Issue link: https://iconnect007.uberflip.com/i/1305670
NOVEMBER 2020 I SMT007 MAGAZINE 53 data may suggest a PCB condition—for exam- ple, stretch or twist, which is an issue around a specific nozzle that has been used—or be related to the specific part itself. Each of these factors comes with its own set of rules as to what can or cannot be tolerated, as well as what potential adjustments in any preceding processes may be applied. In reality, all of these effects happen con- currently with different levels of contribu- tion. With all the analysis done, that is not the end of it, as certain aspects of trend calcula- tions need to be reset when, for example, PCB packs change, nozzles are cleaned or replaced, materials replenished, machine parameters adjusted, etc. A truly challenging AI applica- tion running within a truly big data environ- ment, with data from design, the supply chain, machine programs, MES, and other sources all needed to provide a holistic machine-learning or closed-loop solution. And this was a "sim- ple" example. It is not quite a done deal as to whether we can say that "defects are his- tory" already, though a lot of people are work- ing hard on these algorithms from many direc- tions, so we get closer and closer. Business-wise, to get the buy-in to invest in smart factory inspection and test machines— as well as IIoT-based MES solutions—we need to create a convincing argument that shows that they are an essential and intrinsic part of the value-added process. Direct machines today already include operations that are not strictly of direct added value. Reading fidu- cial locations on the PCB increases the accu- racy of placement, as does the taking of the picked-up component to a camera for recogni- tion and alignment before placement. A signif- icant amount of machine run-time can be ded- icated to these tasks, which are in addition to the basic pickup and placement. As a whole, the industry sees these func- tions as being essential to the SMT placement operation, and that it is only through the use of these technologies that the machines have become capable of placing the newer, smaller, or higher pitch components successfully. The addition of post-placement inspection and test could be positioned as being an evolution of this. The only difference is that it takes a sep- arate machine to perform the test and inspec- tion, as well as external software for the machine-learning, line-learning closed-loop, or factory-learning at the MES level. However, they are all essentially the same in that they facilitate increased direct performance of the line itself. As we have seen with the state of the tech- nology as it evolves, the transition of test and inspection to become a zero-defect driver as opposed to being a filter of defects may not yet be black and white. For some time, there will be an element of both happening. As line and factory layers evolve, the expectation is that defects will be reduced and eliminated over time, as the data captured from test and inspection is improved, linked with increasing sources of data—such as that from MES—and algorithms at the machine. The process has to start somewhere, though. Smart factory management must invest in ini- tiatives that strive to make the transition from defect-based quality management to zero-defect quality management through the use of test and inspection data. I have seen yield losses reduce by an order of magnitude already through the use of software that utilizes test and inspection data to improve placement accuracies—a ben- efit that simply cannot be ignored. Test, inspection, and IIoT-based MES soft- ware that support machine, line, and factory- learning belong in the must column of any As we have seen with the state of the technology as it evolves, the transition of test and inspection to become a zero-defect driver as opposed to being a filter of defects may not yet be black and white.