A capable process fits well inside its tolerance and is centred on it. Narrow the spread or bring the mean back to the nominal value, and Cpk rises.
Normally distributed process, tolerance ±50 µm: Cp = (U − L) / 6σ, Cpk = min(U − μ, μ − L) / 3σ.
Senior Scientist | R&D Project Manager
Institute of Production Engineering and Photonic Technologies, TU Wien
Measuring is half the work. Quality assurance decides what a measured value means for the part and for the process, and feeds that decision back into production.
Integrated quality assurance treats measurement, evaluation and reaction as one system. A measured value is recorded, compared with the specification and turned into a decision: accept the part, correct the process or look for the cause.
Statistical quality assurance provides the tools for that decision: process capability indices such as Cp and Cpk, control charts, and measurement system analysis, which shows whether a measuring system can resolve the variation it is meant to monitor. I teach quality management at TU Wien; before my research career I worked as a project and quality engineer at a supplier of Mercedes-Benz buses.
Process-integrated quality assurance moves these checks closer to the machine: measurements in the process, at the machine and in the cell, linked by computer-aided quality (CAQ) and product and manufacturing information (PMI). A digital twin of the quality task keeps each measured value attached to the part and its nominal model. Without measured data, a twin is only a model.
A capable process fits well inside its tolerance and is centred on it. Narrow the spread or bring the mean back to the nominal value, and Cpk rises.
Normally distributed process, tolerance ±50 µm: Cp = (U − L) / 6σ, Cpk = min(U − μ, μ − L) / 3σ.
Move the sliders to change the spread and the position of the process. Cp compares the width of the tolerance with six standard deviations of the process. Cpk also takes the position of the mean into account and looks at the nearer tolerance limit. Many customers ask for a Cpk of at least 1.33; a centred process with this value leaves about 63 parts per million outside the tolerance.