Back-side Bead Width Research Articles

How to Accurately Monitor the Weld Penetration From Dynamic Weld Pool Serial Images Using CNN-LSTM Deep Learning Model?

This letter illustrates how a challenging problem be solved by assuring the adequacy of the raw information and using an appropriate deep learning network to extract the relevant information. The problem concerned is accurate monitoring of the penetration, in a fully penetrated weld pool, as quantified by the width of the weld on the back-side of the workpiece. This is challenging as the penetration occurs below the workpiece surface and is not visible. A popular method is to use a weld pool image to derive it. Analysis of the physical process suggests that a single weld pool does not contain adequate information but most recent serial weld pools may. As such, although a deep learning model may extract information that is already there, the raw information may not be sufficient. Hence, a model that is capable of extracting information from dynamic serial weld pool images is needed. To this end, a CNN-LSTM (convolutional neural network combined with long-short term memory one) model is proposed. Dynamic weld pools are experimentally generated by changing the welding current and welding speed randomly. The weld pools are imaged using an HDR camera during experiments. Images are also captured from the back-side surface of the workpiece to provide the ground truth for training, validation, and testing. It is found that the highly dynamically changing weld pool can be accurately predicted using serial weld pool images at 0.3 mm for its back-side bead width. Comparison has been made with results from comparative studies to verify the effectiveness of and the contribution from the information adequacy (by using serial images) and the feature extracting capability (by using deep learning).

Vision sensing and feedback control of weld penetration in helium arc welding process

Consistent joint penetration is essential to mechanical properties of weldments, yet difficult to realize in gas tungsten arc welding due to heat accumulation, varying welding conditions, and uncertain disturbance, especially for manufacturing large-scale structural components. The solution is to monitor welding penetration in real-time and implement closed-loop control. In this paper a directional light-assisted vision sensing scheme is proposed to measure the transient profile of the backside weld pool/bead for monitoring and further controlling the joint penetration state. The developing zone and the developed zone of the backside weld bead are recognized, helping determining its final width. And the relationship between the transient backside bead width and the corresponding final width is established. Furthermore, the backside bead profile is employed as a feedback signal and the closed-loop control of weld formation is implemented to achieve full penetration and a constant backside bead width. A multiple-input single-output model predictive controller is utilized in the control system, creatively combined with an anti-windup integrator eliminating the steady-state error caused by nonlinearity in welding. Experimental results have proved that the developed control system is effective in keeping a consistent backside bead width with a rapid response, a small overshoot and robustness against varying heat transfer conditions.

Hybrid machine learning-enabled adaptive welding speed control

Industrial robots have become more diverse and common for automating manufacturing processes, such as welding. Existing robotic control, however, is incapable of adaptively adjusting its operation in response to a dynamic welding environment, whereas a skilled human welder can. Sophisticated and adaptive robotic control relies on the effective and efficient processing of perception data, characterization and prediction of highly dynamic systems, and real-time adaptative robotic reactions. This research presents a preliminary study on developing appropriate Machine Learning (ML) techniques for real-time welding quality prediction and adaptive welding speed adjustment for GTAW welding at a constant current. In order to collect the data needed to train the hybrid ML models, two cameras are applied to monitor the welding process, with one camera (available in practical robotic welding) recording the top-side weld pool dynamics and a second camera (unavailable in practical robotic welding, but applicable for training purpose) recording the back-side bead formation. Given these two data sets, correlations can be discovered through a convolutional neural network (CNN) that is good at image characterization. With the CNN, top-side weld pool images can be analyzed to predict the back-side bead width during active welding control. Furthermore, the monitoring process has been applied to multiple experimental trials with varying speeds. This allowed the effect of welding speed on bead width to be modeled through a Multi-Layer Perceptron (MLP). Through the trained MLP, a computationally efficient gradient descent algorithm has been developed to adjust the travel speed accordingly to achieve an optimal bead width with full material penetration. Because of the nature of gradient descent, the robot would change faster when the quality is further away and then fine-tune the speed when it was close to the goal. Experimental studies have shown promising results on real-time bead width prediction and adaptive speed adjustment to realize ideal bead width.

Closed-loop control of medium-thickness plate arc welding based on weld-face vision sensing

In one-side welding with full penetration, the use of backing strips helps support molten weld metals, but increases manufacturing cycle time and cost. The backing-free welding technology is under investigation for medium-thickness structural components in the aerospace industry. Its crucial challenge is ensuring complete joint penetration and maintaining a suitable and stable backside bead width. This paper presents an innovative method of controlling weld formation based on weld-face vision sensing in arc welding. A standard laser triangulation setup is used to obtain images of weld pool, and a new weld-face feature ‘depression degree of the weld pool’ is proposed. An image processing procedure is developed to extract the depression degree of the weld pool from images in real-time. The depression degree of the weld pool is found to correlate closely with the backside bead width, and is a better indicator of the backside bead width than the weld pool width, with a higher consistency and a smaller time lag. Employing the depression degree as a feedback signal, a backside bead width control method is developed and implemented. Welding experiments on 6-mm-thick aluminum alloy 2219 plates show that the presented method is remarkably effective in achieving full penetration and keeping backside bead width constant.

