Data-Driven Approach to Using Uniform Experimental Design to Optimize System Compensation Parameters for an Auto-Alignment Machine

This paper proposes a data-driven approach using uniform experimental design (DAUED) to optimize system compensation parameters for an auto-alignment machine. The proposed DAUED optimizes system compensation parameters by integrating uniform experimental design with the best combination of parameter values and a stepwise ratio. First, system compensation parameters and a 10-level uniform layout (UL) are used to perform alignment experiments. Second, the best result for the 10-level UL experiments is selected and combined with the stepwise ratio for use in computing the experimental range of each parameter for the next 10-level UL experiments. The steps are repeated until the alignment count remains unchanged. Experiments in industrial examples showed that, compared to the conventional industrial design method, the proposed DAUED requires fewer experiments to obtain the system compensation parameters that minimize the alignment count. For example, to achieve a required alignment accuracy to within $5~{\mu }\text{m}$ , the DAUED can obtain the best system compensation parameters in only 30 experiments and with an alignment count of 1. In addition, in 30 independent runs using the best compensation parameters, the mean and the standard deviations in alignment counts are 1 and 0, respectively. That is, the best system compensation parameters are robust and stable. In contrast, the industrial design method previously used by engineers requires more than 200 experiments to obtain the system compensation parameters, and its alignment count is as high as 4 or 5. In conclusion, compared to the conventional approach to optimizing system compensation parameters, the proposed DAUED is superior in terms of efficiency (i.e., it requires fewer experiments), effectiveness (i.e., it has a lower alignment count), and robustness (i.e., it achieves a standard deviation of zero in alignment count) in online, real-time, and high-precision optimization of an auto-alignment machine.