Optimal design of neural networks for control in robotic arc welding

Robotic gas metal arc (GMA) welding is a manufacturing process which is used to produce high quality joints and has to a capability to be utilized in automation systems to enhance productivity. Despite its widespread use in the various manufacturing industries, the full automation of the robotic GMA welding has not yet been achieved partly because mathematical models for the process parameters for a given welding tasks are not fully understood and quantified. In this research, an attempt has been made to develop a neural network model to predict the weld bead width as a function of key process parameters in robotic GMA welding. The neural network model is developed using two different training algorithms; the error back-propagation algorithm and the Levenberg–Marquardt approximation algorithm. The accuracy of the neural network models developed in this study has been tested by comparing the simulated data obtained from the neural network model with that obtained from the actual robotic welding experiments. The result shows that the Levenberg–Marquardt approximation algorithm is the preferred method, as this algorithm reduces the root of the mean sum of squared (RMS) error to a significantly small value.     

Predicting the depth of penetration and weld bead width from the infra red thermal image of the weld pool using artificial neural network modeling

It is necessary to estimate the weld bead width and depth of penetration using suitable sensors during welding to monitor weld quality. Among the vision sensors, infra red sensing is the natural choice for monitoring welding processes as welding is inherently a thermal processing method. An attempt has been made to estimate the weld bead width and depth of penetration from the infra red thermal image of the weld pool using artificial neural network models during A-TIG welding of 3?mm thick type 316 LN stainless steel plates. Real time infra red images were captured using IR camera for the entire weld length during A-TIG welding at various current values. The image features such as length and width of the hot spot, peak temperature, and other features using line scan analysis are extracted using image processing techniques corresponding to particular locations of the weld joint. These parameters along with their respective current values are used as inputs while the measured weld bead width and depth of penetration are used as output of the neural network models. Accurate ANN models predicting weld bead width (9-11-1) and depth of penetration (9-9-1) have been developed. The correlation coefficient values obtained were 0.98862 and 0.99184 between the measured and predicted values of weld bead width and depth of penetration respectively.     

Simulation and parameter optimization of flux cored arc welding using artificial neural network and particle swarm optimization algorithm

Flux cored arc welding (FCAW) process is a fusion welding process in which the welding electrode is a tubular wire that is continuously fed to the weld area. It is widely used in industries and shipyards for welding heavy plates. Welding input parameters play a very significant role in determining the quality of a weld joint. This paper addresses the simulation of weld bead geometry in FCAW process using artificial neural networks (ANN) and optimization of process parameters using particle swarm optimization (PSO) algorithm. The input process variables considered here include wire feed rate (F); voltage (V); welding speed (S) and torch Angle (A) each having 5 levels. The process output characteristics are weld bead width, reinforcement and depth of penetration. As per the statistical design of experiments by Taguchi L₂₅ orthogonal array, bead on plate weldments were made. The experimental results were fed to the ANN algorithm for establishing a relationship between the input and output parameters. The results were then embedded into the PSO algorithm which optimizes the process parameters subjected to the objectives. In this study the objectives considered are maximization of depth of penetration, minimization of bead width and minimization of reinforcement.     

Optimization of laser butt welding parameters based on the orthogonal array with fuzzy logic and desirability approach

This paper presents the study carried out on 3.5 kW cooled slab laser welding of 904 L super austenitic stainless steel. The joints had butts welded with different shielding gases like argon, helium and nitrogen at a constant flow rate. Super austenitic stainless steel (SASS) normally contains high amount of Mo, Cr, Ni, N and Mn. The mechanical properties are controlled to obtain good welded joints. The quality of the joint is evaluated by studying the features of weld bead geometry such as bead width (BW) and depth of penetration (DOP). In this paper, the tensile strength and bead profiles (BW and DOP) of laser welded butt joints made of AISI 904 L SASS are investigated. Taguchi approach is used as statistical design of experiment (DOE) technique for optimizing the selected welding parameters. Fuzzy logic and desirability approach are applied to optimize the input parameters considering multiple output variables simultaneously. Confirmation experiment has also been conducted for both the analyses to validate the optimized parameters.     

A study on the quality improvement of robotic GMA welding process

With the advance of the robotic welding process, procedure optimisation that selects the welding procedure and predicts bead geometry that will be deposited has increased. A major concern involving procedure optimisation should define a welding procedure that can be shown to be the best with respect to some standard, and chosen combination of process parameters, which give an acceptable balance between production rate and the scope of defects for a given situation.