Dynamic estimation of joint penetration by deep learning from weld pool image

This work aims at a novel approach to estimate the root-pass penetration towards its feedback control, in which the real penetration is measured by the backside bead width. The major challenge is that it happens under the workpiece and likely cannot be directly observable. The dynamic evolution of the weld pool surface has been analysed to design an active vision method monitoring the pool surface, yet fundamentally correlated to the unobservable penetration. The designed convolutional neural network model is trained, validated, and tested for recognising the weld penetration with satisfactory accuracy.

Deep learning-empowered digital twin for visualized weld joint growth monitoring and penetration control

This paper presents an innovative digital twin to monitor and control complex manufacturing processes by integrating deep learning which offers strong feature extraction and analysis abilities. Taking welding manufacturing as a case study, a deep learning-empowered digital twin is developed as the visualized digital replica of the physical welding for joint growth monitoring and penetration control. In such a system, the information available directly from sensors including weld pool images, arc images, welding current and arc voltage is collected in pulsed gas tungsten arc welding (GTAW-P). Then, the undirect information charactering the weld joint geometry and determining the welding quality, including the weld joint top-side bead width (TSBW) and back-side bead width (BSBW), is computed/estimated by traditional image processing methods and deep convolutional neural networks (CNNs) respectively. Compared with single image source, weld pool image or arc image, the CNN model performs better when taking the 2-channel composite image combined by both as the input and the state-of-the-art accuracy in BSBW prediction with mean square error (MSE) as 0.047 mm2 is obtained. Then, a decision-making strategy is developed to control the welding penetration to meet the quality requirement and applied successfully in various welding conditions. By modeling the weld joint cross section as an ellipse, the developed digital twin is visualized to offer a graphical user interface (GUI) for users perceiving the weld joint growth intuitively and effectively.

A tutorial on deep learning-based data analytics in manufacturing through a welding case study

Abstract Over the past decade, machine learning and deep learning have been increasingly reshaping manufacturing towards smart manufacturing. This paper aims to provide a tutorial for researchers to understand the basic principles of deep learning and its applications to manufacturing, using welding as an example. In this tutorial, we first present an overview of welding processes and the advantages of deep learning in solving welding problems, such as process monitoring and product quality prediction. Then, deep learning characteristics are summarized and two representative deep learning techniques, conventional neural networks (CNNs) and recurrent neural networks (RNNs) that are suitable for image processing and sequential modeling, are discussed. A case study on welding quality prediction that predicts the back-side bead width from top-side images through a CNN is demonstrated, with detailed procedures and core codes from building a CNN to testing the network performance. Prospects for deep learning in a manufacturing context are examined from the authors’ perspective.

Weld penetration in situ prediction from keyhole dynamic behavior under time-varying VPPAW pools via the OS-ELM model

In situ monitoring and accurate detecting of welding quality have been one of the common challenges of automatic welding process. This paper contributes an intelligent decision-making framework for the weld penetration prediction from the keyhole dynamic behavior under time-varying VPPAW pools. Initially, a series of dynamic experiments under different welding conditions were conducted to acquire the backside images of keyhole and corresponding backside bead width. Then, the geometry appearance of keyhole was described by the supervised descent method (SDM)–based image processing algorithm. Subsequently, the internal correlation between the keyhole characteristics and the backside width was further derived to help understand the nonlinear and time-varying VPPAW process. Finally, a novel dynamic model based on an online sequential extreme learning machine (OS-ELM) was designed to predict the weld penetration as measured by the backside bead width in real time. Extensive experiment results further verify and validate that the proposed dynamic OS-ELM model is significantly better than other state-of-the-art algorithms in terms of predicting accuracy, efficiency, and robustness.

