SIMULATION APPLIED TO ENGINE UPPER INTAKE MANIFOLD = ASSEMBLY

SIMULATION APPLIED TO ENGINE UPPER INTAKE MANIFOLD ASSEMBLY

Edward J. Williams 206-2 Engineering Computer Center Mail Drop 3 Ford Motor Company Dearborn, Michigan 48121-2053, U.S.A.

Dean E. Orlando Industrial Engineering Department Essex Engine Plant One Quality Way Windsor, Ontario N9A 6X3, CANADA

KEYWORDS

discrete process simulation, manufacturing applications, palletization

ABSTRACT

We describe the application of simulation and statistical analysis to the problems of determining the optimum operating pattern, best number of pallets to use in a recirculating spur line, and the ideal broadcast location within the context of upper intake manifold assembly operations pertinent to the manufacture of automobile engines. Results obtained from these analyses enabled production managers to maximize net jobs per hour (JPH) cost-effectively while minimizing time-in-system and balancing in-process queue lengths.

1 INTRODUCTION

Simulation has been defined as "the process of designing a mathematical or logical model of a real system and then conducting computer-based experiments with the model to describe, explain, and predict the behavior of the real system" (Hoover and Perry 1989). Manufacturing operations are one of the earliest and most perennially popular simulation applications because the complexity of many manufacturing systems defies improvement by "simply thinking and talking about possible approaches," (Clark 1996), and because simulation, unlike traditional closed-form analytical techniques, can assess the effects of interactions among system components, complicated by stochastic variations, on overall system performance (Gogg and Mott 1995; Martinich 1997).

In the study described here, production managers wished to increase production per unit of time cost-effectively by appropriate specifications within the upper intake manifold assembly operations within the overall engine manufacturing process. Two key questions, whose answers were correctly suspected to be highly interrelated even before formal analytical study began, were:

How many pallets should be used for upper-intake-manifold transport along the recirculating spur line within assembly operations?
From which operating station on the main line should a broadcast signal be sent to the recirculating spur line?

We first present an overview of the pertinent engine-assembly operations. Next, we indicate the project goals and performance metrics. Then, we present details of data collection and model construction, verification, and validation. We then describe results of experimentation and analysis undertaken by using the model, and indicate anticipated directions for future work. The steps within this practical application of simulation closely follow those in (Ülgen et al. 1994a), (Ülgen et al. 1994b), and (Banks and Gibson 1996). An example of recent successful use of simulation to analyze overall engine production appears in (Jayaraman and Agarwal 1996).

2 OVERVIEW OF ENGINE UPPER INTAKE ASSEMBLY OPERATIONS

An automobile engine being manufactured travels from operation to operation along a main assembly line. Within this overall manufacturing process, the engine must have its upper intake manifold, a subassembly, attached. Accordingly, a subsidiary upper-intake-manifold assembly process supplies this component to engines traveling along the main assembly line. Since the proper manifold to be received by an engine depends on the parameters of the engine itself, an engine approaching the subsidiary line must "request" a suitable manifold for itself by signaling its requirements from a suitably chosen "broadcast point," creating "just-in-time" modeling challenges like those faced in automobile seat manufacturing processes as modeled by (Venkateswar 1996), since the manifold must be corrected matched with its engine, just as a seat must be correctly matched (color, fabric, etc.) with its automobile.

Within the upper intake manifold pallet loop, the target of detailed analysis in this study, operations occur in the following sequence:

at location A, empty pallets await a signal sent by an approaching engine from the operating station chosen as the main-line broadcast point
having received this signal, the empty pallet, now "knowing" the correct type of upper intake manifold to be placed upon itself, proceeds to location B₁, location B₂, or location B₃ -- whichever location will supply the needed upper intake manifold
the pallet and upper intake manifold then proceed to location C for addition of the appropriate throttle body assembly
the pallet, upper intake manifold, and throttle body assembly then travel to location D, where an operator manually unloads the collective assembly from the pallet and attaches it to the engine which originally sent the signal requesting it.

The flow of components among these operations is diagrammed in Figure 1. Managers and engineers were eager to analyze this system to ensure achievement of palletization advantages such as increased agility, shortened cycle times, and heightened ability to accommodate pending increases in demand (Owen 1996).

