Statistically Testing Inner Layer Yield Improvement Projects 

Estimated reading time: 2 minutes

Can we trust our measurement system to give us reliable data? Is it accurate, repeatable, and reproducible? Measurement is the foundation of quality. We measure for two primary reasons: to make decisions on product quality and to provide data that will inform continuous improvement projects. We can engage in continuous improvement projects if we are confident in our measurement systems.

What is the primary skill for which we are on the payroll? It is our problem-solving skills. Therefore, our function is to solve problems and add value. Dr. Joseph M. Juran stated, "A project is a problem scheduled for solution."¹

Projects need to be connected to business strategies. Their importance must be clear to the organization. They must have the support and approval of management and be tied to the organization's financials. They need a reasonable scope, typically doable in under six months for Plan-Do-Check-Act (PDCA) or Define-Measure-Analyze-Improve-Control (DMAIC) projects. Note that sub-projects may be required first. Project goals must be well-defined, and there must be clear quantitative measures of success. Projects need to be looked at from a statistical point of view.

Statistical thinking views all work as a process and acknowledges that variation exists in all processes. Knowledge and management of variation are crucial for sustained success. For any given process, the output (Y) is a function of the inputs (X): Y = f(X₁, X₂…X_n) + error, where the error is assumed to be random variation.

Statistics and Probability

Statistics is a branch of mathematics that collects, analyzes, interprets, and presents numerical data. It is not merely the science of analyzing data but the art and science of collecting and analyzing data. Statistics are tools to help us; they do not replace the process engineer’s skill and intelligence. Statistics uses hypotheses, experiments, and hypothesis testing.

With hypotheses, a null (H₀) and an alternative (H₁) are put forth. The null hypothesis states that two population means or variances are equal. The alternative hypothesis is a statement that the two population means or variances are not equal (two-tailed). However, the alternative hypothesis can be either one or two-tailed. The one-tail tests for either inferiority or superiority, while the two-tail tests for parity (not equal). Typically, we test for parity, which tests for both inferiority and superiority.

To read this entire article, which appeared in the December 2024 issue of PCB007 Magazine, click here.