This paper represents a new algorithm to establish a mathematical model for predicting top-bead width through a neural network and multiple regression methods, to understand relationships between process parameters and top-bead width, and to predict process parameters on top-bead width in robotic gas metal arc (GMA) welding process. Using a series of robotic GMA welding, additional multi-pass butt welds were carried out in order to verify the performance of the multiple regression and neural network models as well as to select the most suitable model. The results show that not only the proposed models can predict the top-bead width with reasonable accuracy and guarantee the uniform weld quality, but also a neural network model could be better than the empirical models.     

Knowledge-based systems using neural networks for electron beam welding process of reactive material (Zircaloy-4)

Bead-on-plate welding of zircaloy-4 (a reactive material) plates was conducted using electron beam according to central composite design of experiments. Its predictive models were developed in the form of knowledge-based systems in both forward and reverse directions using neural networks. Input parameters considered for this welding of reactive metals were accelerating voltage, beam current and weld speed. The responses of the welding process were measured in terms of bead width, depth of penetration and micro-hardness. Forward mapping of the welding process was conducted using regression analysis, back-propagation neural network (BPNN), genetic algorithm-tuned neural network (GANN) and particle swarm optimization algorithm-tuned neural network (PSONN). Reverse mapping of this process was also carried out using the BPNN, GANN and PSONN-based approaches. Neural network-based approaches could model this welding process of reactive material in both forward and reverse directions efficiently, which is required for the automation of the same. The performance of the neural network models was found to be data-dependent. The BPNN could outperform the other two approaches for most of the cases but not all in both the forward and reverse mappings.     

RBF-NN-based model for prediction of weld bead geometry in Shielded Metal Arc Welding (SMAW)

Welding processes are considered as an essential component in most of industrial manufacturing and for structural applications. Among the most widely used welding processes is the shielded metal arc welding (SMAW) due to its versatility and simplicity. In fact, the welding process is predominant procedure in the maintenance and repair industry, construction of steel structures and also industrial fabrication. The most important physical characteristics of the weldment are the bead geometry which includes bead height and width and the penetration. Different methods and approaches have been developed to achieve the acceptable values of bead geometry parameters. This study presents artificial intelligence techniques (AIT): For example, radial basis function neural network (RBF-NN) and multilayer perceptron neural network (MLP-NN) models were developed to predict the weld bead geometry. A number of 33 plates of mild steel specimens that have undergone SMAW process are analyzed for their weld bead geometry. The input parameters of the SMAW consist of welding current (A), arc length (mm), welding speed (mm/min), diameter of electrode (mm) and welding gap (mm). The outputs of the AIT models include property parameters, namely penetration, bead width and reinforcement. The results showed outstanding level of accuracy utilizing RBF-NN in simulating the weld geometry and very satisfactorily to predict all parameters in comparison with the MLP-NN model.

     

Real-time simulation for a virtual reality-based MIG welding training system

This paper describes a real-time welding simulation method for use in a desktop virtual reality simulated Metal Inert Gas welding training system. The simulation defines the shape of the weld bead, the depth of penetration, and the temperature distribution in the workpiece, based on inputs from the motion-tracking system that tracks the position of the welding gun as a function of time. A finite difference method is used to calculate the temperature distribution, including the width of the weld bead and the depth of penetration. The shape of the weld bead is then calculated at each time step by assuming a semi-spherical volume, based on the width of the weld bead, the welding speed, and the wire feed rate. The real-time performance of the system is examined, and results from the real-time simulation are compared to physical tests and are found to have very good correlation for welding speeds up to 1,000 mm/min.     

Bead geometry prediction for robotic GMAW-based rapid manufacturing through a neural network and a second-order regression analysis

The single weld bead geometry has critical effects on the layer thickness, surface quality, and dimensional accuracy of metallic parts in layered deposition process. The present study highlights application of a neural network and a second-order regression analysis for predicting bead geometry in robotic gas metal arc welding for rapid manufacturing. A series of experiments were carried out by applying a central composite rotatable design. The results demonstrate that not only the proposed models can predict the bead width and height with reasonable accuracy, but also the neural network model has a better performance than the second-order regression model due to its great capacity of approximating any nonlinear processes. The neural network model can efficiently be used to predict the desired bead geometry with high precision for the adaptive slicing principle in layer additive manufacturing.     

An object detection and recognition system for weld bead extraction from digital radiographs

With base in object detection and recognition techniques, we developed and implemented a new methodology to perform the first head-function of a weld quality interpretation system: the weld bead extraction from a digital radiograph. The proposed methodology uses a genetic algorithm to manage the search for suitable parameters values (position, width, length, and angle) that best defines a window, in the radiographic image, matching with the model image of a weld bead sample. The search results are verified in a classification process that recognize true detections using image matching parameters also proposed in this work. To test the proposed methodology, two groups of images were used; one consisting of 110 radiographs from pipelines welded joints and the other containing 6 images with different numbers of radiographs per image. The tests results showed that, besides automatically check the number of weld beads per image, the proposed methodology is also able to supply the respective position, width, length, and angle of each weld bead, with an accurate rate of 94.4%. As a result, the detected weld beads are correctly extracted from the original image and made available to be inspected through others algorithms for failure detection and classification.     

