到专家系统的设计中,是进行智能ＣＡＤ的一条有效途径。介绍了ＡＮＮ在概念设计、设计过程中形象思维的模拟、知识的获取和表示、回溯问题的模拟等方面的应用。     

A new approach for semiconductor parameter extraction using cathodoluminescence and artificial neural networks

In this paper, a new parameter extraction technique that jointly extracts four semiconductor-related parameters from theoretical/experimental cathodoluminescence data collected as a function of electron-beam energy is presented. The extraction technique is based on feed-forward artificial neural networks (ANN) where the ANN is trained to learn the inherent relationship between the input parameters (absorption coefficient α, diffusion length L, dead layer thickness Zt, and relative quantum efficiency Q) and the output parameter (CL intensity versus electron beam energy). After the training of the ANN, it is possible to observe the reverse process and extract the four parameters from any CL curve using an exhaustive search method. One of the main advantages of the proposed method is that the optimum set of values for the four parameters (α, L, Zt, Q) are obtained because the exhaustive search is performed in the search space spanned by all four parameters. Computational results on an n-type GaAs free defect semiconductor sample show that a unique set of parameter values with errors less than 5.5% from the nominal values can be obtained for each set of the experimental data points using the proposed algorithm.     

A simulated annealing/local search to minimize the makespan and total tardiness on a hybrid flowshop

Scheduling is a major issue faced every day in manufacturing systems as well as in the service industry, so it is essential to develop effective and efficient advanced manufacturing and scheduling technologies and approaches. Also, it can be said that bi-criteria scheduling problems are classified in two general categories respecting the approach used to solve the problem. In one category, the aim is to determine a schedule that minimizes a convex combination of two objectives and in the other category is to find a good approximation of the set of efficient solutions. The aim of this paper is to determine a schedule for hybrid flowshop problem that minimizes a convex combination of the makespan and total tardiness. For the optimization problem, a meta-heuristic procedure is proposed based on the simulated annealing/local search (SA/LS) along with some basic improvement procedures. The performance of the proposed algorithm, SA/LS, is compared with a genetic algorithm which had been presented in the literature for hybrid flowshop with the objective of minimizing a convex combination of the makespan and the number of tardy jobs. Several computational tests are used to evaluate the effectiveness and efficiency of the proposed algorithm against the other algorithm provided in the literature. From the results obtained, it can be seen that the proposed algorithm in comparison with the other algorithm is more effective and efficient.     

Automatic recognition of machining features using artificial neural networks

We report on the development of an intelligent system for recognizing prismatic part machining features from CAD models using an artificial neural network. A unique 12-node vector scheme has been proposed to represent machining feature families having variations in topology and geometry. The B-Rep CAD model in ACIS format is preprocessed to generate the feature representation vectors, which are then fed to the neural network for classification. The ANN-based feature-recognition (FR) system was trained with a large set of feature patterns and optimized for its performance. The system was able to efficiently recognize a wide range of complex machining features allowing variations in feature topology and geometry. The data of the recognized features was post-processed and linked to a feature-based CAPP system for CNC machining. The FR system provided seamless integration from CAD model to CNC programming.     

Application of finite element method and artificial neural network to the die design of radial forging processes

This investigation adopts the finite element method (FEM) and the artificial neural network (ANN) to plan the radial forging of work-hardened materials to yield the optimal designed die. The process parameters considered herein are die corner radius (R), ring gap height (H), friction factor (m), work-hardening coefficient (n), gap between the billet and die (c) and the punch load (f). The accuracy of the FEM model constructed herein is established. Fifty sets of processing parameters are simulated by the FEM, and the results, together with the outer rims of the flange after forming, are taken as the learning file in ANN. Then, based on the range that is set by the learning file, another 20 sets of flange with different shapes than those in the test file are selected to obtain a combination of parameters of the die, materials and lubricants and other factors. During the design of the die, many tests are conducted, and flanges of similar shapes are found to be obtained with various combinations of processing parameters. This result indicate that the learning pattern presented herein meets the needs of all types of parameter combinations. Finally, based on the required specification of the shape of the outer rim of the flange, this work uses ANN to obtain all the specified processing parameters. Finite-element analysis is then used to confirm the accuracy of the results and further investigate the effect of the related parameters on the flange shape. The following conclusion is drawn: The design of the die can yield finished flange products with similar shapes using different parameter combinations. During the forming process, a suitable range of parameters is selected for the die, the materials and the lubricant. Then, according to the strength of their effects, their inputs and output values are appropriately adjusted and the most suitable combination of processing parameters identified according to the similarities in the flange shapes they produce.     

改进遗传算法优化的神经网络在智能故障诊断中的应用

设计了用模拟退火的混合遗传算法代替BP网络的反向传播过程的改进算法,解决了在机械敲障诊断系统中BP算法容易陷入局部极小值的问题.该算法是在遗传法中引入模拟退火机制,将其同BP算法结合,形成一个混合的优化算法.新算法既有神经网络的学习能力和鲁裤性,又有遗传算法的强的全局随机搜索能力.仿真结果表明,这种改进算法极大提高了内燃机故障诊断系统的效率和准确性.