POSMEC 214 Postgrdute Symposium in Mechnicl Engineering Deprtment of Mechnicl Engineering Federl University of Uerlândi Novemer, 26th to 28th 214, Uerlândi - MG - Brzil MONITORING OF RESISTANCE SPOT WELDING PROCESS WITH DECREASED SPOT-TO-SPOT DISTANCE Yevgeni Chvertko 1, e.chvertko@kpi.u Mykol Shevchenko 1, n.shevchenko@kpi.u Andriy Pirumov 1,.pirumov@kpi.u Mksym Zierov 2, zierov@hotmil.com 1 Ntionl Technicl University of Ukrine Kyiv Polytechnic Institute, Welding Deprtment, 6/2 Dshvsk Street, BOX 56, Kyiv-Ukrine. 2 Federl University of Uerlândi, Deprtment of Mechnicl Engineering, Av. João Nves de Ávil, 2121, BOX 3-2, Uerlândi-MG-Brzil. Astrct. Resistnce spot welding is welding process with high level of utomtion nd high productivity. This rises numer of tsks relted to development of qulity evlution nd process monitoring systems operting in rel-time mode which would llow to detect non-complint joints during the process run of shortly fter it is finished. The more complex tsk is to mke such system s much universl s possile. In cse of resistnce spot welding side from the se metl chrcteristics, prts design nd welding prmeter the spot-to-spot distnce effects significntly the joint formtion. The uthors mde efforts to develop the system for monitoring the prmeters of resistnce spot welding with vrying distnce etween spots. The on-line monitoring system sed on rtificil neurl networks ws developed for evlution of process devitions. This system is elieved to e dequte for determintion of process violtions resulting in disturnces of welding prmeter nd cn e used for prediction of possile defects in the welded spots even with vritions of distnce etween them. Keywords: Stnce Spot Welding, Dynmic Resistnce, Neurl Network, Clssifiction of Joints 1. INTRODUCTION Resistnce spot welding is widely used in modern industry to produce metl structures. The process of resistnce spot welding itself requires rther high level of utomtion due to very short process time, limited ccess to the weld zone (this lso limits the possiilities for opertionl control), high (up to melting point) tempertures, etc. Usully the structures re designed so tht severl spots should e welded on them their loction eing defined y the designer to ensure the overll strength nd other mechnicl properties required. Sometimes this leds to spots loction eing different from the stndrd requirements (Gost 15878-, 19). Monitoring of resistnce spot welding processes my e orgnized upon opened or closed circuit, ech hving its own dvntges nd disdvntges for different conditions of welding (Steinmeier, 2). The type of feedck should e considered s well: the min prmeters cn e current, voltge, power. Ech of them hs its own limittions, so the finl choice is lwys eing mde s kind of compromise. It should e lso mentioned tht up-to-dte mens of qulity ssurnce for resistnce spot welding usully re designed to stilize the joints qulity while the monitoring of process dynmics still remins uncovered. This leds to oligtory use of destructive testing methods in every production-run, though defects still cn occur in finl products. Recent reserch hs proven tht monitoring of dynmicl resistnce could e good tool for qulity evlution in resistnce spot welding (Livshits, 1997). It cn e used to trck the mount of het generted in weld zone nd in this wy to predict the chrcteristics of the welded spot (Bi et l. 28). The method works in rel-time mode which is lso huge dvntge. Though the spot-to-spot distnce will effect dynmics of its chnges during the welding process which mkes it hrd to develop universl monitoring system which could operte effectively despite distnce etween welded spots. The ojective of this work ws to develop methods of qulity evlution of welded joints produced y resistnce spot welding with decresed distnce etween them y mens of monitoring of dynmicl resistnce. 2. EXPERIMENTS DETAILS The specimens to e welded were produced from low-cron steel,8 mm thick, their dimensions eing mm. Welding ws performed on МТ-1215 resistnce spot welding mchine with РКС-1 welding timer (oth re products of Kkhovk plnt of electric welding equipment, Ukrine). To void edge effects during welding the
Chvertko Y., Shevchenko M., Pirumov A., Zierov M. Monitoring of resistnce spot welding process with decresed spot-to-spot distnce welded spots were locted on the longitudinl symmetry xis, the outside ones eing plced on the distnce 2 mm from the specimen edge. According to requirements Gost 15878- (19). for metl thickness,8 mm the distnce etween spots should e 2 mm with minimum dimeter 6 mm. Experiments were performed for optiml spot-to-spot distnce nd for severl reduced ones: 15 mm, mm, 6 mm (spots re locted next to one nother) nd 3 mm (spots overlp one nother on hlf of their dimeter). For every distnce three series of experiments were performed. In the first one (the se set) welding prmeters were set up so tht despite decrese of spot-to-spot distnce the welded spots otined were 6 mm di, their vlues were chosen ccording to recommendtions (Smirnov, 2). The other two were designed to simulte violtion of welding technology which led to