Hence the system under study here, the upper intake manifold pallet loop, is the server and the main assembly line is its customer. Additionally, the main assembly line is a "pull" system, whereas the pallet loop is a "push" system. Detailed analysis of one system component (the pallet loop) determines how it can best support the overall system. The challenge of analyzing a system by decomposition into components is typical of the "industrial-strength" simulation challenges described by (Seppanen 1995). The interface between the "pull" system being supported by a subsidiary "push" system is a variant of a similar prototype interface described and analyzed in (Cochran and Kim 1995).

3 PROJECT GOALS AND PERFORMANCE METRICS

The objective of this simulation study was the determination of the optimal broadcast point location (station) and the concurrently optimal quantity of pallets to use for operation of the upper intake manifold pallet loop at the engine plant. The lower and upper station limits for the broadcast point location were chosen from operational considerations. Specifically, the lower station limit of #148 was chosen heuristically, and the upper station limit of #168 was chosen because, working back (upstream) from station #187, station #168 is the first station from which a broadcast signal can be sent with the associated upper intake pallet reaching the pallet unloading point in time to be unloaded. Stations #154 and #162 were chosen as intermediate possibilities approximately equally spaced between stations #148 and #168. The lower and upper plausible limits for the number of pallets (15 and 36 respectively) were determined from heuristic and historical considerations. Intermediate pallet quantities of 22 and 29 were chosen to be equally spaced among four palletization levels. These decisions produced a total of sixteen alternatives, four broadcast-station locations and four levels of palletization.

The performance of the overall system was evaluated by four metrics established jointly by plant managerial and engineering personnel. The first and most fundamentally important metric was system throughput in jobs per hour (JPH), which had to meet or exceed production quotas. The second metric was average time in system (makespan). Reduction of makespan aims to increase utilizations of costly equipment and resources (Harrell and Tumay 1995), a criterion of particular importance in this context due to the high desirability of moving engines along the main line as quickly as possible. The third metric was queue length (both average and maximum) for each of the three different queues of pallets defined within the upper intake pallet loop:

those pallets which have been unloaded and are en route to the loading point, or awaiting loading there
those pallets which have been loaded, but not yet processed at the throttle body operation
those pallets which have been loaded and processed at the throttle body operation, and are en route to the unloading point or awaiting unloading there.

The fourth metric was "queue disparity," the difference between the average length of the longest of these three queues and the average length of the shortest of these three queues. For example, if the three queues listed above had average lengths of 10.1, 4.3, and 5.6 pallets respectively, the queue disparity would be 10.1-4.3 = 5.8 pallets. Naturally, both managerial and operational personnel were eager to make the third and fourth metrics as small as reasonably possible consistent with strong performance on the first two metrics.

4 COLLECTION OF DATA

Well before specific data collection began, a team of Essex Engine Plant industrial, process, and controls engineers, simulation consultants (both internal to Ford and external), and engineers from the machine-tool vendor met to specify precisely the scope of the project, as described in section 3.

Next, initial data collection confirmed the following fundamental constraints:

the geometric relationship of the upper intake manifold pallet loop and the main assembly must be that shown in Figure 1
the number of pallets must not exceed 36 due to size restrictions
availability of a throttle body assembly would never be a constraint
the broadcast point must be located between station #148 and station #168 inclusive, as described in section 3.

Details of operation times required at locations A, B₁, B₂, B₃, C, and D (see Figure 1 and section 3), required for this model, were available from archives of previous, more localized simulation studies of individual areas and/or analogous operations elsewhere in this plant or other plants. This a priori availability of required data from archives, which greatly accelerated the progress of this project to meet tight timing constraints, illustrated the importance of treating data as a valuable corporate resource, as advocated in (Ülgen et al. 1995). These operational data comprised cycle times and two recurring dunnage timing overheads. Dunnage, the protective material guarding the parts against unwanted movement or shock (K. W. Tunnell Company, Incorporated 1995), required removal from each layer of parts on the pallet (four seconds). Similarly, the manual replacement of each emptied pallet with the next full one required twelve seconds. Since maximum queue lengths were a metric of importance, these details could not validly be modeled as an average cycle time. These data are summarized in Table 1 below. Table 1: Cycle and Dunnage Timing Data

