Online control of stencil printing parameters using reinforcement learning approach

This research proposes a novel approach to control the stencil printing process (SPP) parameters online in surface mount technology (SMT) of printed circuit boards (PCBs). Several external variables induce variations in stencil printing quality including environment conditions, operator faults, and others. This research aims to build an optimal adaptive controller that captures these variations and consequently adjusts the controllable and significant printing parameters to enhance the solder paste volume transfer efficiency (TE) during actual production run. Q-learning which is a reinforcement learning (RL) approach is used to control the main printing parameters (printing speed and pressure, and the separation speed) online. The results show that Q-learning converges to the optimal policy for the SPP problem, and the optimal sets of actions for different states are retrieved using Q-table. Moreover, the developed controller is capable to reach the terminal state for several testing examples with taking few actions. (C) 2018 The Authors. Published by Elsevier B.V.