Date of Award
Master of Science (MS)
The objective of this research is to model the relationship between force, sound, and motion signals in manual assembly environments through a wearable sensor glove and the resultant quality of vehicle connections made on the assembly line. Many tasks in production assembly are still completed manually due to the intuition needed by the associate, complex automation steps, or time constraints. This is largely observed in automotive assembly environments. With the amount of variability in manually completed processes, the possibility for error increases. These processes include hose and electrical connections which can loosen over time after passing initial quality testing, resulting in costly, time-consuming rework and a diminished brand image. It is the intent of this work to utilize multidimensional operator force signatures and movements exhibited to understand the primary forces acting in the direction of the connector locking and additional measured forces acting in other directions. The sensor signals feed into the classification algorithm for rapid postprocessing to enable real-time feedback indicating a completed connection or a connection that needs further investigation. These classifications can later act as a steppingstone for automating manually completed manufacturing processes by implementing the findings into autonomous systems to yield an automatic verification of the process. This research captured data physically exerted by the operator as a means of accountable process quality evaluation where there are limited marketable products and research. The work also introduced a sensor glove system capable of capturing operator applied shear force in a robust and durable way fit for a manufacturing environment. Marketed products and research shear force sensing are extremely limited in breadth, and force sensing gloves are unsuitable for an assembly environment due to cost, measurement capabilities, durability, and/or operator encroachment. The sensing system developed in this research is coupled with a classification algorithm capable of discerning incomplete or rework connections from successful ones demonstrated on an OEM assembly line. The developed sensor glove capable of capturing shear and normal force, acceleration, and gyroscopic information was successfully tested on an OEM assembly line for 250+ vehicles of work. This includes the completion of hard plastic connections, tool usage, and tasks completed outside of the takt. Five classification models using the gathered data yielded accuracies of 91% or above using a 60/40 train/test split. The best performing model, Na¨ıve Bayes, achieved a balanced accuracy of 97.6%.
Kerner, Scott, "Evaluation of Manually Completed Manufacturing Assembly Processes Through a Wearable Force and Motion Sensing System Integrated Into a Glove" (2022). All Theses. 3875.