Measurement scheme and classification methods for the development of a 'Product DNA' concept in manufacturing.

Today's manufacturing is characterized by an abundance of data continuously collected across the manufacturing process and throughout the product's life cycle. Inspired by the Human Genome Project, this research proposes a 'Product DNA' concept to extract the useful information from the product quality characteristics, which will define a product's unique characteristics to help the process fault diagnosis and product functionality prediction. To ensure the successful development of the 'Product DNA' concept, this research focuses on the measurement scheme and classification methods for dimensional and surface quality characteristics in powertrain manufacturing processes. A novel measurement scheme to collect dimensional quality data is first proposed to ensure the acquisition of an appropriate 'Product DNA' with the minimum amount of time and effort. A methodology to reduce the total measurements on one single part by alternating the measurements of various feature groups in order to increase the part sampling frequency is designed. In addition, a practical solution to use the dimensional genome in the 'Product DNA' and identify process faults is investigated. A procedure to deal with the diagnosibility problem with the 'Stream of Variation' model has been designed. Unique estimates of the root causes from part quality measurements can be obtained by representing the information contained in a large set of process fault parameters with a smaller array of variables named station-level errors. Furthermore, a method to characterize the product surface patterns with histogram estimators of surface parameters for the use of surface classification is proposed. Compared with the conventional characterization method, more surface pattern details are captured and the classification accuracy is improved with the proposed method. Finally, a set of dimensional reduction and coefficient shrinkage methods are introduced in case the dimensionality problem is encountered in surface classification. Results from a case study illustrated that not only is the classification accuracy improved but also the selection of the most significant surface parameters influencing the classification results could be useful for process fault diagnosis.Ph.D.Applied SciencesMechanical engineeringOperations researchUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/126604/2/3253444.pd