Location (Figure 1)	Cycle Time (seconds)	4-sec Layer Change @	12-sec Pallet Change @
A	n/a	n/a	n/a
B₁	18	9th part	63rd part
B₂	18	6th part	24th part
B₃	18	6th part	24th part
C	18	40th part	280th part
D	20	n/a	n/a

5 CONSTRUCTION, VERIFICATION, AND VALIDATION OF MODEL

The model was constructed using the WITNESS simulation software package. This package provides an interactive model-building environment which allows a user knowledgeable about the process being analyzed to build and verify a complex model rapidly and conveniently. Further, use of WITNESS permits the modeler to develop a screen animation of the process concurrently with building the logical model (Thompson 1996).

Several techniques were used to verify and validate the model. Walkthroughs of the model conducted by the modeling team members exposed errors for early correction. Examination of the animation and step-by-step model traces, both within the modeling team and subsequently with the plant engineers, confirmed the model correctly mirrored the process logic of the actual system. Preliminary runs of the model with all stochastic variation removed allowed numerical validation of output statistics (Schriber 1974).

The manufacturing system is inherently steady-state, not terminating. Therefore, the question of initial-transient length required resolution. Graphical examination as suggested in (Welch 1983) of system-state parameters as functions of time confirmed that a warmup time (no data collected) of one hour (followed by a data-collection interval of thirty-nine hours) for each replication was sufficient to remove all initialization bias.

Although the manual operations on the upper intake manifold pallet loop had no downtime, the operations on the main assembly line did. Since detailed performance statistics on queue lengths within the pallet loop were needed, these downtimes required inclusion in the model at a high level of detail. Accordingly, the modeling team attempted to fit various canonical densities, such as the gamma, Weibull, or lognormal, to the extensive empirical downtime data available. Various goodness-of-fit tests, such as the chi-square, Kolmogorov-Smirnov, and Anderson-Darling (Law and Kelton 1991) were unanimous in rejecting all such closed-form density functions. Therefore, the empirical data were used directly for random generation of availability-time-based MTTR and MTBF intervals. Since operations personnel at the plant confirmed that the extensive history of actual MTTR and MTBF data encompassed the full range of conceivable values from minimum to maximum, the usual objection to using empirical data directly (Williams 1994) was overridden.

6 ANALYSIS OF RESULTS

As indicated in section 3, there were sixteen possible alternatives to consider, resulting from the orthogonal combination of four possible broadcast-point locations and four possible pallet quantities. Five replications of each alternative (a total of 80 simulation runs completed) were used to construct 95% confidence levels for each performance metric. All confidence levels were narrow enough to foster high confidence in decision-making based on the model; for example, the widest of the sixteen time-in-system confidence intervals was less than 1½ minutes wide.

The analysis of results began by noticing that the combination of placing the broadcast point at Station #168 and having 36 pallets in the recirculating loop was completely unable to keep up with production. This alternative was hence rejected forthwith ("X" in Table 2, below). However, the throughput metric alone distinguished no further among the remaining fifteen alternatives under consideration.

Five other alternatives performed poorly by causing a long time-in-system. Specifically, having 36 pallets in the loop led to this problem irrespective of whether the broadcast point was placed at station #148, #154, or #162. Likewise, using 29 pallets and placing the broadcast point at station #162 or #168 produced this problem ("L" in Table 2).

Seven other alternatives produced higher "queue disparities" than desired. As explained in section 3, one of the system performance metrics sought approximate equality of average length among the three queues in the upper intake pallet loop. Specifically, having 15, 22, or 29 pallets in the loop and placing the broadcast point at station #148 or #154, or having 22 pallets and placing the broadcast point at station #168 ("Q" in Table 2) led to high queue disparaties.

