Title :Manufacturing System Realization

Our research focuses on the integration of manufacturing system CAD/CAM models with statistics-based methods for design, control and diagnostics of multistage processes.  To this end our research addresses the following:

• Developing statistical methods driven by engineering models (CAD/CAM) to achieve manufacturing system quality improvement during system ramp-up.

Recent studies show that major challenges in design and implementation of complex manufacturing systems are related to system failures and quality deterioration during ramp-up phase. As high as 72% of root causes of process variation in automotive assembly are due to fixture failure [2].  However, relatively little was done on systematic fixture failure diagnosis during ramp-up of a new manufacturing system.

Traditional quality improvement methods do not provide systematic approaches for fault root cause diagnostics. Until recently, existing methods for quality improvement were primarily based on statistical analysis of in-process measurements detecting process change rather than determining root cause. Such methods are not always effective for complex systems such as auto body assembly and may not be applicable for new product/process launches (ramp-up phase) due to the unavailability of historical statistical data, pre-enumerated faults or identified case studies.  Yet, diagnostics methodologies consider limited discrete definition of fault (fault/no fault or 1/0), which is insufficient for quality improvement and ramp-up of new manufacturing systems. 
 

Our research has led to the development of a methodology for quality improvement by root cause diagnosis of dimensional variability in multiple station manufacturing processes (MMP) [1, 3, 17, 20].  This requires developing statistical methods driven by engineering models (CAD/CAM) to isolate fault root cause.  This is a departure from classical approaches in statistic, which utilize previously captured data or solved case studies.  Additionally, fault definition is expanded to cover cases where fault is understood as a continuous function spanning between 0 and 1. This integration holds a promising direction for quality engineering research and is especially relevant given the lack of historical data in new and short-run manufacturing processes.

We proposed a new methodology for in-process dimensional fault diagnostics and quality improvement of multi-station body assembly systems by combining process variability detection, isolation and root cause determination [1, 3, 12, 16, 17, 20].  In this research, a new hierarchical model of assembly architecture and sequence is developed.  This represents a multi-level body assembly system [1, 3, 17, 52, 59, 60] based on which a set of correlation clustering and diagnostic reasoning techniques are studied.  These techniques are based on the developed fixture failure model (CAD/CAM) and its relationship with measurement data represented as principal components analysis (PCA) [3, 17, 55].  Our research proved the inherent relationship between fixture failures and PCA [3, 55].  This builds a foundation for fixture fault diagnosis that further led to the knowledge-based dimensional fault diagnosis methodology, used during the launch of new automotive body assembly processes. The implementation of this methodology helped to: (a) accomplish the 2-mm dimensional variation level for automotive body, achieved for the first time by a US auto assembly (JNAP/DaimlerChrysler); (b) receiving the 1998 J.D. Powers award by Bramalea Plant, DaimlerChrysler for best quality vehicle in the US market.  Additionally, it became part of the Chrysler Operating System (COS) in 1998 for all new car and truck launches and of GM’s “BIW Data Analyzer” software.

We feel that our work provides a theoretical basis for process variability diagnosis for sheet metal assembly. This has created the dual benefits of continuously improving quality while at the same time reducing overall product development time through rapid fault root cause detection. It also provides a model for optimal sensor placement for single fixture [9] and for multistage manufacturing processes (MMP) with end-of-line sensing [7] or with distributed in-process sensing [11, 23] (please also see (2) Sensor Networks in Manufacturing). This further led to evaluation and optimizing of assembly design [8, 19, 38, 44, 46] and tolerancing study [21, 28, 10, 55 ] presented in (3) Integration of Product and Process Characteristics.