YuKang Liu

Dynamic Neuro-Fuzzy-Based Human Intelligence Modeling and Control in GTAW

Human welders experiences and skills are critical for producing quality welds in manual gas tungsten arc welding (GTAW) process. In this paper, a neuro-fuzzy-based human intelligence model is constructed and implemented as an intelligent controller in automated GTAW process to maintain a consistent desired full penetration. An innovative vision system is utilized to real-time measure the specular 3D weld pool surface under strong arc light interference. Experiments are designed to produce random changes in the welding speed and voltage resulting in fluctuations in the weld pool surface. Adaptive neuro-fuzzy inference system (ANFIS) is proposed to correlate the human welders response to the 3D weld pool surface as characterized by its width, length and convexity. Closed-loop control experiments are conducted to verify the robustness of the proposed controller. It is found that the human intelligence model can adjust the current to robustly control the process to a desired penetration state despite different initial conditions and various disturbances. A foundation is thus established to explore the mechanism and transformation of human welders intelligence into robotic welding systems.

Modeling and control of quasi-keyhole arc welding process

Quasi-keyhole is a novel approach proposed to operate the keyhole arc welding process. Because the methods effectiveness depends on the amperage of the peak current used to establish the keyhole, this paper proposes adjusting the amperage based on the duration of the peak current, which equals the keyhole establishment time. A nominal model structure has been selected from those identified using experimental data and been used in the design of an adaptive predictive control system. Closed-loop control experiments have been conducted to verify the effectiveness of the developed system under varying set-points and varying travel speeds.

Iterative Local ANFIS-Based Human Welder Intelligence Modeling and Control in Pipe GTAW Process: A Data-Driven Approach

Combining human welder (with intelligence and versatility) and automated welding systems (with precision and consistency) can lead to intelligent welding systems. This paper aims to present a data-driven approach to model human welder intelligence and use the resultant model to control automated gas tungsten arc welding process. To this end, an innovative machine-human cooperative virtualized welding platform is teleoperated to conduct training experiments. The welding current is randomly changed to generate fluctuating weld pool surface and the human welder tries to adjust his arm movement (welding speed) based on his observation on the real-time weld pool feedback/image superimposed with an auxiliary visual signal which instructs the welder to increase/reduce the speed. Linear model is first identified from the experimental data to correlate welders adjustment on the welding speed to the 3-D weld pool surface and a global adaptive neuro-fuzzy inference system (ANFIS) model is then proposed to improve the model accuracy. To better distill the detailed behavior of the human welder, K -means clustering is performed on the input space such that a local ANFIS model is identified. To further improve the accuracy, an iterative procedure has been performed. Compared to the linear, global and local ANFIS model, the iterative local ANFIS model provides better modeling performance and reveals more detailed intelligence human welders possess. To demonstrate the effectiveness of the proposed model as an effective intelligent controller, automated control experiments have been conducted. Experimental results verified that the controller is robust under different welding currents and welding speed disturbance.

Toward Welding Robot With Human Knowledge: A Remotely-Controlled Approach

This paper presents a remotely controlled welding scheme that enables transformation of human welder knowledge into a welding robot. In particular, a 6-DOF UR-5 industrial robot arm is equipped with sensors to observe the welding process, including a compact 3D weld pool surface sensing system and an additional camera to provide direct view of the work-piece. Human welder operates a virtual welding torch, whose motion is tracked by a Leap sensor. To remotely operate the robot based on the motion information from the Leap sensor, a predictive control approach is proposed to accurately track the human motion by controlling the speed of the robot arm movement. Tracking experiments are conducted to track both simulated movement with varying speed and actual human hand movement. It is found that the proposed predictive controller is able to track human hand movement with satisfactory accuracy. A welding experiment has also been conducted to verify the effectiveness of the proposed remotely-controlled welding system. A foundation is thus established to realize teleoperation and help transfer human knowledge to the welding robot.

Real-time measurement of the weld pool surface in GTAW process

This study aims at real-time measurement /reconstruction of the specular three-dimensional (3D) weld pool surface under strong arc in gas tungsten arc welding (GTAW). A vision-based sensing system is used to project a dot-matrix laser pattern on the weld pool surface which is able to specularly reflect the laser pattern. This reflection pattern, deformed by the weld pool surface has been utilized to reconstruct the 3D shape of the weld pool surface. Based on careful analysis, the underlying reconstruction problem is formulated mathematically. An analytic algorithm is proposed resulting in the weld pool surface be reconstructed in real-time on average in 3.22 milliseconds. In order to verify the effectiveness of the proposed reconstruction algorithm, it is tested on objects with a priori knowledge of geometry having a specular surface. Then the robustness of the proposed reconstruction scheme is further validated using a real-time experiment.The detailed analysis of measurement error validates the accuracy of the proposed algorithm.

