While 2019 ended on a fairly positive note for the mechanical engineering industry in Germany, expectations for 2020 were scaled back considerably in light of the global coronavirus crisis. A number of production slumps were due to the collapse of finely tuned supply chains. As a result, companies are once again focusing on the search for hidden savings potential. At the end of the production chain in particular, final quality control is an area that has a significant impact on the market position, reputation and cost recovery of the entire company. Deevio wants to use AI-based image processing to automate this process, which is often still carried out manually - and also save investments in image processing systems that have already been purchased.

The dilemma of final quality control

The situation to date has been as follows: On the one hand, numerous, otherwise highly automated production companies still rely almost exclusively on the decisions of their comprehensively qualified inspectors when it comes to final quality control. These inspectors have an excellently trained eye, years of experience and the ability to distinguish an actual production error from a supposed defect or tolerable variation.

Nevertheless, even the most attentive and motivated employees sooner or later suffer from fatigue and hunger and can no longer concentrate or are absent for other reasons. However, these factors - which are completely understandable from a human point of view - have a merciless effect on the quality of the products, which then often go straight to the customers. In addition, companies in structurally weak areas in particular are finding it increasingly difficult to find qualified employees who are able and willing to take on this enormous responsibility. The documentation of decisions made by quality inspectors is also often a major problem. If the iO or niO criteria cannot be traced in the required level of detail, no optimization options can be identified and implemented.

On the other hand, many investment managers are finding that the hopes raised by the acquisition of rule-based, automated image processing systems in the area of final quality control are not being fulfilled. In fact, the reject rate has risen to over 20 percent due to pseudo defects. With these results, it is no longer possible to cover costs - quite apart from the amortization of the consistently high acquisition costs, which is at least a medium-term goal. Although these systems have very powerful and high-resolution cameras, their purely rule-based approach means that they cannot keep pace with the range of variants that actually occur. Experience has shown that optics that are set too sharply, for example, sort out parts that are actually free of defects simply because a reflection of light or a color fluctuation has been misinterpreted as a scratch. Nevertheless, rule-based processes do offer some specific advantages that make them more suitable for use during or at the start of production.