It happens again and again that bonds produced in a series process no longer meet the requirements placed on them. In the rarest of cases, this occurs in the form of failure of all bonded parts, but only in a more or less high percentage. The lower this percentage is, the more difficult it is to determine the toot cause for the failure.
In one specific case, the parts manufactured by a supplier at just under 1‰ failed to bond completely, either during further assembly to the finished product or shortly after delivery to the customer. With production volumes of several thousand parts per day, the seemingly low failure rate of 1‰ is understandably unacceptable.
On the other hand, at such a low failure rate, a process parameter cannot be “just” changed to check whether the error can be eliminated. In order to be able to make a statistically reasonably reliable statement, several thousand parts would have had to be manufactured according to the changed conditions and tested. Apart from the associated costs, this would have meant that almost a day’s production could not have been delivered to the customer. Also an increase in the random inspection of parts carried out within the framework of the final inspection of finished goods would have made little sense.
In such cases, troubleshooting is more like detective work. In a first step, the documentation of the incoming goods inspections carried out, the in-line inspection of important process parameters and the inspection of work samples produced during production, as well as the results of the finished goods inspection must be reviewed and correlated with the occurrence of the defect. With a bit of luck, initial approaches can already be seen here, e.g. the first occurrence of the defect could coincide with the batch change of a raw material or a drift in a process parameter could be observed, which in turn correlates with the defect frequency.
With such low error rates, however, the cause is usually due to an unfavorable coincidence of various parameters, each of which alone does not lead to the error. In addition, as a rule not all process parameters can be documented. The “human factor” must be considered here in particular. Even if hardly any worker will willfully do something wrong, errors can occur simply because he carries out a work step somewhat differently than his colleagues.
My approach in such cases is to observe the production over a longer period of time, i.e. over several shifts, in order to detect any individual details that may be unfavorable for the bonded joint. In the case described above, several potential causes of defects were identified in this way and appropriate corrective measures were proposed. After their implementation, the error no longer occurred.