100t焊接滾輪架設(shè)計(jì)-總體設(shè)計(jì)和從動(dòng)滾輪座設(shè)計(jì)【說(shuō)明書+CAD】

100t焊接滾輪架設(shè)計(jì)-總體設(shè)計(jì)和從動(dòng)滾輪座設(shè)計(jì)【說(shuō)明書+CAD】,說(shuō)明書+CAD,100t焊接滾輪架設(shè)計(jì)-總體設(shè)計(jì)和從動(dòng)滾輪座設(shè)計(jì)【說(shuō)明書+CAD】,焊接,滾輪,架設(shè),總體,整體,設(shè)計(jì),以及,從動(dòng),說(shuō)明書,仿單,cad Online evaluating on quality of mild steel joints in resistance spot welding 33 Online evaluating on quality of mild steel joints in resistance spot welding Zhang Pengxian and Chen Jianhong 張鵬賢，陳劍虹料 Abstract A method was developed to realize quality evaluation onevery weldspot in resistance spot welding based on information processing of artificial intelligentFirstly，f signals o廠welding current and welding voltage，as information source，were synchronously collectedInput power and dynamic resistance were selected as monitoring waveform$Eight characteristic parameters relating to weld quality were extracted from the monitoring wavefortTt$Secondly，tensileshear strength 0廠the spotweldedjoint was employed as evaluating target ofweld qualityThrough correlation analysis between every two parameters of characteristic vector，five characteristic parameters were reasonably selected to found a mapping model 0廠 weld quality estimationAt last，the model was realized by means 0廠the algorithms of Radial Basic Function neural network and sample matrixesThe results showed validations by a satisfaction in evaluating weld quality of mild steel joint online in spot welding process Key words resistance spot welding，weld quality，characteristic parameter，quality evaluating，Radial Basic Function neural network 0 Introduction Resistance spot welding(RSW)is one of the most dominant sheet metal joining processes applied in the auto mobile，aerospace and other appliance industriesSeveral factors may influence the joint quality，such as the weld tip geometry，dirt and corrosion on the electrodes andor on metal surfaces，and fluctuation of AC supply voltage Especially，the invisibility and instantaneity in the process of a nugget formation cause weld quality instable and diffi cult for detectionll-41 Today，one of the nlajor problems faced by the process control engineers is the lack of a reli able，lowcost，nondestructive technique for identifying bad welds and predicting weld strengths in real time The signals of spot welding process involve abundant information which is directly or indirectly correlative with the nugget formation and joint qualityPatange et a1I s suggested a method of evaluating weld quality as an in process system with the dynamic resistance monitored by using a microprocessor in the secondary circuit of the welding machineHao et a1 。 performed the multiple linear regression analysis on the characteristic analysis of aluminum RSW to estimate the nugget diameter and weld strength，which might be a lowcost，non-destructive and effective approach to estimate quality based on the infor mation processing in RSW In the paper，mild steel sheet metal ST1 2 was select ed as substrate in resistance spot welding processin which welding current and welding voltage were collected on real time and treated as the basic information source Therefore，a method was explored to evaluate weld quali tv 1 Monitoring signals and its capturing Spot welding machine YF0201 Z2 was used in the ex- periments，in which the welding current and welding volt age were synchronously collected by meads of the data col lecting systemThe system is made up of sensors，the cir euit of signal obtained and converted，AD card and com putersHall Voltage Transducer was selected to measure welding voltageRogowski coil with frequency response of This paper is supposed by National Natural Science Foundation of China(No50275028) $Zhang PengxianKey Laboratory of Nonferrous Metal Alloys and Processing of Ministry of Education，Lanzhou University of Technology，Lanzhou，730050 Email：zpxlutcn(Zhang Pengxian) Chen Jianhong，State Key Laboratory of Gansu Advanced Non ferrous Metal Materials，Lanzhou University of Technology，Lanzhou，730050 34 CHINA WELDING Vo117 No4 1)ecember 2008 1 00 kHz was used to measure welding cun-ent of which the measure scope can meet with gathering for welding current under 100 kAA 12一digit converter of AC一61 15 was se lected as AD card，of which the synchronous sampling rate of the card is 40 kHzAt the same time，MATLAB software was used to analyze signal characteristics and set up network reeognition mode1For ease of measure and compare，all weld experiments were accomplished under the same conditionThe work diameter of the