RESONANT TRANSITION SWITCHING WELDING POWER SUPPLY N. Frohleke, H. Mundinger, S. Beineke, P. Wallmeier, H. Grotstollen Institute for Power Electronics and Electrical Drives University of Paderborn, FB 14 - LEA Pohlweg 47-49, 33095 Paderborn, Germany Phone: ( 4 9 ) 5251-603157 Fax: (+49) 5251-603157 E-mail: nfroel @lea.uni-paderborn.de Abstract - By means of parallel connected passive networks, charged only modestly by energy for commutations, resonant transition switching is ensured for this modified full-bridge topology used in a welding power supply. A new driving scheme adapts the resulting power circuitry for both the droplet and the short-circuiting transfer welding modes occurring in the gas metal arc welding process. Less voluminous and cheaper magnetic components together with lower conduction losses of transistors and diodes yield a cost effective, light weight power supply for 12kW l350A. The analysis of the circuit, the design and the verification results are presented. I. 1NTRODUCTION Modern high-frequency welding power supplies (WPS) feature low weight, robustness, high reliability and flexibility at reasonable prices. Publications [ 1,2,3,4] demonstrate that above mentioned features have already been achieved for low power welding supplies by soft switching topologies. For the higher output power required for the gas metal arc welding (GMAW, MIGMAG) process, however, classical resonant converters are not optimal. Some of their drawbacks are high reactive power, high conduction losses, and a bulky tank inductor as revealed in [5,61. In order to extend the range of high frequency WPS to GMAW applicatioic (e.g. 10kW) without the need to parallel smaller units, a modified resonant transition switching fullbridge is proposed. The proposed full-bridge is denoted as Passive Resonant Commutated Poles (€33-PRCP), following the terminology used i n [6]. The F€-PRCP is based on the pseudo-resonant full-bridge [7], but avoids the high peak current charge of the commutation inductors by employing the transformers leakage inductance as well as a modified driving scheme. Thus the size of the inductors, the current charge of the switches, and the reactive power circulating in the converter is reduced. Pulsed gas metal arc welding power supplies While the gas tungsten arc welding (GTAW) process requires a constant welding current, the GMAW process needs for a self-adjusting arc a power supply with a constant voltage characteristic [8]. For puls-welding WPS, however, the welding current is pulsed to steer the droplet transfer. Therefore, the power stage is embedded into a currentcontrol loop as shown in fig. I . The set point for the current control-loop 0-7803-3932-0 is governed by a slower voltage control-loop, which is not shown, with pulse-generating features. As the current set point will change rapidly, the current control-loop must be fast. For a currentcontrolled power stage a current sensor is obligatory. Hence, the new control scheme can use the current signal at almost no cost penalty. inverter (ZVS) [deadtime adiustm. t t t t I set point current controller I I Fig. 1 Current-controlled welding power supply The publication divides into the following major chapters: After explaining the operating principle, a circuit analysis is summarized, which leads to designs equations for the control signals for two modes of operation. Finally selected simulation results are given for a WPS, which was built as an example prototype featuring the above mentioned characteristics and additionally a good efficiency at 12kW/350A supplied from 400V three-phase mains. Calculated and simulated results were proved by measurements. 