Kristinn Andersen

Weld modeling and control using artificial neural networks

Artificial neural networks were evaluated for monitoring and control of the variable polarity plasma arc welding (VPPAW) process. Three areas of welding application were investigated: weld process modeling, weld process control, and weld bead profile analysis for quality control. Experiments and analysis confirm that artificial neural networks are powerful tools for analysis, modeling, and control applications. They are particularly attractive in view of their capabilities to process nonlinear and noisy data, learn from actual welding data, and execute at relatively high speed. It is shown that neural networks are capable of modeling parameters of the VPPAW process to on the order of 10% accuracy or better. The same was observed when neural networks were used to select welding equipment parameters and the resulting bead geometries were estimated. These performance figures suggest that a VPPA welding control system can be implemented based on neural network models and control mechanisms. >

Synchronous weld pool oscillation for monitoring and control

A novel approach for inducing and monitoring oscillations in a molten weld pool is presented. Research efforts have illustrated that the weld pool resonates at natural frequencies that are related to its dimensions and state of penetration. This phenomenon may be used to monitor the weld pool, and particularly its depth of penetration, in a closed-loop feedback control system. The approach used to induce pool oscillations was to excite the weld pool with current pulses synchronized to the natural oscillations of the pool. Implementation of this synchronous weld pool pulsing technique was based on the use of a phase locked loop (PLL) system. The natural weld pool oscillations are used as the reference frequency source and a pulsing circuit is controlled by the PLL oscillator so that the arc current pulses repeatedly impact the pool after a fixed number of reference oscillation periods. An optical sensor detects the pool oscillations which are amplified, filtered, and limited to eliminate amplitude variations from the optical signal. A model of the weld pool is developed which uses a fluid droplet formulation for the relation of weld pool geometry and other physical parameters to the natural frequencies of the weld pool. Comparison of the weld pools actual resonant frequency with the expected resonant frequency as predicted by weld pool geometry models and measurements of the pool width (or area) allows and assessment of the state of penetration of the weld pool into the workpiece.<<ETX>>

Automated visual inspection and interpretation system for weld quality evaluation

The visual inspection of weld beads and subsequent evaluation of the weld quality is an integral part of the commercial welding environment. However, this inspection process tends to be both time and labor intensive. An automated system for the performance of this task has been developed. The sequence of events in the operation of the vision-based system are: (1) image capture; (2) image enhancement; (3) image processing; and (4) image evaluation. To minimize cost and complexity, the system uses conventional video camera and related hardware and software for the image-capture and image enhancement portion of the evaluation process. Various weld processes were observed to have certain characteristic features which were most relevant for the inspection and evaluation of the particular process. Image processing codes were written to extract those features of the weld beads and store this information in data files for subsequent assessment. Numerical algorithms were written, tailored to each of the weld processes, to perform the image evaluation portion of the quality evaluation process. This result is expressed as a relative quality rating which was found to correlate well with a quality rating derived by direct observation by a human inspector of various quality welds. The vision-based weld quality evaluation system has potential for use as a post-weld quality evaluation system, or, due to the high update rate of the overall vision system (>10 Hz), as part of a real-time control system.

Microprocessor-based arc voltage control for gas tungsten arc welding using gain scheduling

Limitations of the traditional automatic voltage controller (AVC) for gas tungsten arc welding are explained and the characteristics of the welding arc that cause these limitations are presented. The design of an adaptive AVC using a microprocessor-based controller is described. The enhanced AVC uses gain scheduling to maintain optimum gain in the feedforward loop of the system and offers a constant arc length control mode which is useful under conditions where the welding currents is varied over a wide range.<<ETX>>

Neural network methods for the modeling and control of welding processes

While welding processes are of great importance in manufacturing, their modeling and control is still subject of research. The highly nonlinear, strongly coupled, and multivariable nature of these processes renders the use of analytical tools practically impossible. In this article a novel approach is presented which employs networks of simple nonlinear units: a neural network. A widely used welding process, the Gas Tungsten Arc Welding is presented and the problem of its modeling and control is exhibited. A very brief introduction to neural networks is followed by presenting the experimental results for modeling the static and dynamic behavior of the process, as well as some practical recommendations regarding the use of the neural network techniques for controlling these processes.

Multiple-robot programming for coordinated motion, end-effector calibration, and part localization

The authors present an algorithmic solution to the problem of programming a redundant multiple-robot system for concurrent motion. Arc welding processes are used as application examples. Specifically, the weld path programming algorithm is illustrated, using the GTAW (gas tungsten arc welding) process as an example and assuming a 6-DOF (six-degree-of-freedom) welding robot and a 2-DOF positioner. The algorithm is designed to simplify, and to some extent automate, the task of programming the robotic mechanisms of the welding workcell. The algorithm treats the multiple robots as a single kinematic mechanism. Redundant joints are used to satisfy constraints imposed by the welding process. Approaches to solving the practical problems of calibration and localization are presented within the framework of the algorithm. Simulation examples for specific tasks are demonstrated and discussed.<<ETX>>

Dynamic modeling and control of nonlinear processes using neural network techniques

An adaptive network architecture of nonlinear elements and delay lines is proposed, which can be taught to model the time responses of a nonlinear, multivariable system. The structure has been applied to the modeling and control of a highly coupled multivariable process, namely, gas tungsten arc (GTA) welding. The authors present the architecture, learning algorithm, and experiments which showed the feasibility of the approach, and propose a controller architecture that can regulate a nonlinear, multivariable plant such as GTA welding.<<ETX>>

New techniques for modeling and control of GTA welding

Solutions to modeling the gas tungsten arc (GTA) welding process are presented. A scheme of modeling using an adaptive filtering approach is compared to a scheme using neural networks. The adaptive filtering scheme is a black-box modeling approach using finite impulse response filters to model the multivariable system. This approach assumes the system to be linear in the specified region of operation. The neural network approach has the potential to model nonlinearities in the system. Potential problems with this approach are also discussed. A solution to the control problem using a second neural network is also suggested.<<ETX>>

An implementation of a generic algorithm for robotic welding programming

A methodology for programming a redundant, multiple-robot system for concurrent motion is discussed. A general programming algorithm is proposed which allows the user to treat multiple robots as a single kinematic mechanism. The algorithm eliminates joint redundancy by imposing task-specific constraints on the manipulator motions. As a result the programming process is greatly simplified. The authors review the algorithm and demonstrate its use for robotic gas tungsten arc welding where appropriate constraints are imposed. Implementation and simulation results are presented.<<ETX>>