nd. Incomplete fusion ws otined y reducing the welding current nd y reducing the pressure together with decrese of the preliminry compression time. Mesuring system included nlog-digitl converter E-1 (L-Crd, Russi), voltge nd current sensors nd the system for primry dt collecting nd processing developed y the uthors. In ll the experiments the smpling frequency ws set t khz which is sufficient for signl recover for the welding methods investigted (Chvertko, Pirumov, Shevchenko, 213). The current sensor used ws Rogowski loop which is common one for resistnce welding equipment. The voltge mesurement ws performed etween two electrode-holders (the signl otined included the welding zone nd two electrodes). The mesuring system ws designed ccording to the following requirements: the equipment used should e uniform, with no specil djustments to use it for resistnce spot welding process; no specil trining should e required to work with the system designed; the equipment should e le to work in conditions of resistnce spot welding: short welding time, high energy pplied to the prts to e welded, high level of electromgnetic disturnces, etc.; dt processing should e considered due to huge mount of dt recorded, including intermedite opertions (integrtion, filtrtion, etc.). Due to complex dt processing needed it ws decided to use digitl system insted of nlog one. Such system is esily operted nd cn e used to store huge dt rrys nd to process them rther quickly. Dt recording ws performed using L-Grph v.2. progrmme, dt processing ws performed with MtL R213. Signls were filtered using digitl Butterworth filter. After welding of specimens with different spot-to-spot distnces nd different sets of welding prmeters destructive tests were performed y mking welded spots mcrosections (Fig. 1). 3. DATA PROCESSING c Figure 1. Welded spots: ; ; c Dt otined vi mesuring current nd voltge hd to e processed to form rrys of vlues of dynmicl resistnce which could e dequtely nlyzed y mens of rtificil intellect. Neurl networks re widely used for monitoring different welding processes Chvertko, Pirumov, Shevchenko (214) which mkes it rtionl to pply the in this cse s well. The min requirements to the dt to e nlyzed with neurl networks re (Osovskiy, 22): two dt rrys should e developed: the trining sequence used for network trining nd control one used to check the network opertion; the trining sequence should contin s mny possile ptterns of welded spots formtion s possile; dt plcement in the trining sequence should not e ligned with ny kind of reltionship; dt in trining sequence should e normlized to void chnges of welding prmeters to effect the clssifiction; the informtion should e tken from the signl form, not its vlue (Khikin, 26). According to requirements listed it ws decided to used integrted y the time equl to hlf of min power period vlues of electrode-to-electrode dynmicl resistnce. Additionlly script ws developed to identify the time intervls corresponding to joint heting nd removl of puses etween them s fr s when welding mchine is operted y humn such puses my significntly vry in time. Signls on different stges of processing re given on Fig. 2.
POSMEC 214 Postgrdute Symposium in Mechnicl Engineering Novemer, 26th to 28th 214, Uerlândi - MG - Brzil c d Figure 2. Signl processing: s recorded; filtered signl; c efore puses removl; d with puses removed Dynmicl resistnce is clculted y dividing recorded vlues of voltge nd current. As fr s Rogowski loop ws used s current sensor dditionl signl preprtion ws needed. First of ll, the signl hs very shrp edge which results in loss of dt during recording (mximum vlue isn t lwys registered, see Fig. 3, ). To compenste this script ws developed to seek the mximum vlue of signl in every hlf-period time intervl nd to replce it with mximum one, now equl to ll periods recorded (see Fig. 3, ). Results of dynmicl resistnce clcultions nd forming dt rrys for further nlysis y neurl networks re given on Fig. 4. Figure 3. Signl from Rogowski loop: s mesured; fter ugmenter Figure 4. Clcultion results
Chvertko Y., Shevchenko M., Pirumov A., Zierov M. Monitoring of resistnce spot welding process with decresed spot-to-spot distnce 4. APPLICATION OF ARTIFICIAL NEURAL NETWORKS FOR QUALITY EVALUATION Neurl networks cn e used to find nd nlyze implicit reltions etween dt rrys. In our cse the min tsk of network ws to clssify dt ccording to disturnces in welding process despite distnce etween spots nd divide them into three clsses:, nd. Two networks were chosen to perform this tsk: Rdil Bsis Functions network (RBF) nd Proility Neurl Network (PNN). Both types of networks provide clssifiction of dt rry sed upon its djcency with other rrys tking into ccount peculirities of distriution of previously nlyzed rrys. Network trining ws performed using two trining sequences: one contining informtion out 96 spots, nother 167. Lesser one (symmetric sequence) contined equl numer of every spot clss signls, igger one (non-symmetric) ws uilt s symmetric sequence plus dditionl rrys from clss joints. This ws done due to PNN network ppliction: when trining sequence contins