11. OPERATING PRINCIPLE The FB-PRCP shown in fig. 2 is made up of two legs A and B, a center tapped transformer with a small discrete inductor Ls, the output rectifier, and the output filter inductor Lf, which includes the wire inductance. At steady state the GMAW welding arc can be represented by a resistor Ra and a voltage source V, [I]. Leg A consists of an IGBT-half-bridge (Qap, Qan) with two freewheeling diodes (Dap, Dan), two small resonant capacitors (Cap, Can), an inductor La and two auxiliary capacitors ( C I , C2). Leg B is identical but uses different part values. The auxiliary capacitors are relatively large (a few pF) and provide approximately half the input voltage. In order to obtain constant relative current ripple, the filter inductance Lfis nonlinear. As shown in fig. 3, the IGBTs are driven in a phase-shifted 615 Leg A I ’ 1 i N +?’ State: P P - N P ( p a r t 1 I Leg B I N P-N N P P - N P (part! I I I Qap Qan I Q ~ P Vin , , 1 Qbn . . : , , . , , NN-PN (partl) N N - P N (part!= , I . . . . ~ I I I PN-PP . I , , . . I 0 I I I I I I Fig. 2 Power circuitiy of FB-PRCP with load equivalent manner, so that each leg voltage (vcu, vcb) has a duty cycle of about 50% and the volt-second-product across the transformer primary winding is proportional to the phase-shift dr. The time duration tcu and tCh for commutation in legs A and B depend on the load current and are adjusted to maintain ZVS for the IGBTs. I I *I le dt . Q ~ P . . . , 1 I . I I , I On off ‘O‘la‘2a ‘ 3 ‘4 I I 8 , , , , \ , I , I , ‘ 5 ‘68 ‘7a ‘8 ‘9 t ‘IO Fig. 4 Typical waveforms at heavy load On Off mutation intervals are named with a pair of two letters, according to the initial and final network state. On Qbn I , On Off to t3 t5 f t,, Fig. 3 Control signa’s for FB-PRCP At light load, the energy stored in La and Lb ensures the commutation of the resonant phase legs from rail to rail similar to the FB-ARCP [9] but without the need for switching devices with symmeirical blocking capability and associated driver and logic circuitry. At heavy Zoad, the energy of Ls is predominantly utilized to commutate the resonant capacitors. Compared to the conventional phase-shifted resonant full-bridge Ls is much smaller for a given minimum load current, and therefore the effective duty ratio - thus also the transformers turns ratio - can be increased. Fig. 4 shows typical waveforms for heavy load conditions. These waveforms d o not differ much from the conventional phase-shifted resonant full-bridge, as the commutations are primarily based on ih. The conversion cycle divides into 10 topological states indicated in fig. 4 and fig. 6. Network state PN denotes that Sap and Sbn conduct resulting in respective transistor voltages. Note, that equal symbols NN or PP denote freewheeling and NI’ or PN denote energizing states. Com- 111. CIRCUIT ANALYSIS For steady state analysis of the FB-PRCP the following assumptions are set: a) semiconductors, capacitors, and inductors are ideal, lossless components b) the transformer is replaced by its magnetizing inductance Lm and an ideal transformer with tums ratio n. The leakage inductance is included in the value of inductance Is, and the auxiliary capacitors CI, C2, C3, and C4 are equal and large enough to be replaced by voltage sources of V,, /2 c) the phase-shift dr is restricted to values suitable for converter operation d) the output current is perfectly smoothed; this assumption holds, since the welding currents ripple must be small e) the commutations are fast enough that i h ( t ) , i d t ) , and i L d t ) are constant during commutation intervals. With these assumptions, fig. 5 results, representing the equivalent power circuit, where Ca = Cap+Can and Cb= Cbp+Cbn. For simplification reasons the ideal transformer is eliminated and the output current is transferred to the primary as Zup= /,Z n, with n being the turns ratio of primary to secondary (np/ns). 