A SYSTEMATIC STRATEGY FOR OPTIMIZING MANUFACTURING OPERATIONS

A manufacturing optimization strategy is developed and demonstrated, which combines an asset utilization model and a process optimization framework with multivariate statistical analysis in a systematic manner to focus and drive process improvement activities. Although this manufacturing strategy is broadly applicable, the approach is discussed with respect to a polymer sheet manufacturing operation. The asset utilization (AU) model demonstrates that efficient equipment utilization can be monitored quantitatively and improvement opportunities identified so that the greatest benefit to the operation can be obtained. The process optimization framework, comprised of three parallel activities and a designed experiment, establishes the process-product relationship. The overall strategy of predictive model development provided from the parallel activities comprising the optimization framework is to synthesize a model based on existing data, both qualitative and quantitative, using canonical discriminant analysis, to identify main effect variables affecting the principal efficiency constraints identified using AU, operator knowledge and order-of-magni-tude calculations are then employed to refine this model using designed experiments, where appropriate, to facilitate the development of a quantitative, proactive optimization strategy for eliminating the constraints. Most importantly, this overall strategy plays a significant role in demonstrating, and facilitating employee acceptance, that the manufacturing operation has evolved from an experienced-based process to one based on quantifiable science.     

PROCESS IMPROVEMENT: AN EXPLORATORY DATA ANALYSIS APPROACH WITHIN AN INTERVAL-BASED OPTIMIZATION FRAMEWORK

This article revisits an old problem; “systematically explore the information contained in a set of operating data records and find from it how to improve operational performance by taking the appropriate decisions in the space of operating conditions,” thus leading to continuous process improvement. A series of industrial case studies within the framework of the internships in the Leaders for Manufacturing (LFM) program at Massachusetts Institute of Technology led us to a reexamination of the traditional formulations for the above problem. The resulting methodology is characterized by the following features: (1) problem statement and solutions are expressed in terms of hyperrectangles in the decision space, replacing conventional pointwise results; (2) data-driven, nonparametric learning methodologies were advanced to produce the requisite mapping between performance and decisions; (3) operating performance is in essence multifaceted, leading to a multiobjective problem, which is treated as such. The proposed methodology has been applied to a number of industrial examples and in this paper we provide a brief overview only of those that can be discussed in the open literature.     

A MODEL FOR SWITCHING DISPATCHING RULES IN REAL TIME IN A FLEXIBLE MANUFACTURING CELL

We model choice of dispatching rules in real time (system state dependent) as a pattern recognition problem, using a modified version of Data Envelopment Analysis. A data base of system state and performance values is created from extensive simulation, and this data base is used to train the pattern-recognition model. Our results show that the model is very effective in choosing a mix of dispatching rules over a period of time, varying the mix with system objectives, and performing better than the strategy of using fixed rules. We show how “If-Then” decision rules can be created from the model and portrayed in a decision-tree-like diagram. Since such decision rules are based on rigorous mathematical foundations, optimization will be ensured in our approach.     

A NEW LEARNING APPROACH TO PROCESS IMPROVEMENT IN A TELECOMMUNICATIONS COMPANY

Redesigning and improving business processes to better serve customer needs has become a priority in service industries as they scramble to become more competitive. We describe an approach to process improvement that is being developed collaboratively by applied researchers at US WEST, a major telecommunications company, and the University of Colorado. Motivated by the need to streamline and to add more quantitative power to traditional quality improvement processes, the new approach uses an artificial intelligence (AI) statistical tree growing method using customer survey data to identify operations areas where improvements are expected to affect customers most. This AI/statistical method also identifies realistic quantitative targets for improvement and suggests specific strategies predicted to have high impart. This research, funded in part by the Colorado Advanced Software Institute (CASI) to stimulate profitable innovations, has resulted in a practical methodology used successfully at US WEST to help set process improvement priorities and guide resource allocation decisions throughout the company.     

AN OVERVIEW OF TRADEOFF CURVES IN MANUFACTURING SYSTEMS DESIGN

In this paper we review the use of tradeoff curves in the design of manufacturing systems that can be modeled as open queueing networks. We focus particularly on the tradeoff between expected work-in-process (or product leadtime) and capacity investment in job shops. We review the algorithms in the literature to derive tradeoff curves and illustrate their application in evaluating the efficiency of the system, in deciding how much capacity to have, how to allocate resources between the reduction of uncertainty and the introduction of new technologies, and how to assess the impact of changes in products throughput and product mix. The methodology is illustrated with an example derived from an actual application in the semiconductor industry.     

STATE-OF-THE-ART SURVEY: OPEN QUEUEING NETWORKS: OPTIMIZATION AND PERFORMANCE EVALUATION MODELS FOR DISCRETE MANUFACTURING SYSTEMS

In this survey we review methods to analyze open queueing network models for discrete manufacturing systems. We focus on design and planning models for job shops. The survey is divided in two parts: in the first we review exact and approximate decomposition methods for performance evaluation models for single and multiple product class networks. The second part reviews optimization models of three categories of problems: the first minimizes capital investment subject to attaining a performance measure (WIP or lead time), the second seeks to optimize the performance measure subject to resource constraints, and the third explores recent research developments in complexity reduction through shop redesign and products partitioning.     

MULTISTAGE SAFETY STOCK PLANNING WITH ITEM DEMANDS CORRELATED ACROSS PRODUCTS AND THROUGH TIME

Determining safety stocks in multistage manufacturing systems with serial or divergent structures, where end-item demands are allowed to be correlated both between products as well as in time, is my focus. I show that these types of correlation have contrary effects on the distribution of safety stocks over the manufacturing stages and that neglecting the correlation of demand can lead to significant deviation from the optimal buffer policy. Using base-stock control and assuming total reliability for internal supplies, I present a procedure for integrated multilevel safety stock optimization that can be applied to arbitrary serial and divergent systems even when demand is jointly cross-product and cross-time correlated. As I demonstrate in an example for autocorrelated demands of a moving average type, there are specific solution properties that drastically reduce the computational effort for safety stock planning. Safety stocks determined in that way can be used as an appropriate protection against demand uncertainties in material requirements planning systems.