electrode tip is set as 6 mmwhereas the welding time was fixed as 30 cvcles antthe electrode force as 3 800 NThe sample plates，after welding，were cut into the same shape，of which the plate is 130 mm in length，30 nlnl in width， with the thickness varied from 07 Him to 14 mmFig1 20 l0 0 1O -20 7 6 4 3 lO 6 2 0 0 5 10 15 20 25 30 Welding time Tcycle (a)Welding current wavetorm shows all example of the synchronous waveforms of welding current and welding voltage at the welding current of 4 000 A Heat input，power input and dynamic resistance waveforlns were selected as monitoring objectsFirstly， average effecti ve values of welding current and welding voltage in every halfcycle were monitored and defined as ， and U respectivelyThus，the power input P is equal to muhiplying j by U Dynamic resistance Rp is equal to the corresponding voltage divided by the current at the time of the peak point of welding current waveformThis algo rithm can elirainate the innuence of n1utual inductanee l 。 Fig2 shows the waveforms of， ，U ，P and RpWhen welding process was interfered by external or internal fac一 2 0 -2 0 5 10 15 20 25 30 Welding time Tcycle (b)Welding voltage waveform Fig1 Signals of welding current and welding voltage 0 5 10 l5 20 25 30 Welding time?Tcycle (a)Welding CHirellwaveform 0 5 10 15 2O 25 30 Welding time 77cycle (e)Input power waveform 14 12 1O 08 06 04 0 5 10 15 20 25 30 Welding lime7cycle ()Welding vohage wavefornl 0 5 Fig2 Monitoring waveforms of， ，U ，P 10 15 20 25 30 Wediug time Tcycle _ 。 一一。 一 一 )i k 。 一 一 一。 一I)一 |，lllJ0音l1I】 IJ 一 i。0I】IJ旦 蘭 u】 0 I壬()0 T1(1 _ m r e V a e v a 啦 S e r C m a n V D ) d p R d n a Online evaluating on quality of mild steel joints in resistance spot welding 35 tors，there would be some singularity changes in these waveformsFig3 shows the four waveforms monitored un一 7 6 5 4 3 4 2 0 0 5 10 15 20 25 3O Welding time Tcycle (a)Welding current waveform O l0 15 20 25 30 Welding time Tcycle (C)Input power waveform der the condition of which the fluctuations of AC supply voltage appeared in the welding process 14 12 10 08 06 04 220 160 140 0 5 10 15 20 25 30 Welding time Tcycle (b)Welding voltage wavefoFIn O 5 lO 15 2O 25 30 Welding time Tcycle (d)Dynamic resistance waveform Fig3 Monitoring waveforms of singularity changes 2 Extraction of characteristic parameters Because the waveforms of dynamic resistance and power input are related to nugget formation of spot weld ing，characteristic parameters extracted from the wave forms will give clearer expression to metal mehing and nugget growthTotal heat input Qt and average value of in put power P directly indicate variation of heat source in welding process SO that they were selected as main parame tersSix characteristic parameters are extracted from dy namie resistance waveformFig4 is a sketch map of the parameters describedHerein，t represents time rate， namely the ratio of time when peak value appears in dy namic resistance waveform to total welding timetOt repre sents heat effect caused by internal resistance of welding area in current welding conditionsThe peak value of dy- namic resistance Rp is a beginning sign of the metal melt ing d represents dropout rate after the peak indicating nugget growth and expanding degree of plastic circleR is resistance value when power supply is cut off，as it can monitor nugget size and plastic circle sizeR is gap be tween Rp and R R is average value in welding process 180 140 120 0 tO 5 10 15 20 25 30 Welding time Tcycle Fig4 Sketch map of characteristic parameters extracted 3 Algorithm of quality evaluating 31 Mapping model of joint quality Mapping pattern and algorithm were founded on a large number of data samplesTable 1 shows a certain A 0 )I I10-I T10 II一 一0 j 。II 一 】一_【葛時(shí)蠱 口 0 8 6 u 享0cI 4IIH i |】IIB1s1啦2 0一魯 II 36 CHINA WEI DING VoI1 7 No4 December 2008 number of arrays which colle from monitoring signal on four different experiment conditionsEvery array in Table 1 re presents certain conditionvc304001 is an example to ex plain naming rules of the arrayHerein，VCindicates the experimental condition of which the supply voltage is unstable in welding process；3040indicates the welding current as 3 000 A with welding time 40 cycles；01is an array numberThusvc304001一vc304030is is sued 30 groups of data in the same experiment condition Each array is made up of42 dimensional vectors(， ，t)， (U ，t)，(R ，t)and(P，t)，where sample matrixes are composed of characteristic vectorsTable 2 shows a form ing process of sample matrix in normal welding condition In