616 Finally, at uCrrbecomes zero and the energizing interval NP (fig. 6d) begins. C. Energizing interval NP: t 2 I t I t j As soon as the freewheeling of the output rectifier ends at t / b or t2,, power is transferred to the output as depicted in fig. 6d. The voltage across Lm is divided due to the the ratio of Lml(Lm+Ls), whereby with reduced LmlLs a lower output voltage results. Hence, the designer should not reduce I,, by selecting a small Lm. At t3, which is identical to dt+to , the modulator terminates the energizing interval NP by opening switch Sbp; ucb swings from V,, to zero voltage. Fig. 5 Equivalent circuit A. Abbreviations and Acronyms o,:=JZz, w2:=J7GZ3%- D. Commutation interval NP-NN: t3 5 t 5 t4 During commutation interval NP-NN fig. 6e represents the power circuit. Assuming ib is approximately constant with iLb(t3)= ILb3we obtain B. Commutation interval PP-NP: to 5 t I t z The analysis starts at the end of the freewheeling interval PP, hence the initial conditions are as follows: switches Sap (transistor) and Sbp (diode) are conducting the output rectifier is freewheeling, thus shorting the transformers magnetizing inductor. immediately after opening switch Sap at to fig. 6a represents the power circuit. With assumption e) iLFand uca result as: (1) i , ( t ) = I , cos(wo(t - t o ) ) - I ~ o and ucu(t)= U,,, - Z o l , s i n ( ~ ~ ( t - t ~ ) ) . (2) Mode A: Under heavy load, i.e. if I , , is large, Can is rapidly discharged and the freewheeling diode of San starts to conduct at 1 tlu = -arcsin i L s ( t )= -zZ2 sin(w2T)-lxb cos(w2t")+ I,,, (7) and ucb(t) = U,, c0s(02t")-Z2lxb sin(o2t") . (8) I where t:=t-t3. Applying sin(x) z x and cos(x) and (8) to i,(t) = -~ Ls+ Lm 2 f for x << f simplifies (7) t" - ( I a p + ILmO) and uCb(t)= Ui,- - tIxb . Cb The commutation NP-NN is completed as soon as comes zero at uin tq =- (10) UCb be- Cb+t3 . Ixb (3) WO with circuit topology depicted in fig. 6b being active until when iLr5 I, - lap. The freewheeling of the output rectifier is terminated. Mode B: If lap is small, the freewheeling of the output rectifier ends before uca becomes zero. Thus, starting at iLFand iLocompete charging and discharging Ca, as shown in fig. 6c. E. Freewheeling interval NN: t-, I t I t5 During freewheeling intervals no power is transferred to the output and the output current is freewheeling. Theoretically, ih and ib would sustain their level at the end of commutation interval NP-NN. In reality, however, the voltage drop across the rectifier diodes will occur on the transformers primary winding and add to the voltage drop across the switches to reduce the currents, and, more inconvenient, cause losses! (A lot of work has been invested on reducing these freewheeling losses. For cost-sensitive applications with wide input voltage range none of them seemed suitable. E.g. it is not possible to use saturable inductors for inductors La and Lb, because even small variations in their part value or the input voltage would cause large variations of their peak current.) 617 F. Symmetries: 1 5 I t I t l o The FB-PRCP is symmetrical i n regard to upper and lower half of the legs, but, due to the driving scheme, there is no symmetry between leg A and leg B. The symmetry within the legs can be exploited to derive waveforms for the time interval r5 to r/,]. Because the auxiliary capacitors (CI...C4) balance the currents ,i and iLh, some steady state peak values can be approximated as transition topologies and have [nus been investigated i n more detail. Fig. 7 shows the principle waveforms of ucu and iLrfor heavy and light load modes. In order to generate proper gating signals it is important to find the respective mode for a given I,. When the mode of operation changes from A to B, t l , and t l h are equal, and so are rlh and t2h. Using (5)and (6) yields where Iuphmis the limit of Iq between modes A and B. As (12) is based on the assumption of short commutation intervals, the currentli have to be chosen somewhat higher to compensate for losses due to finite commutation time. G. DC voltage tnznsfer ratio For modes A and EI the average voltage across the rectifier output can be derived as r +dt V, ( 0 ) = k . - t l U ] for mode A and (13) - [%- - 2 K,, Lm Ts n L s + Lm is constant for a fixed input voltage. As t l b , tru, tZhrand t4 all depend on the load current, ( 1 3) and (14) are evaluated numerically. IV. DESIGN ASPECTS The existence of two load dependent operation modes A and B makes the FB-PRCP different from other resonant for mode B, where k = d) t2 ... 