Supervised Learning of Human Welder Behaviors for Intelligent Robotic Welding

Current industrial welding robots are mostly articulated arms with a pre-programmed set of movement, which lack the intelligence skilled human welders possess. In this paper, human welders response against 3-D weld pool surface is learned and transferred to the welding robots to perform automated welding tasks. To this end, an innovative teleoperated virtualized welding platform is utilized to conduct dynamic training experiments by a human welder whose arm movements together with the 3-D weld pool characteristic parameters are recorded. The data is off-line rated by the welder and a fuzzy classifier is trained, using an adaptive neuro-fuzzy inference system (ANFIS), to automate the rating. Data from the training experiments are then automatically classified such that top rated data pairs are selected to model and extract “good response” minimizing the effect from “bad operation” made during the training. A supervised ANFIS model is then proposed to correlate the 3-D weld pool characteristic parameters and welders adjustment on the welding speed. The obtained model is then transferred to the welding robot to perform automated welding task as an intelligent controller. Experiment results verified that the proposed model is able to control the process under different welding current as well as under disturbances in speed and measurement. A foundation is thus established to selectively learn “good response” to rapidly extract human intelligence to transfer into welding robots.

Neuro-fuzzy based human intelligence modeling and robust control in Gas Tungsten Arc Welding process

Human welders experiences and skills are critical for producing quality welds in Gas Tungsten Arc Welding (GTAW) process. Modeling of the human welders response to 3D weld pool surface can help develop next generation intelligent welding machines and train welders faster. In this paper, a neuro-fuzzy based human intelligence model is constructed and implemented as an intelligent controller in automated GTAW process to maintain a consistent desired full penetration. An innovative vision system is utilized to real-time measure the specular 3D weld pool surface under strong arc interference in GTAW process. Experiments are designed to produce random changes in the welding speed and voltage resulting in fluctuations in the weld pool surface. Adaptive Neuro-Fuzzy Inference System (ANFIS) is proposed to correlate the human welders response to the 3D weld pool surface. Control experiments are designed to start welding using different initial current and have various disturbances including variations of arc length and welding speed. It is found that the proposed human intelligence model can adjust the current to robustly control the process to a desired penetration state despite different initial conditions and various disturbances. A foundation is thus established to explore the mechanism and transformation of human welders intelligence into robotic welding system.

Data driven modeling of human welder intelligence: A neuro-fuzzy approach

Modeling of skilled human welders response to 3D weld pool surface can help develop next generation intelligent robotic welding systems and train welders faster. In this paper, neuro-fuzzy based data driven modeling of human welder intelligence is conducted. An innovative vision system is utilized to real-time measure the specular 3D weld pool surface under strong arc interference. Experiments are designed to produce random changes in the welding speed and voltage resulting in fluctuations in the weld pool surface. Adaptive Neuro-Fuzzy Inference System is proposed to correlate skilled human welder response to the fluctuating 3D weld pool surface. It is found that the proposed neuro-fuzzy model can model the human welder intelligence with good accuracy. Comparison of the novice and skilled human welder also reveals detailed adjustments made by the skilled human welder and help train the novice welder. A foundation is thus established to explore the mechanism and transformation of human welders intelligence into robotic welding system.

Dynamic neuro-fuzzy estimation of the weld penetration in GTAW process

The weld pool contains abundant information about the welding process and can thus be utilized to accurately monitor the weld penetration. This paper addresses the dynamic estimation of the weld penetration in GTAW process. A machine vision based weld pool sensing system is utilized and the 3D weld pool surface is reconstructed in real-time. Various dynamic experiments under different welding conditions are conducted to acquire data pairs for establishing the correlation between the front-side weld pool characteristic parameters and the weld penetration specified by its back-side bead width. Due to the substantial inertia of the welding process, the weld penetration can be more accurately estimated if the adjacent weld pools are used. Hence, a nonlinear dynamic Adaptive Neuro-Fuzzy Inference System (ANFIS) model is developed to estimate the weld penetration in real-time. It is found that the weld penetration can be estimated with satisfactory accuracy by the proposed online monitoring system.

Estimation of weld penetration using parameterized three-dimensional weld pool surface in gas tungsten arc welding

A skilled welder determines the weld joint penetration from his/her observation on the weld pool surface during the welding process. This paper addresses the estimation of weld joint penetration (i.e. determining back-side bead width that measures the degree of the weld joint penetration in full penetration welding) using the parameterized 3D weld pool surface in gas tungsten arc welding (GTAW). To this end, an innovative machine vision system is used to measure the specular weld pool surface in real-time. Various experiments under different welding conditions have been performed to produce full penetration welds with different back-side bead widths and acquire corresponding images for reconstructing the weld pool surface and calculating candidate estimation parameters. Through the least squares algorithm based statistic analyses, it was found that the width, length, and convexity of the 3D weld pool surface provides the optimal model to predict the back-side bead width with an acceptable accuracy. A foundation is thus established to effectively extract information from the weld pool surface to facilitate a feedback control of the weld joint penetration.