Intelligent modelling of back-side weld bead geometry using weld pool surface characteristic parameters

In manual welding process, skilled welders can ensure the weld quality through compensating for deviation observed from the weld pool surface. In this paper a three dimensional vision sensing system was used to mimic the human vision system to observe the three-dimensional weld pool surface in pipe GTAW process. Novel characteristic parameters containing information about the penetration state specified by its back-side weld pool width and height were proposed based on the reconstructed three dimensional weld pool surfaces. Then, variation in characteristic parameters and their relationships with the back-side parameters were studied through experiments under different welding conditions. Direct measurement of penetration is not preferred in a manufacturing site, soft-sensing method was thus proposed as an alternative to obtain it in real time due to established soft-sensing model and auxiliary variables which can be sensed in real time. In order to obtain the penetration status in real time conveniently, back-propagation neural network, principle component analysis based back-propagation neural network and global best adaptive mutation particle swarm optimization based back-propagation neural network models were established to estimate the penetration based on the proposed characteristic parameters. It was found that the top-side characteristic parameters proposed can reflect the back-side weld pool parameters accurately and the models are capable of predicting the penetration status in real time by observing the three-dimensional weld pool surface.     

Microcomputer controlled arc oscillator for automated TIG welding

A magnetic arc oscillation system has been developed to control the fusion characteristic of arc welding. A single-pole electromagnet generates a transverse magnetic field parallel to the weld line. The weave pattern of the magnetic field, hence the bead geometry, is controlled by a custom-built computer. The applied magnetic field has a marked influence on the bead width, but less on the bead depth. Experiments also show that bead appearance is improved by magnetically oscillating the arc.     

自适应PID的薄壁筒体TIG焊接熔透控制研究

在薄壁筒体环缝焊接中,工艺参数以及坡口尺寸的波动都会导致焊接熔透状态的变化。为保证稳定的熔透,需要对焊接熔透状态进行实时检测和控制。针对薄壁筒体环缝TIG焊,建立了基于视觉传感的直接传感法,获取了清晰的熔池背面图像,经图像处理得到了满足熔透要求的熔池背面熔宽范围,并通过自整定PID控制器调节焊接速度,对熔透状态进行自适应控制。结果表明,该方法对熔透状态信号提取可靠,基本消除了熔透状态间接传感法的控制盲区。     

Effect of welding sequences on residual stresses

Accurately predicting welding residual stresses and developing an available welding sequence for a weld system are pertinent tasks since welding residual stress is inevitably produced in a welded structure. This study analyzes the thermomechanical behavior and evaluates the residual stresses with various types of welding sequence in single-pass, multi-pass butt-welded plates and circular patch welds. This is achieved by performing thermal elasto-plastic analysis using finite element techniques. Furthermore, this investigation provides an available welding sequence to enhance the fabrication process of welded structures.     

Optimization of friction stir welding process using NSGA-II and DEMO

In welding processes, the selection of optimal process parameter settings is very important to achieve best weld qualities. In this work, neuro-multi-objective evolutionary algorithms (EAs) are proposed to optimize the process parameters in friction stir welding process. Artificial neural network (ANN) models are developed for the simulation of the correlation between process parameters and mechanical properties of the weld using back-propagation algorithm. The weld qualities of the weld joint, such as ultimate tensile strength, yield stress, elongation, bending angle and hardness of the nugget zone, are considered. In order to optimize those quality characteristics, two multi-objective EAs that are non-dominated sorting genetic algorithm II and differential evolution for multi-objective are coupled with the developed ANN models. In the end, multi-criteria decision-making method which is technique for order preference by similarity to the ideal solution is applied on the Pareto front to extract the best solutions. Comparisons are conducted between results obtained from the proposed techniques, and confirmation experiments are performed to verify the simulated results.

