Machine learning is a field of artificial intelligence (AI) that allows a computer-controlled software or system to take decisions and learn based on the analysis of empirical data from a database or physical sensors. The ease of use of Vi TECHNOLOGY’s 3D SPI is made possible thanks to machine learning. The Pi series 3D SPI inspection systems’ revolutionary ergonomics and award-winning programming simplicity are made possible using patented machine learning algorithms. The principle is simple but radical: only information that the system cannot detect or reconstruct by itself is requested from the operator:
• In order to optimize solder paste detection, the lighting level is automatically adjusted according to the color of the printed circuit board, white or black. In addition, the color and therefore the shape and position of the screen printing are also learned during the programming phase. • The machine is also able to learn how to recognize glue dots when scanning the first printed circuit board and automatically optimizes its internal parameters accordingly. • The same applies to the recognition of “skip marks”: the system learns to identify them at the first inspection, regardless of the shape of the sticker or marker trace on subsequent boards. • the precise “stop” position of the board conveyor, regardless of the geometry or position of the board on the conveyor, is calculated in real time without any operator-managed learning being required.
Unique algorithm measures exact height of paste deposits The R&D team at Vi TECHNOLOGY developed an algorithm allowing the system to learn how to locate the paste deposits, without relying only on the location patterns, which are often insufficient due to the stretch or warpage of the board. Moreover, this algorithm allows to measure precisely and individually the “0” height reference point of each pad taking into account the copper thickness, in order to obtain an extremely accurate height measurement of each solder paste deposit.
This patented algorithm gives the PI series its exceptional performance in terms of accuracy and repeatability, while featuring ergonomics and programming simplicity that are unique on the market, offering the lowest operating costs.
3D AOI programming assisted by machine learning Noticed at the recent Productronica exhibition, the new software platform of Vi TECHNOLOGY’s K3D AOI systems now integrates new generation algorithms for programming.
• Auto MatchMaker: When creating the inspection program for a new board, all the components available in the library are automatically recognized by this new algorithm. There is no longer any need to do this search manually, one by one: the automatic component recognition takes care of it and the programming time is almost halved.
• Auto Teach: For the addition of a new component in the central library, the machine learning algorithm recognizes the shapes of the body and leads, if any, and creates the reference model for inspection by associating the required AOI tests. This new model is then submitted to the operator for approval before being saved in the central library.
Thanks to this technology, the quality of 3D AOI inspection programs is constant and no longer depends on the skill level of the programmer.
Machine learning guarantees the accuracy of 3D AOI Only a system capable of adapting in real time to the measured physical conditions is in a position to meet the challenges of increasing complexity of electronic boards and the miniaturization of components, respecting the takt time of the production line while preserving first pass yield.
The K-series 3D AOI integrates machine learning technology to;
• Provide outstanding analytical capability: The scan of each board is instantly adjusted to its actual topography, and not to the theoretical or average one. This technical feature can only be achieved when the machine is able to measure the deformation of each board in real time and adapt its behavior independently.
• Achieve a first pass yield of 99%: An ultra-telecentric optical column with a sub-pixel resolution of 4.75µm in XYplane feeds the algorithm with extremely accurate input data and images allowing it to deliver results with a metrological quality. This gives the system a high process stability at pre- or post-reflow inspection, with an optimal test coverage.
• Optimize inspection cycle time: During the programming phase, it is necessary to associate re-flattened areas with their components, and to deduce the best inspection path offering the shortest cycle time. Thanks to our POPE algorithm (“PCBI Optimal Path Estimation”), the machine adapts its behavior according to what it needs to scan and calculates the optimal acquisition path without degrading scanning quality.