Latin square design for chip length machine vision measurement system analysis

ABSTRACT Machine vision systems are widely employed to provide data for analysis and interpretation in many industries, where measurement system analysis (MSA) is used to ensure the validity of these data. In a common MSA, each part is measured by each operator several times, resulting in a two-factor crossed design. However, when other factors in machine vision system must be considered, the MSA becomes more complex. In a chip length vision system, chip locations, chip placement angles, and variability among different chips have been identified as factors that influence measurement variability. A full factorial crossed design on this vision system would call for 1536 trials and cost about 32 hr, which is unacceptable to the practitioners. However, by designing a replicated Latin square experiment, which omits the interaction effects, the number of trials is dramatically reduced to 96 runs. Then, ANOVA is used to estimate variance components. %P/T and are the criteria to finally determine the capability of this machine vision system. The Latin square method for MSA is verified by simulation and applied to a chip manufacturing application.