Neuro-fuzzy algorithm for quality assurance of resistance spot welding

Resistance spot welding is widely used in the field of plate assembly. However, there is currently no satisfactory nondestructive quality evaluation for this type of welding either in real-time or on-line. Moreover, even though the rate of welding under conditions of expulsion has been high until now, there is still no established method of quality control against expulsion. Accordingly, this paper proposes a quality assurance technique for resistance spot welding using a neuro-fuzzy algorithm. Four parameters from an electrode separation signal in the case of nonexpulsion and dynamic resistance patterns in the case of expulsion are selected as the fuzzy input parameters. These parameters are determined using a neuro-learning algorithm, and then are used to construct a fuzzy inference system. When compared with the real strength for the total strength range, the fuzzy inference values of strength produced a specimen error within /spl plusmn/4%, plus the percentage of specimen errors within /spl plusmn/1% was 88.8%. The tensile-shear strength limit for electrically coated zinc is 400 kgf/mm/sup 2/. When evaluating whether the quality of the welding was good or bad according to this criterion, the probability of misjudgement that a good quality weld was a poor one was 0.43%, and the reverse was 2.59%. Finally, the proposed neuro fuzzy inference system can infer the tensile-shear strength of resistance spot welding with a high efficiency in cases of both nonexpulsion and expulsion. It is also anticipated that an on-line welding quality inspection system will be realized in the near future.