Table 2: Elimination of Combinations

Broadcast	15	22	29	36
Station	Pallets	Pallets	Pallets	Pallets
148	Q	Q	Q	L
154	Q	Q	Q	L
162	OK	OK	L	L
168	OK	Q	L	X

7 CONCLUSIONS AND INDICATIONS FOR FURTHER WORK

In view of Table 2 above, the three most promising alternatives (denoted "OK" in the table) were:

Locate the broadcast point at station #162 and use 15 pallets
Locate the broadcast point at station #168 and use 15 pallets
Locate the broadcast point at station #162 and use 22 pallets

The first two alternatives seemed intuitively more appealing than the third, because they used fewer pallets which had to be purchased, maintained, and stored. Nevertheless, alternative 3 was chosen by plant engineers jointly with design engineers from the machine-tool vendor. This alternative performed slightly better, when simulated, than alternatives 1 and 2. Additionally, alternative 3 contained more "built-in" protection against potential future demands for increased throughput, and also held the advantage that one-time purchase and installation of pallets was more cost-effective than their incremental purchase and installation. After implementation, the actual performance of alternative 3 matched simulation predictions in all four metrics to within 5%.

However, this simulation study examined no processing-time variations on the pallet loop. As a result of further modeling and analysis, such variations may be proved sufficient to require the higher palletization level chosen. In that case, the superiority of alternative (3) above may become more marked.

An additional, unanticipated benefit of this simulation study was its assistance to controls-design engineers in determining location of control switches in the spur (subsidiary) line. These improvements in control-switch placement, achieved early in the implementation process, greatly reduced installation debugging time and expense.

Since the assembly line analyzed in this study performs multistage production, this simulation work, now used to incorporate day-to-day operational improvements, may be extended to real-time production control using concepts and methods such as those explained and advocated in (Reinhart and Heitmann 1995), and illustrated in an application to oil production in (Ogård and Eikaas 1996).

ACKNOWLEDGMENTS

Eric R. Haan, Applications Engineer, Production Modeling Corporation, made valuable contributions to the content, clarity, and organization of this paper.

A previous version of this paper appeared in Proceedings of the Southeastern Simulation Conference '96, editor Joseph S. Gauthier, pages 16-22.

APPENDIX: TRADEMARK

WITNESS is a trademark of the Lanner Group.

REFERENCES

Banks, Jerry, and Randall R. Gibson. 1996. Getting Started in Simulation Modeling. IIE Solutions 28(11):34-39.

Clark, Gordon M. 1996. Introduction to Manufacturing Applications. In Proceedings of the 1996 Winter Simulation Conference, eds John M. Charnes, Douglas J. Morrice, Daniel T. Brunner, and James J. Swain, 85-92.

Cochran, Jeffery K., and Sung-Soo Kim. 1995. A Simulation Approach To Determine an Optimum Junction Point in Hybrid Push and Pull Manufacturing Systems. In Simulators International XII, eds. Maurice Ades and Ariel Sharon, 271-276.

Gogg, Thomas J., and Jack Robert A. Mott. 1995. Improve Quality & Productivity with Simulation. 2nd edition. Palos Verdes, California: JMI Consulting Group.

Harrell, Charles R., and Kerim Tumay. 1995. Simulation Made Easy a Manager=92s Guide. Norcross, Georgia: Industrial Engineering and Management Press.

Hoover, Stewart V., and Ronald F. Perry. 1989. Simulation a Problem-Solving Approach. Reading, Massachusetts: Addison-Wesley Publishing Company.

Jayaraman, Arun, and Arun Agarwal. 1996. Simulating an Engine Plant. Manufacturing Engineering 117(5):60-68.

K. W. Tunnell Company, Incorporated. 1995. Containerization. In Standard Handbook of Plant Engineering, 2nd edition, ed. Robert C. Rosaler, 8-11=978-28. New York, New York: McGraw-Hill, Incorporated.

Law, Averill M., and W. David Kelton. 1991. Simulation Modeling and Analysis, 2nd edition. New York, New York: McGraw-Hill, Incorporated.

Martinich, Joseph S. 1997. Production and Operations Management : an Applied Modern Approach. New York, New York: John Wiley & Sons, Incorporated.

Ogård, Ommund, and Tor Ivar Eikaas. 1996. Integrated Flexible Process Simulation. In Proceedings of the 8th European Simulation Symposium, Volume 1, eds. Agostino G. Bruzzone and Eugène J. H. Kerckhoffs, 365-369.