equl numer of ech clss dt, these networks in their further work will think tht proility of otining good nd non-complint joints is equl s well. The qulity of network trining ws performed y the effectiveness fctor []. After the network trining the elements used for it were presented to the network input nd the numer of right responds ws clculted. The precision of trining sequence nlysis ws clculted from Eq. (1): B Т T (1) R where В numer of right responds, R totl numer of locks for trining. For oth networks Т T ws equl to 1. The evlution of effectiveness of the neurl networks developed ws performed y presenting to their inputs dt mtrices which were not used for trining (the control sequence). Effectiveness of the neurl network ppliction ws clculted from Eq. (2) (Dykonov nd Kruglov, 2): Т Т C Е (2) T where Т C precision of control sequence processing. As fr s for ll the networks Т T = 1, Е = Т C. Overll effectiveness of spots clssifiction (different distnces mixed in one rry) ws s follows: 1. RBF network, non-symmetric sequence (167 signls): frq1 =,9737 ; frq2 =,5247 ; frq3 =,8261 ; 2. RBF network, symmetric sequence (96 signls): frq1 =,6667; frq2 =,46; frq3 =,7232; 3. PNN network, non-symmetric sequence (167 signls): frq1 =,8131; frq2 =,5323; frq3 =,8575; 4. PNN network, symmetric sequence (96 signls): frq1 =,6667; frq2 =,4523; frq3 =,8667. Detlized dt on every spot-to-spot distnce re given on Fig. 5. It is cler tht neurl network ppliction for clssifiction of joints hs proilistic nture nd its precision is less then %. Anywy, they cn e used to detect the joints for which dditionl testing methods should e pplied. For exmple, symmetric RBF network hs shown good results in identifying spots with (more then 84 %) while non-symmetric RBF network s results prove it to e useful to detect non-complint joints. The fct tht PNN networks hs shown good results for some clsses mens tht there is certin reltionship etween those dt nd they cn e clssified (Dykonov nd Kruglov, 2). Effectiveness cn e improved y re-forming the trining sequences nd ppliction of dditionl preliminry dt processing. Also deep neurl networks operting on specil computer clusters seem to e useful in this cse ecuse of their huge numer of links.
POSMEC 214 Postgrdute Symposium in Mechnicl Engineering Novemer, 26th to 28th 214, Uerlândi - MG - Brzil % % 2 2 % % 2 2 c d Figure 5. Effectiveness of neurl networks opertion: PNN with symmetric () nd non-symmetric () sequences, RBF with symmetric (c) nd non-symmetric (d) sequences 5. CONCLUSIONS 1. Electrode-to-electrode resistnce ppers to e one of the key chrcteristics which defy the conditions of heting in resistnce spot welding nd, respectively, ffect the joint formtion. Dynmics of this resistnce chnges cn e used to evlute the qulity of welded spots. This dynmic depends not only on welding prmeters, ut on the spot-to-spot distnce s well. 2. Artificil neurl networks cn e used for qulity estimtion in resistnce spot welding with vrying spot-to-spot distnce. Anlysis of effectiveness of their opertion hs shown tht PNN network with non-symmetric trining sequence cn e used to define joints of when distnce etween spots is decresed (its effectiveness eing 81 %). If the spot-to-spot distnce is igger the est results cn e otined using RBF network with non-symmetric trining sequence (97 %). Non-complint joints cn lso e detected: PNN with symmetric trining sequence hs shown high effectiveness for (81 92 %, differs with distnce increse) while RBF network with symmetric sequence is good for detecting (up to 78 %). 3. The finl choice of network type nd trining sequence should e mde tking into ccount tsks from qulity ssurnce progrm. For exmple, the tsk could e to divide ll the joints into two groups (complint nd noncomplint), to identify the joints with only, etc. 6. REFERENCES GOST 15878-. Resistnce welding. Welded joints. Structurl elements nd dimensions, Vlid since 1.7.19. Steinmeier, D., 2, Resistnce Welding Power Supply Feedck Mode Selection, microjoining Solutions - microtipstm, Vol.1, p.152. Livshits, A.G., 1997, Universl qulity ssurnce method for resistnce spot welding sed on dynmic resistnce, Welding Journl, Vol.76, p. 154. Bi, J., Slem, M., Kuntz, M., Brown, Dr.L.J., 28, Improved Consistency of Resistnce Spot Welding (RSW) vi Power Supply Control Strtegy, Auto21 Conference. Vol.28, p. 26. Smirnov, V.V., 2, Resistnce welding equipment: Reference ook, Ed. Energotomizdt, Moscow, 848 p. Chvertko, Ye., Pirumov, A., Shevchenko, M., 213, Monitoring of the process of Flsh-Butt Welding, Soldgem & Inspeção, Vol.18, No.1, pp. 31-38. Chvertko, Ie.P., Pirumov, A.Ie., Shevchenko, M.V., 214, Monitoring of welding processes with ppliction of rtificil neurlnetworks, Bulletin of the NYUU KPI, No.2, pp. 88-93. Osovskiy, S., 22, Neurl networks for informtion processing, Ed. Finnce nd sttistics, Moscow, 344 p. Khikin, S., 26, Neurl networks, Khrkov, Vilims, 3 p. Dykonov, V.P., Kruglov, V.V., 21, Mthemtic pplictions for MtL, Ed. Piter, Sint-Petersurg, 4 p. 7. RESPONSIBILITY NOTICE The uthors re the only responsile for the printed mteril included in this pper.