Table 2，characteristic vectors are made up of tensile shear strength r，of welding spot and characteristic parame- tersTable 3一Table 5 show the result of correlative anal ysis On four kinds of welding conditions in Table 1 Table l Data samples of monitoring signals with different weld conditions Table 2 Forming process of sample matrix Table 3 Result of correlative analysis between parameters in the characteristic vector Table 5 Result of COrrelative analysis between parameters in the characteristic vector C0rrelative coefficient ris listed in the tables，which shows that the bigger the value of叼is，the more distinct the correlative degree isThe biggest value of 叼 is 1000WhereI一1Vrepresents four welding COn。 diti0ns mentioned in Table 1Based on the value of叩， five characteristic parameters Qt，t，R ，R and R are selected to map 32 Realization of evaluating algorithm Based on neura1 network toolbox of MATLAB，Radial Basic Function f RBF)neural network was used to set up a n0nlinear mapping model from Qt， t，R ，R and R to RBF network is a type of forward network on basis of fll】nction approach theoryIts learning process is to look for bestfitting plane of data samples in hyperspace And it is better than BP network in terms of learning rate， the function approach capability of network and perform ance 0f Dattern reeo gnition and classification So， RBF network is designed to evaluate welding qualityQt， t， R ，R and R are selected as input vectorsRadbas，a kind of Gauss function，is selected as transfer function of hide laverThere is a nerve cell in output layerTransfer functi0n 0f the 0utput layer adopts a kind of linear function purelinIts structure expression is as follows： 口 =radbas(Il 一P ll 6 ) 。 =purelin(W 。 +b。) (2) Where，。 i represents the Noi element。f。utput Veetor。 in the hide layer；W is the weight vector of nerve cell in the Noi hide layer，ie，the No row of weight matrix； p the input vect。r；b the thresh。ld；and 0 the。utput vec- tor of output layer 4 Validation of evaluating algorithm In order to train and simulate RBF network，240 groups 0f data samples were obtained from welding test Before trainingevery numerical value of data samples was coITespondingly normalized to l 0，1With training tar get err0r set as 000 1，the nerve cell number in hide layer is 5Through iterative learning and training，a stable net w0rk structure was set upTo validate the availability of e valuating algorithm，linear regression analysis lS accom Dlished between 0utput of network simulation“Aand tar get 0utput of testing value“V”Fig5 shows the result of 38 CHINA WELDING V011 7 No4 December 2008 linear regression analysis，where，R represents correlative coefficientits value is 0928and this indicates that there iS unifoFruity between A and VThis result SHOWS strategy of quality evaluation is feasibleAnd it can get sarisfactorresult that RBF network iS employed in weld quality estimation 50 30 30 40 50 Fest vahle VMPa Fig5 Result of linear regression analysis 5 Conclusions In resistance spot welding process，weld quality is mainly decided by nugget formation processDynamic re sistance，power input and other signals contain abundant dynamic informationThe characteristic parameters extrac ted from the waveform of monitoring signals can be made as the parameters reflecting weld quality The strategy of quality evaluating is a feasible based on characteristic parameters，sample matrix and cmTelative analysisThis method is effective to reduce the dimensions of information space and to estimate weld quality in time The validation result shows that RBF network algorithm can realize quality estimation of mild steel joints and attain satjsfled resuh RefeFences 1 Gedeon S A，Sorensen C D，Ulrich K T，et a1Measurement of dynamic electrical and mechanical properties of resistance spot weldingWelding Journal，1987，66(12)：378s一 385S 2Dickinson D W，F(xiàn)ranklin J E Characterization of spot weld ing behavior by dynamic electrical parameter monitoring Welding Journal，1980，59(6)：170s一176s 3 Cho H S，Rhee SExperimental study of nugget formation in resistance spot weldingWelding Journal，2003，82(4)： 195s200s 4Tsai C L，Dai W L，Dieknson D WAnalysis and develop一 nent of a realtime contml methodology in resistance spot weldingWelding Journal，1991，70(12)：339s一351S， 5Patange S R，Reddy G PMicroprocessorbased resistance spot welding monitoringWelding Journal，1 985，64(12)： l70s一176s Hao M，Osman K A，Roomer D R，et a1Development in characterizalion of resistance spol welding of aluminum Welding Journal，1996，75(1)：1 S一8s 窆 0 一 H鼉一 0 cI 0