Owen, Jean V. 1996. Profits Through Pallets. Manufacturing Engineering 116(1):33-42.

Reinhart, Gunther, and Knut Heitmann. 1995. Simulation-Based Production Control Within Complex Multistage Production Plants. In Proceedings of the 7th European Simulation Symposium, eds. Mario DalCin, Ulrich Herzog, Gunter Bolch, and Ali Riza Kaylan, 634-638.

Schriber, Thomas J. 1974. Simulation Using GPSS. New York, New York: John Wiley & Sons, Incorporated.

Seppanen, Marvin S. 1995. Developing Industrial Strength Simulation Models. In Proceedings of the 1995 Winter Simulation Conference, eds. Christos Alexopoulos, Keebom Kang, William R. Lilegdon, and David Goldsman, 936-939.

Thompson, W. Bruce. 1996. Introduction to the WITNESS Visual Interactive Simulator and OLEII Automation. In Proceedings of the 1996 Winter Simulation Conference, eds. John M. Charnes, Douglas J. Morrice, Daniel T. Brunner, and James J. Swain, 547-550.

Ülgen, Onur M., John J. Black, Betty Johnsonbaugh, and Roger Klungle. 1994a. Simulation Methodology in Practice -- Part I: Planning for the Study. International Journal of Industrial Engineering, 1(2):119-128.

Ülgen, Onur M., John J. Black, Betty Johnsonbaugh, and Roger Klungle. 1994b. Simulation Methodology in Practice -- Part II: Selling the Results. International Journal of Industrial Engineering, 1(2):129-136.

Ülgen, Onur M., Defne Berkin, Mark Brazier, Roger Klungle, Anil S. Menawat, and Hwa-Sung Na. 1995. Management of Simulation Technology in Large Companies a Panel Discussion. In Proceedings of the 1995 Summer Computer Simulation Conference, eds. Tuncer I. Ören and Louis G. Birta, 361-366.

Venkateswar, Kalluru. 1996. Design Framework for Just-in-Time and In-Sequence Automobile Seat Manufacturing Systems. In Proceedings of the 1996 Summer Computer Simulation Conference, eds. V. Wayne Ingalls, Joseph Cynamon, and Annie V. Saylor, 217-221.

Welch, P. D. 1983. The Statistical Analysis of Simulation Results. In The Computer Performance Modeling Handbook, ed. S. S. Lavenberg, 268-328. New York, New York: Academic Press.

Williams, Edward J. 1994. Downtime Data -- its Collection, Analysis, and Importance. In Proceedings of the 1994 Winter Simulation Conference, eds. Jeffrey D. Tew, Mani S. Manivannan, Deborah A. Sadowski, and Andrew F. Seila, 1040-1043.

AUTHOR BIOGRAPHIES

EDWARD J. WILLIAMS holds bachelor's and master's degrees in mathematics (Michigan State University, 1967; University of Wisconsin, 1968). From 1969 to 1971, he did statistical programming and analysis of biomedical data at Walter Reed Army Hospital, Washington, D.C. He joined Ford in 1972, where he works as a computer software analyst supporting statistical and simulation software. Since 1980, he has taught evening classes at the University of Michigan, including undergraduate and graduate statistics classes and undergraduate and graduate simulation classes using GPSS/H, SLAM II, or SIMAN. He is a member of the Association for Computing Machinery [ACM] and its Special Interest Group in Simulation [SIGSIM], the Institute of Electrical and Electronics Engineers [IEEE], the Institute of Industrial Engineers [IIE], the Society for Computer Simulation [SCS], the Society of Manufacturing Engineers [SME], and the American Statistical Association [ASA]. He serves on the editorial board of the International Journal of Industrial Engineering -- Applications and Practice.

DEAN E. ORLANDO holds a bachelor's degree in industrial engineering from General Motors Institute (1984). During the past fifteen years, he has held various manufacturing positions in the automotive industry. He is now an industrial engineer at Ford Motor Company currently working as a simulation engineer in various cross-functional teams. These teams have used, and are using, simulation both as a tool for facilities-and-tooling investment strategy and, concurrently, for forward planning of manufacturing. He is a member of the Society of Manufacturing Engineers [SME].