Analysis Based On The Signal Characteristics Of Spot Welding Quality Online Monitoring Method

The on-line quality monitoring of spot welding is desiderated in automobile assembly process. In the process of resistance welding, the signals of voltage and current reflect both the input energy and the dynamic variation for joint impedance of forming the nugget. The random influence factors of welding quality must be present in the variety of welding signals directly or indirectly, but can't be observed simply because of its indetermination, non-linearity and coupling with each other. Purposed on on-line monitoring and controlling of resistance spot welding quality,the modern analysis methods of signals were adopted to analyse the characteristcs of dynamic welding current and voltage, and picked up the informations in them. Artificial Neural Networks(ANN) were used to predict the nugget sizes of resistance spot welder. The work was done as follows.A signal collection system was developed with KS2062. The welding voltage and current were measured synchronously, which carried out displaying and presetting with the C+ +. Time-domain, frequency-domain and wavelet were used to analyse current, voltage, displacement signals in this thesis in order to research the relativity between the character of signals and the process of resistance spot welding. Enriched the recongition of the process of welding by the point of view signal analysis, the basic will be established for welding defect on-line identifying and quality classifying. It is concluded that the waveform of current, voltage signals and the variation of dynamic resistance, power of joint have related to nugget formation closely, so they can be used for on-line monitoring and controlling of joint quality of resistance spot welding. Because the variation of the signals of voltage and current is not obvious in frequency domain, the signal characters were stressly analysed and picked up in time domain.The two different neural networks were used to predict the nugget sizes of resistance spot welder, in which the input vectors were constructed by the time sequences of cycle parameters normalized. The common BP Network was analysed. The effective methods were advanced to improve the rapidity and accuracy of the model, which have predicted nugget sizes availably. The Radial Basis Function (RBF) neutral network is a non-linear map by altering the parameters of non-linear activation functions of neuron so that it can improve the rapidity of networks and avoid the local minimality. On the other hand, the RBF neutral network can realize the classification of multi-dimension because it adopts the multivariate interpolation way based on the Radial Basis Functions. Therefore, the RBF neutral network will act a significant part in on-line monitoring and controlling of welding quality.