9 In mode B the inductor current I k o has to ensure the commutation of leg A. The critical Iup is denoted Iaphm,because at lower output current Ih,, and at higher current iLs will accelerate the discharging of Ca. Thus, Ik0 has to be larger than ( Iuptim-IDnO),which leads to a simple but important equation. In the original pseudoresonant full-bridge the peak current through La must be above the maximum primary transformer current to achieve soft switching. In the FB-PRCP, however, it can be chosen to meet requirements set by the designer. Selecting a large Cu, thus represents an effective turn off snubber, gives a low Z,, affecting a high Izo on one hand, hence Ih0 must be high. On the other hand, Iolo can be lowered by selecting a larger Ls for the price of a smaller volt-seconds-product at the transformers primary winding. Though a larger Ls will have a negative effect on the converters overall performance, it limits dildt for the transformers primary winding and the rectifier diodes as positive side effects. t) ' 4 e) '3 ... t4 ... '5 Fig. 6 Topological network states for the FB-PRCP and notation (network states for ts...tlo are not shown): f5.4 Commutation NN-PN; initiated when Commutation PP-NP; initiated when modulator modulator opens Sun(with Sun comprising opens Sup (with Sup comprising Qup and Dup) Qan and Dan) at t5 = %Ts; part 1 at to; part 1 Commutation NN-PN; part 2; heavy load Commutation PP-NP; part 2; heavy load t6a..t7u Commutation NN-PN; part 2; light load Commutation PP-NP; part 2; light load r6k..r7h Energizing PN Energizing NP t7 ...t# Commutation PN-PP; initiated when Commutation NP-NN; initiated when modulator t8...t9 modulator opens Shn at t8 = %Ts + dt opens Sbp at t j = dt Freewheeling PP Freewheeling NN tg ...t / o 618 p,y:.; large and switches of leg B should be gated as soon as the voltage across them is zero. vm .....,........ ....... .. ,_,,._ V. SIMULATIONS t v.m- xI a 20 .....4 ....: LI...CL.. I I Extensive PSpice simulations were performed before building a prototype WPS for 350AllOkW from 400VAC three-phase mains, out of which one is depicted i n fig. 9. Circuit data were: La= I20pH. Lb= 400pH. Cup= Can= Cbp= Cbn= IOnF, n= 12, Ls= 12pH. Lm= 0.7mH,fs= 4OkHz, V,,= 540VDC 1.- - - - - - - II I I 2 '0, .. I 2.0, Fig. 7 Typical waveforms during commutation PP-NP Delay times tca and f c b First, switches and diodes were modeled by conduction losses and linearized switching characteristics. Losses in passive components were modeled by appropriate e m . Second, manufacturer provided semiconductor models were used using IGBT-half-bridge modules SKM7SGBIOfBN and MUR20020CT rectifier diodes. Delay times tca and tCh have to be adjusted according to the load current. Hence, to maintain ZVS of leg A, Qan may not be gated before ucu is zero; using (3) and (6) gives this minimum time delay tCa. An upper limit for rCa is given, because the Qan must be gated before the anti-parallel diode Dan stops to conduct. Typical ,,may-be-gated" and ,,has-to-begated" curves for legs A and B are depicted in fig. 8 for the prototype converter as specified in chapter V. The gap between ,,may-be" and ,,has-to-be" of leg A can be controlled by the inductance of La. The lower La, the bigger the gap, and the larger the window of tolerable part values. Selecting a wide gap even allows to use a constant tCa. -6OOV 1 - 360~s Fig. 9 *..-.... / "may be gated" leg B -.._ *...._. -e... ....._..*. ......._ ..__.+*.. 