Study On On-line Weld Quality Monitoring For Resistance Spot Welding Based On Neural Networks

Resistance spot welding is used extensively for the fabrication of sheet metalassemblies, especially in the automobiles and aviation fields because of its particularadvantages. The process of resistance spot welding is regarded as a nonlinear,uncertain dynamic time-varying process and a complicated controlled course.Simultaneously, for a spot welding system, it is not easy to measure with on-line theoutput, the nugget size by usual measuring method. And it is also difficult to controlor identify the welding process with on-line because the welding process is very shortand the output is non-measurable. Many countries dived a mass of manpower andmaterial resources to research, and obtained some development.There are many technical parameters and other parameters effect on the qualityof resistance spot welding, including welding current, welding time and electrodeforce. Because of these factors, the welding process becomes complex. Especially, theinstantaneous and the non-visibility of the welding process make the controllingquality of the spot welding very difficultly. The traditional controlling ways havemore or less localizations. So to get a reliable capacity controlling way and system hasbecome important which scholars regard to.As the artificial intelligence developing, the expert system, the ANN and thefuzzy controlling are becoming mature. So the scholars get to use these to monitor thequality of spot welding on real-time repair the fluctuation because of diversifiedfactors to get eligible welding spot.It has important significance to apply intelligence controlling into the field ofwelding has. In the welding process, the factors are coupling with each other. Becausethe traditional controlling ways need accurate math former, to get perfect results withthe traditional controlling ways is very difficult. This former is approximate, even it isrough, and that in order to advance the controlling precision, it often makes thecontrolling system complex and recedes the real-time quality of the controllingsystem.The Neural Networks copy human's cerebra function. It has abilities ofdiscursion, judge and study, and it can be used to gain on any sequence functions.Because the welding process is indetermination, non-linearity and coupling with eachother, to describe the relation between the quality and parameters of spot weldingbecomes difficult. But we can solve this problem with the Neural Networks. We canmonitor the quality of welding spot with trained Neural Networks.To build a system which has dynamic electricity parameters as its input and hasthe capability of the nugget as its output has important significance. As the NeuralNetworks becomes more and more perfect, it has been applied to many welding fields,such as the choice of the technical parameters, the controlling of the welding processand so on.This paper describes the now-development of monitoring and controlling thespot welding, and points out that the popular monitoring spot welding with artificialneural networks can over come the flaws of monitoring with traditional ways.This paper analyses the relations between the parameters of the dynamicresistance and the dimension of nugget, and selects the first-rank input parameters forthe neural networks.In this study, a system for real-time quality inspection of resistance spot weldingwith artificial neural networks is suggested, aiming at the lag circs in this field in ourcountry. The mostly function of this system is to acquire the quality communicationsof the spot welding process and inspect the dimension of spot welding nugget.Thereinto acquiring the quality communications is realized with PC7483 and thesoftware of C language and inspection of the dimension is realized with artificialneural networks. We established and trained the networks, and then we replanted thedatum of the networks into C language to establish the system of real-time inspectionof nugget diameter for resistance spot welding. We also made out some elementaryexperimentation to test the system, and the results indicate that the system issteady-going.