Machine learning of weld joint penetration from weld pool surface using support vector regression

Skilled human welders can control the weld joint penetration through observing the molten pool. This suggests that a model may be developed to predict the backside bead width, that quantitively measures the weld joint penetration, from the weld pool surface. However, the weld pool surface is specular and subject to the radiation of the arc such that its measurement is challenging. At the University of Kentucky, the weld pool surface is measured using an innovative a 3-D vision sensor that can overcome the challenges caused by the specular nature and arc radiation; and the measured surface is characterized by three parameters. Because of the lack of physics based model, neural networks would typically be used to approximate the unknown correction, which is nonlinear in general, between the backside bead width and the characteristics parameters. Unfortunately, neural networks require large amount of data to train for adequate model accuracy. While the weld pool surface can be measured using the innovative 3D sensor, the ground truth for the backside bead width needs to be measured off-line after the experiment and to this end the work-work needs to appropriately cleaned/processed. Large amount of training data needed may not be easily obtained. To improve the critical ability to accurately predict the backside bead width, models need to be established from relatively small amount of training data. To this end, the authors propose to use the support vector regression (SVR) method and hypothesize that a SVR model trained using the small amount of the training data available would perform better than that a multi-layer perceptron (MLP) artificial neural network model trained using the same data. Modeling results show that for the relatively small training data available, the optimized SVR model provides a more accurate prediction to the backside bead width. As such, the authors systematically advanced the ability to accurately predict the weld joint penetration. The use of the innovative 3D sensor to obtain the 3D weld pool surface and the proposed use of the support vector method to address the small data issue played crucial roles.

Model-Based Predictive Control of Weld Penetration in Gas Tungsten Arc Welding

Skilled welders can estimate and control the weld joint penetration, which is primarily measured by the backside bead width, based on weld pool observation. This suggests that an advanced control system could be developed to control the weld joint penetration by emulating the estimation and decisionmaking process of the human welder. In this paper an innovative 3-D vision sensing system is used to measure the characteristic parameters of the weld pool in real-time in gas tungsten arc welding. The measured characteristic parameters are used to estimate the backside bead width, using an adaptive neuro-fuzzy inference system (ANFIS) as an emulation of skilled welder. Dynamic experiments are conducted to establish the model that relates the backside bead width to the welding current and speed. The dynamic linear model is first constructed and the modeling result is analyzed. The linear model is then improved by incorporating a nonlinear operating point modeled by an ANFIS. Because the weld pool needs to gradually change, being controlled by a skilled welder, a model predictive control is used to follow a trajectory to reach the desired backside bead width and the control increment is penalized. Because the weld pool is not supposed to change in an extremely large range, the resultant model predictive control is actually linear and an analytical solution is derived. Welding experiments confirm that the developed control system is effective in achieving the desired weld joint penetration under various disturbances and initial conditions.

Closed-Loop Control of Robotic Arc Welding System with Full-penetration Monitoring

The real-time detection of the state of the gap and weld penetration control are two fundamental issues in robotic arc welding. However, traditional robotic arc welding lacks external information feedback and the function of real-time adjusting. The objective of this research is to adopt new sensing techniques and artificial intelligence to ensure the stability of the welding process through controlling penetration depth and weld pool geometry. A novel arc welding robot system including function modules (visual modules, data acquisition modules) and corresponding software system was developed. Thus, the autonomy and intelligence of the arc welding robot system is realized. Aimed at solving welding penetration depth, a neural network (NN) model is developed to calculate the full penetration state, which is specified by the back-side bead width (Wb), from the top-side vision sensing technique. And then, a versatile algorithm developed to provide robust real-time processing of images for use with a vision-based computer control system is discussed. To this end, the peak current self adaptive regulating controller with weld gap compensation was designed in the robotic arc welding control system. Using this closed-loop control experiments have been conducted to verify the effectiveness of the proposed control system for the robotic arc welding process. The results show that the standard error of the Wb is 0.124 regardless of the variations in the state of the gap.

CMAC-based modelling for HPDDL welding process control

A Cerebellar Model Articulation Controller- (CMAC-) based modelling method and closed-loop control system were developed to estimate and control the weld fusion for measuring the topside and backside bead widths of the weld pool in High Power Direct Diode Laser (HPDDL) welding. Because of the difficulty in using backside sensor, a topside weld pool geometry is measured using a topside vision sensor and is also used to measure the backside bead width. The drive current of the laser and the welding speed are taken as the control variables. The basic function of CMAC-based steady-state models, which are more easily obtained than dynamic ones, of the non-linear controlled process can predict the control variables with satisfactory accuracy to shorten output response transient time, and two simple Proportional and Integral (PI) controllers are included in the control system to adjust the control variables to maintain outputs at the desired levels. The results of closed-loop control simulations of laser welding process demonstrate that the developed control system is effective and robust to fluctuations or variations that lead to the changing parameters of the non-linear model.

Intelligent control of arc welding process

Weld joint penetration is crucial in generating sound welds. Successful control of penetration requires accurate measurement of back-side width of the weld pool. For practical application, the sensor used should be attached to the welding torch. The back-side bead width is thus invisible. It is known thatskilled operators can estimate joint penetration based on the geometry of the weld pool. In this work, a neurofuzzy system is developed to infer the back-side width from pool geometry. Both the back-side and weld-face widths of the weld pool can therefore be feedback controlled. Welding current and welding speed are selected as the control variables. The controlled 2-input 2-output non-linear process is described using a neurofuzzy model. Based on the neurofuzzy dynamic model, a predictive control algorithm is designed. Experimentation verified the effectiveness of the developed system.

