Unsupervised machine learning application in the selection of measurement strategy on Coordinate Measuring Machine

It is indisputable that some type of coordinate measurement system (CMS) is generally used to assess the quality of dimensional and geometric characteristics. Considering the required accuracy, flexibility, and speed of measurement, a CMM with a scanning sensor may offer the best performance. These measurement systems are very complex, and many factors affect the reliability of the measurement results. A Metrologist’s choice represents the greatest variability in the measurement strategy. Previous research has shown that the measurement results can be changed up to 100 % by choosing a different measurement strategy when evaluating the form error. This paper conducts a detailed study of the impact of the measurement strategy on the cylindricity error when measuring eleven workpieces with the same nominal characteristics, but different real characteristics described by roughness and the reference value of cylindricity. To examine the influence and importance of certain factors and their levels, design of experiment (DoE) and unsupervised machine learning techniques of PCA (Principal Component Analysis) and Multiple Correspondence Analysis (MCA), were used. The results suggest that depending on the real geometry of the workpiece, different factors with different percentages influence the output characteristic. The objective was to choose a uniform measurement strategy when measuring cylindricity on the CMM, while the prior information about the actual geometry of the workpiece is lacking.