     

Application of integrated soft computing techniques for optimisation of hybrid CO2 laser–MIG welding process

In this paper, artificial neural networks (ANNs), genetic algorithm (GA), simulated annealing (SA) and Quasi Newton line search techniques have been combined to develop three integrated soft computing based models such as ANN–GA, ANN–SA and ANN–Quasi Newton for prediction modelling and optimisation of welding strength for hybrid CO₂ laser–MIG welded joints of aluminium alloy. Experimental dataset employed for the purpose has been generated through full factorial experimental design. Laser power, welding speeds and wires feed rate are considered as controllable input parameters. These soft computing models employ a trained ANN for calculation of objective function value and thereby eliminate the need of closed form objective function. Among 11 tested networks, the ANN with best prediction performance produces maximum percentage error of only 3.21%. During optimisation ANN–GA is found to show best performance with absolute percentage error of only 0.09% during experimental validation. Low value of percentage error indicates efficacy of models. Welding speed has been found as most influencing factor for welding strength.     

Intelligent modelling and optimization of the gas tungsten arc welding process

This paper describes an application of neural networks and simulated annealing (SA) algorithm to model and optimize the gas tungsten arc welding (GTAW) process. The relationships between welding process parameters and weld pool features are established based on neural networks. In this study, the counter-propagation network (CPN) is selected to model the GTAW process due to the CPN equipped with good learning ability. An SA optimization algorithm is then applied to the CPN for searching for the welding process parameters with optimal weld pool features. Experimental results have shown that GTAW performance can be enhanced by using this approach.     

Numerical simulation of the electron beam welding process

Electron beam welding is a highly efficient and precise welding method that is being increasingly used in industrial manufacturing and is of growing importance in industry. Compared to other welding processes it offers the advantage of very low heat input to the weld, resulting in low distortion in components. Modeling and simulation of the laser beam welding process has proven to be highly efficient for research, design development and production engineering. In comparison with experimental studies, a modeling study can give detailed information concerning the characteristics of weld pool and their relationship with the welding process parameters (welding speed, electron beam power, workpiece thickness, etc.) and can be used to reduce the costs of experiments. A simulation of the electron beam welding process enables estimation of weld pool geometry, transient temperature, stresses, residual stresses and distortion. However this simulation is not an easy task since it involves the interaction of thermal, mechanical and metallurgical phenomena. Understanding the heat process of welding is important for the analysis of welding structure, mechanics, microstructure and controlling weld quality.In this paper the results of numerical simulation of electron beam welding of tubes were presented. The tubes were made of 30HGSA steel. The numerical calculation takes into consideration thermomechanical coupling (TMC). The simulation aims at: analysis of the thermal field, which is generated in welding process, determination of the heat-affected zone and residual stresses in the joint. The obtained results allow for determination both the material properties, and stress and strain state in the joint. Furthermore, numerical simulation allows for optimization of the process parameters (welding speed, power of the heat source) and shape of the joint before welding. The numerical simulation of electron beam welding process was carried out with the ADINA System v. 8.6. using finite element method.     

Phenomenological model-based study on electron beam welding process,and input-output modeling using neural networks trained by back-propagation algorithm,genetic algorithms,particle swarm optimization algorithm and bat algorithm

High power density welding technologies are widely used nowadays in various fields of engineering. However, a computationally efficient and quick predictive tool to select the operating parameters in order to achieve the specified weld attribute is conspicuously missing in the literature. In the present study, a computationally efficient inverse model has been developed using artificial neural networks (ANNs). These ANNs have been trained with the outputs of physics-based phenomenological model using back-propagation (BP) algorithm, genetic algorithm (GA), particle swarm optimization (PSO) algorithm and bat algorithm (BA) separately to develop both the forward and reverse models. Unlike data driven ANN model, such approach is unique and yet based on science. Power, welding speed, beam radius and power distribution factor have been considered as input process parameters, and four weld attributes, such as length of the pool, depth of penetration of the pool, half-width of the pool and cooling time are treated as the responses. The predicted outputs of both the forward and reverse models are found to be in good agreement with the experimental results. The novelty of this study lies with the development and testing of five neural network-based approaches for carrying out both forward and reverse mappings of the electron beam welding process.     

Finite element analysis of residual stress induced by shot peening process

The aircraft industry has only recently begun to explore possible application of welding as an alternative joining method for the design of future large civil airliner wing. One of the main obstacles, encountered in the past years, to welding application within the aircraft industries were due to failure in the weldments, caused by high tensile residual stresses present in the region of the weld, reducing drastically fatigue strength of welded joints. Improvement in the fatigue life of the welded joint can be obtained if compressive residual stresses are introduced at the weld region.Shot peening is a manufacturing process intended to give aircraft structures the final shape and to introduce a compressive residual state of stress inside the material in order to increase fatigue life. This paper presents the modeling and simulation of the residual stress field resulting from the shot peening process. The results achieved show that a significant decrease of welding induced tensile residual stress magnitude can be obtained. Good agreement between experimental and numerical results was achieved.