Issue link: https://iconnect007.uberflip.com/i/1345210
12 SMT007 MAGAZINE I MARCH 2021 perhaps resolvable with a refinement of the rules within the closed-loop or machine learn- ing soware algorithms. e real loss of trac- tion, however, is caused by significant external changes, such as: • e change of any specific materials being used from one supplier to another, introducing a small dimensional change • A new batch of PCBs that have slightly different positional distortions inherent to their manufacturing processes • Just a change in the product being produced as the schedule for highly mixed production is followed Incorporating Data Expanding our horizons once more, to include logic around these sources of vari- ance, means bringing in MES factory-level data, allowing the feedback analysis soware to understand the root-cause of many more sources of variation and associated potential effects, enabling smart soware to go beyond its current limitations, based on the existing scope. is next step puts us higher again on the AIQ scale, as we are incorporating more data into the decision-making process—again made possible using CFX-based best practices to eliminate needless costs and complexities. All the steps discussed so far are referred to as being "smart," but clearly there is quite a dif- ference in the degree of smartness, depending on each implementation, bearing in mind that this was just a simple example. Each successive step brought us up the AIQ scale, but has each step really been "smart" in terms of incorpo- rating intelligence? e soware still cannot think for itself; it is following pre-programmed rules and flows, just like the phone and sewing machine. It is simply the inclusion of more and more data, allowing more and more algorithms to be developed and applied. Some argue that people appear intelligent simply because of the sophistication of the sheer amount of data that we capture through our senses, and how the rules within us have evolved as we interact back to make sure that the operational move- ment meets expectation, avoiding excessive or incorrect actions. Machine Learning Expanding on that principle of direct feed- back in a smart way is machine learning (ML). For example, an inspection machine will mod- ify its assessment of passes and fails as a result of changed settings determined from analy- sis of prior judgements. is expands out to become "closed-loop" feedback where two or more different machines are involved. For example, deviations in X, Y, and rotation posi- tion of placed SMT components are measured by an inspection machine. en, by an analy- sis of patterns and trends in the data, param- eters are changed to reduce the deviation in placement. Subtle live alterations to the place- ment process can be done automatically, or the alarm is raised to call for an operator to replace a worn nozzle, for example. Instances of defects are then avoided by not allowing any deviations to go beyond control limits. ese smart technologies, available today, are used to correct and refine actions that are performed, enhancing the effect of the origi- nal sensors. It is interesting to note that prac- tical implementation of these steps has been enabled by expanded communication of data between machines that are likely to have come from different vendors. Technologies such as IPC-CFX (Connected Factory Exchange) have revolutionized the ability to share and utilize actionable data in a singularly defined format and meaning, without the need for middleware and IP exchange. is has been smart. We are certainly now higher up the AIQ scale, as dem- onstrated in this simple example, by the order of magnitude of defects found at test. But the losses are not yet reduced to zero. Our degree of "smart" is not yet perfect. Are we not curious as to why these closed- loop technologies do not produce perfect yield? Clearly there is more variation going on than we thought, some of it subtle, and