3701.15 380~s 390~s 400~s Simulation results at steady state and at nominal operating point VI. EXPERIMENTAL RESULTS A prototype was built with part values given in chapter V, I I I 1 b using the UC387.5 controller and analog circuitry for the load 1OOA 200A 300A load current[A] dependent adjustment of the dead-times. Fig. 10a to fig. 10d Fig. 8 Range for gating signals show waveforms of transistor Voltages vca and VCb. resulting voltage at the transformers primary winding "Tr* and primary For leg B there is only a minimum delay time t C b , Because Current ~ L Fat different load conditions. Both vca and V C b beZrh depends o n Zq, the duration of the commutation interval have like predicted by the analysis, providing ZVS for the NP-NN can be estimated from (1 1) as IGBTs. Note, the slight increase of ih during energizing U,, .Cb U,, .Cb St4 -t3 5 . (17) states, mostly caused by the magnetizing current i b . A drop ILb3 I L m O -IIapMar ILb3 + I,, of ih can be observed during freewheeling intervals, as preThe bigger the peak resonant current ILb3 the smaller the dicted by simulations. Values for diLr/dt are largely limited maximum commutation time. Note, that based o n Zho the due to Ls, and almost n o voltage overshoot of the leg voltages circuit would work even without any Lb. The commutation is visible. Fig. 1Oe shows the zoomed commutation PP-NP, time, however, would be very long and limit the usable range with ih reversing its direction before uco decays to zero. Nevfor dt, so that it is not suitable for proper circuit operation. ertheless, Cu is discharged and ZVS is maintained. Finally, in Therefore, to obtain a large range for dr, Lb should not be too fig. 10f transistor voltage ucu and the auxiliary inductor cur"may be gated" leg A **'*... 619 VIII. ACKNOWLEDGEMENT The authors gratefully acknowledge the support by Cloos SchweiRtechnik, Haiger, Germany. IX. REFERENCES [I] L. Malesani a. o., “Electronic Welder with High Frequency Resonant Inverter, ” IEEE Trans. Industry Applicutions, vol. 31, no. 2 , March/April 1995, pp. 273-279. [2] T.-F. Wu, a. o., “Analysis and Design of Variable Frequency and Phase-Shift Controlled Series Resonant Converter Applied for Electric Welding Machines,” in Proceedings of the 1995 IECON, pp. 656661. [3] P.C. Theron, a. o., “Welding power supplies using Partial Series Resonant Converter,” in Proceedings ofthe 1995 IECON, pp. 1319-1321. [4] H. Pollock, J.O. Flower, “Series-parallel loadresonant converter for controlled-current arc welding power supply,“ in IEE Proceedings-Electric Power Applications, vol. 143, iss. 3, May 96, pp. 2 1 1-21 8. [ 5 ] R.L. Steigerwald, a. o., “A comparison of high power DC-to DC soft-switched converter topologies,” in Proceedings of ?he 1994 IAS, pp. 1090-1096. [6] R. Fanington, a. o., “Analysis of reactive power in resonant converters,” in Proceedings of rhe 1992 PESC, pp. 191-205. [7] O.D. Patterson, a. o., “Pseudo-Resonant Full-Bridge DClDC Converter, ” IEEE Trans. on Power Electronics, vol. 6, no. 4, Oct. 1991. Fig. 10 Measurements: a) at load current 20A, b) I50A, c) 250A, d ) 375A (resulting in ari output power of 13kW), e ) commutation PP-NP at 20A, f) inductor current iL, at 1SOA rent ib is shown, which proves that during commutation intervals of leg A. i k is almost constant VII. CONCLUSION A full-bridge topology for the gas metal arc welding process is introduced featuring high power density in conjunction with low costs. This is achieved by using resonant transition switching of the four transistors and relatively low reactive power processing due to parallel connected passive networks, which are only modestly charged by the energy for commutations. The low cost fe.iture arises momentarily by omitting the symmetrical blocking devices incl. drives of the branch for commutation in the auxiliary resonant pole converter (whose costs need to be weighted against the passive means for commutation.) The so called passive resonant commutated poles full-bridge (FB-PRCP) is facilitated by a new low cost driving scheme adapting the resulting power circuit according to the load conditions. 620 [8] J.F. Lancaster, The Physics of Welding, Pergamon Press, I.ed., 1984, p. 150. [9] R.W. De Donker, J.P. Lyons, “The Auxiliary Resonant Commutated Pole Converter,“ in Proceedings of the IAS 1990, pp. I 128-1235.