Application Of RBF Neural Network In PCB Drilling Process

PCB just printed circuit board, it's very important to semiconductor industrial since it provides connection for electronic components. PCB applied electronic print technology that's why it is so-called"printed"circuit board.Paul Eisler, Austrian, the inventor of PCB. He applied printed circuit board in a radio receiver device in 1936. American applied this technology in radio receiver for military use in 1943. And 1948, US officially ratified this invention in business use. From the 50s of the twentieth century, PCB technology has been put to use widely. Before that, connection of electronic components relies on wire directly. Nowadays, PCB technology occupies a dominate position in electronic industrial.PCB is the most important and active link in electronic industrial, It is the biggest subsection of the whole industrial since it's output value take 1/4 of it. Normally, the growth of PCB is 3% higher than electronic component industrial. After the burst of Dot-com bubble in 2001, PCB industrial got completely recovery with globe IT industrial in 2003. Chinese PCB output value hit 50 billion RMB in 2003, 32.4% increasement over the past year and also leap to second place in the world. China keep 30% annual growth rate till 2007 while PCB got considerably down trend in 2008 and 2009 due to financial crisis. Semiconductor industrial rebound in 2010 while PCB just got 10.5% increasement. But Chinese PCB back to 30% high speed growth own to industrial relocation. 2011, lot of uncertain factors still limit the PCB.Upgrade requirements has been the strong impetus for PCB. In order to adapt the trend of latest electronic product like compact, lightness and multifunction, requirements for next generation of PCB is high density, high integration, embedded and multi-layer. HDI, subdued, IC assembly products will be the new growth point. Their commonality is high density. Therefore, accuracy has to be improved accordingly for PCB process. The purpose of drilling process is providing connection between layers, so the drilling accuracy determine the PCB accuracy.This paper start with brief introduction of PCB process, then analyze the drilling process in depth and the root cause for hole shift. After then, list all the factors for hole shift and find out the key factors by using statistical method. And last, decide to use the RBF neural network to solve the problem by comparison with other tools. This paper specially focus on how to use Matlab Toolbox to buildup RBF neural network model and how to use RBF tools to analyze the hole shift and model optimization and network training. According to the drilling data of last 3 month and analysis for RBF theory and key factors, finally decide to buildup the RBF model by using collet torque force and spindle run out as input and X/Y axis hole shift as output. Reset the process spec base on the result, and conduct DOE for 3 lots. Result shows that the new parameters from the model can control the hole shift well. That means this model has high practical value.