A Parallel Distributed Processing Technique

---

By: Charalabos Doumanidis, Nikos Fourligkas, Brian Marquis

[Intoduction][Process][Modeling and Control][Conclusion][Reference][Back]

Abstract:

Traditional welding methods with a concentrated, sequentially-moving heat source generate a standard temperature field which does not allow for independent control of all weld-quality attributes. The new scan welding technique provides a continuous, simul taneous heat distribution on th eentire accessible weld surface, as this is swept by a rapidly reciprocating torch, with its power and motion modulated inprocess, based on thermal feedback from an infrared pyrometer and a distributed-parameter adaptive c ontrol strategy. The process produces the full bead leangth simultaneously, and its smooth, regulated temperature field results in closely-controlled favorable thermal weld properties as well as several productivity benefits.

In modern welding practice, the process conditions, such as the torch power and motion, are selected according to empirical recommendations, in order to obtain the specified quality characteristics of the final weld, such as the bead geometr y, weld material properties and residual stress and distortions. However, since all weld features result from the same welding thermal field developed by a single, localized, sequentially moving torch, they cannot be specified independently of each other. This coupling between the weld features is an impediment for process optimization, and clearly compromises the overall weld quality and productivity. Since this is a consequence of the lumped-source process arrangement, multiple torch schem es, with independently modulated heat inputs, have been recently proposed [1]. Howecer, their actual multitorch implimentation is rather impractical and costly, and they provide only limited additional flexibility in the welding conditions.

An alternative to this classical mechanized, concentrated-source, sequential welding method, involves a simultaneous, continuous heat distribution on the weld. The innovative scan welding technique impliments this distributed heat imput by fast rep etitive cycling of a single robot-driven torch along the entire accesible weld surface. By power modulation of this timeshared torch, this novel parallel method provides any specified heat imput distribution through scanning of a raster patt ern on the weld surface, or through vector motion of the torch on dynamically scheduled trajectories. Thus, its advantage over conventional multitorch methods lies in its flexibility of the thermal boundary conditions on the weld surface, ensuring a continuous, dynamic reshaping of the resulting temperature field history, which generates the desired geometric, amterial, and stress/strain characteristics of high-quality welds. In contrast to traditional serial welding, where the weld bead is produce d in longitudinal steps by a steep, nonuniform temperature hill, the rapid scan weldign reciprocation of the torch covers the total weld length, and deposits the entire weld bead simultaneously, in progressive cross-sectional increments. In addition, as i t can be seen, scan welding can generate a unifor thermal field in the centerline direction, with smooth temperature gradients, which yields favorable thermal properties of the joint.

Scan welding is dynamically described by analytical, numerical, and experimental modeling methods for the temperature, phase and stress field in the weld. A real-time quasiliniear analytical model is obtained by spatial-temporal convolution of the thermal conduction of Green's function, which may be derived in closed-form, developed computationally or measured on the weld surface. For more flexibility, a finite-difference numerical simulation of the process provides for separate meshes for the weld bead a nd heat affected zone, as well as for materials with temperature dependent properties and latent transformation effects. This computational model can accomodate various heat loss effects from the surfaces and convective heat transfer in the pool through e quivalent conduction [2]. Last, off-line experimental measurement of the thermal responses of the scanned weld during various power transient tests formulates a linearized, autoregressive moving average (ARMA) model with variable parameters, due to proces s nonlinearities, thermal drift of material properties, and various disturbaces in weld condtions.

These models serve as a foundation of a closed-loop, distributed-parameter, adaptive control system, with real-time identification of model parameters, through feedback of non-contact surface thermal measurements. The torch trajectory planning and power a djustment strategy drives the heat source along the loci of maximum deviation of the measurment from the specified temperature field. this scheme is augmented by external loop filtering with anti-reset windup and by a Smith-predictor scheme to cope with long transportational delays. After computational validation of the closed-loop system performace in following setpoint commands and rejecting various disturbances, an automated scan welding prototype is currently implemented in the laboratory, based on r obotic Plasma Arc Welding (PAW) of thin stainless steel (304) plates. The experimental setup employs an infrared pyrometry camera as the thermal sensor, and the timeshared torch reciprocation is obtained by a dedicated servomechanism.

[1] Doumanidis C, "Modeling and Control of Timeshared and Scanned Torch Elding", ASME Journal of Dynamics Systems, Measurementr and Control (accepted for publication).
[2]Marquis B, Doumanidis C,"Distributed-Parameter Simulation and Adaptive Control of Welding", IASTED Intl. Conference on Modeling and Simultion, Pittsburgh PA, May 1993