Neurofuzzy model-based predictive control of weld fusion zone geometry

A closed-loop system is developed to control the weld fusion, which Is specified by the top-side and back-side bead widths of the weld pool. Because in many applications only a top-side sensor is allowed, which is attached to and moves with the welding torch, an image processing algorithm and neurofuzzy model have been incorporated to measure and estimate the top-side and back-side bead widths based on an advanced top-side vision sensor. The welding current and speed are selected as the control variables. It is found that the correlation between any output and input depends on the value of another input. This cross coupling implies that a nonlinearity exists in the process being controlled. A neurofuzzy model is used to model this nonlinear dynamic process. Based on the dynamic fuzzy model, a predictive control system has been developed to control the welding process. Experiments confirmed that the developed control system is effective in achieving the desired fusion state despite the different disturbances.

Neurofuzzy model-based weld fusion state estimation

Proper fusion is crucial in generating a sound weld. Successful control of the fusion state requires accurate measurements of both the topside and back-side bead widths. A top-side sensor-based system is preferred so that the sensor can be attached to and moved with the torch. Thus, the system must be capable of estimating the back-side bead width with high accuracy. Because skilled human operators can estimate the fusion state from the observed weld pool, a neurofuzzy system is developed to infer the backside bead width from the pool geometry in this work. It is found that the back-side bead width can be estimated with satisfactory accuracy by the identified neurofuzzy model. Thus, accurate feedback of the fusion state can be provided for its control.

Numerical Analysis of Fully Penetrated Weld Pools in Gas Tungsten Arc Welding

Full penetration welding is widely used in metal joining, but it has been ignored in previous convective numerical models. In addition to the free surface on top of the pool, an additional free surface appears on the bottom of the workpiece. It can be shown that the top surface, temperature distribution and fluid flow field in the weld pool are all coupled with the pool's bottom surface. This complicates the numerical process and therefore no convective models have previously been developed for fully penetrated weld pools. In order to improve the numerical solution for the fully penetrated weld pool, a three-dimensional model is proposed. Free top and bottom pool surfaces have been included. The electromagnetic force, buoyancy force and surface tension gradient (Marangoni) are the three driving forces for weld pool convection. Welding parameters are changed in order to analyse their effects on weld pool geometry. It is found that the depression of the top surface contains abundant information about the full penetration state as specified by the back-side bead width.

Infrared sensing of full penetration state in gas tungsten arc welding

The objective of this study is to present an applicable top-side infrared sensing technique for the prospective closed-loop control of weld penetration in gas tungsten arc welding (GTAW). A model is developed to calculate the full penetration state, which is specified by the back-side bead width, from the sensed infrared images. To ensure the model validity in the prospective closed-loop control, the experiments, which generate the data for the model identification, are conducted under the experimental conditions that will be encountered during practical closed-loop control of the welding process. The heat transfer condition and electrode tip angle may vary during welding or from case to case. Also, the control variables which are used to adjust the weld process in order to reach the required weld penetration will also change. In many cases, the current can be employed as an on-line adjustable control variable because of the implementation ease, when the welding speed and arc length are maintained at the preset values. Thus, different currents, workpiece sizes, and electrode tip angles are arranged in the experiments to emulate the possible current adjustments, case to case heat transfer variations, and electrode wear. The infrared characteristics of the effects of these parameters are extracted to regress the full penetration state. Finally, the back-side weld width (the full penetration state) is calculated using the resultant model from the sensed infrared data.

母板，裏当て銅板間電圧の成因と裏ビード幅との関係 Ｉ フラックス銅裏当て方式（ＦＣＢ）片面溶接における裏ビード幅自動制御

A possibility of a back bead width control system have been proposed to stabilize the process of one side FCB SAW in which a stationary copper backing plate covered with a flux layer of several milimeter thickness is used as backing.When a stable welding process is maintained under an adequate welding condition, a signal voltage between the backing copper plate and the plate being welded have beeen detected. The polarity and phase of the detected voltage having a few volt level is well corresponding to those of the arc voltage in case of AC current welding.In a stable FCB SAW process, the back side bead is mostly formed by a key-hole action of the arc. It is assumed for the reason that the arc penetrates through' the welding groove and the flux layer, then it contacts the copper plate and is reflected to the welding plate. The correlation between the detected voltage level and the back side bead width is recognized.The similar phenomenon is observed in a key-hole welding by using a transfer type plasma arc welding.In order to maintain the back bead width constant the detecting voltage is compared with a reference voltage, and then the welding current is controlled by the difference of the above voltages.