DSES6620 Simulation Modeling and Analysis - Term Project Report: Rapid Prototyping Cell Simulation
DSES6620 Simulation Modeling and Analysis
Professor: Ernesto Gutierrez-Miravete
Term Project Report: Rapid Prototyping Cell Simulation
Hanspeter Bayer
Telephone: 203-492-8051
Email:
21 December 2000
Table of Contents
Table of Contents 2
Abstract 4
Introduction 5
Goals 5
Scope 5
Requirements 5
The System 5
The Model 6
Entities 7
Locations, Routings and Processes 7
Resources 7
Simplifications and Assumptions 7
No rejected work orders 7
Transit times are negligible 7
Model Maker Utilization is the same as Workstation Utilization 7
Number of Parts Requested per Work Order is not Handled Explicitly 7
SLA Machines Can Handle Two Work Orders Simultaneously 8
No Downtime 8
SLA Service Time = In-Process Time – Model Maker Service Time 8
Model Makers Do Not Leave An SLA Job to Work on Another Job 8
Data 8
Table 1: Relevant Model Shop Work Order Fields 8
Input Analysis 9
Performance Analysis 10
Verification, Validation and Resulting Refinements 12
Shifts 12
Gate 12
Disk Drives 12
UV Oven 12
Validation Metrics 12
Table 2: Validation Metrics 13
Variations 13
Addition of a Third Workstation (and Model Maker) 13
Forced Batching 13
Addition of a Third SLA Machine 13
Results 13
Comparison of Systems 14
Conclusions 14
References 14
Appendices 15
Appendix 1: Text Printout of the Original Model 15
Appendix 2: Output Results (Averaged) from the Original Model Simulation 18
Abstract
This paper presents the modeling and simulation of a stereolithography-based rapid prototyping work cell. Model development, assumptions, input data, verification and validation are discussed. Mean time in system is identified as the major performance metric. The results of the simulation are presented, as are the results of three variations that were modeled in an attempt to reduce mean time in system. Conclusions are drawn based on the results.
Introduction
The system being modeled and simulated is a set of two stereolithography apparatuses and the equipment and people associated with them. This system of components can be considered as a rapid prototyping cell in which prototype parts are “grown” out of a plastic resin. The cell is part of a larger model shop in a large, manufacturing company.
Work for the cell is sent to the shop by design engineers via the creation of an electronic work order. These work orders are often referred to as “jobs.” The work order is accompanied by a 3D parametric solid model, which the engineer has created and which can be read directly by the rapid prototyping software. The work order typically requests between one and ten two samples be made of a part, and may include several different parts. There is a perception among the design engineers who feed work orders into this system that the system cycle time is too long.
Each SLA machine is dedicated to producing parts using a certain resin. The first resin, referred to as “5170”, is very stiff and somewhat brittle. The second resin, referred to as “SOMOS” is more pliable. The resin is chosen by the design engineer, depending on the characteristics required by the design.
Goals
The goals from the original project proposal were simplified and reduced to:
· Develop a model of the rapid prototyping cell
· Validate the model; that is show that the model is accurate
· Make several changes to the model and compare their resulting performance to that of the original model to see if any of the changes predict a reduction in time in system.
Scope
The scope of the project is confined to the modeling and simulation of the operation of the two stereolithography apparatuses (SLA’s or SLA machines) and the resources required for their operation.
Requirements
The following were available to execute the project:
· The author’s time
· The student version (4.2) of ProModel
· Excel Spreadsheet software
· A database containing information for a year’s worth of model shop workflow
· Limited time with model makers and the model shop manager
The System
Figure 1 shows a schematic of the system. Work orders are fed into the system and are assigned to model makers. Once the model maker is free to work on a work order, they process the solid model file on a workstation, save their work and then send the file to the appropriate SLA when it is free. The SLA grows the part or parts. Once they have been grown the parts are placed in a UV oven for one hour to finish curing. Once they have cured, they are manually cleaned by the model makers. At this point the work order is complete.
Figure 1: System Schematic
The Model
Figure 2: Model in ProModel
Entities
Two entity types have been created in the model; both are work orders. The first represents a work order requesting SLA parts grown using the 5170 resin (job_5170), the second represents a work order requesting parts grown using the SOMOS resin (job_somos).
Two more entities, batch entities, were required when the variation using forced batching (described later) was developed.
Locations, Routings and Processes
The model has the following locations:
· A gate at which the 5170 jobs arrive – from here they go to the queue
· A gate at which the SOMOS jobs arrive – from here they go to the queue
· A queue (known informally as “the shelf”) – from here the jobs go to a workstation when either is free and a model maker is available
· Two workstations – model maker service time is used to make the job “wait” here – from here the jobs are stored on one of the workstations’ hard drives
· The hard drives for these workstations – modeled as queues which store the jobs while the SLA’s are occupied
· Two SLA Machines – service times with a mean of 20 hours and a standard deviation of 8 hours
· A UV oven (modeled as a conveyor) – it takes one hour for a part to traverse this conveyor
· Two cleaning stations – the jobs require a model maker at this point - they wait here for a half-hour to represent the model maker cleaning the parts and the jobs exit the system.
Resources
Several model makers are available to spend at least a part of their time working on SLA work orders. Typically no more than two are assigned at any given time to work on SLA work orders.
Simplifications and Assumptions
No rejected work orders
The time taken by the manager to assign or reject work orders is negligible. Typically, the manager can process a work order in a matter of minutes. It is assumed that this does not affect the time in system for a given job, nor do work orders queue up at the manager. That is, the manager’s service time is much less than the work order interarrival times.
Transit times are negligible
Since time in system values are of the order of several days and service times are of the order of several hours, it was assumed that the time taken for a work order to move from location to location is negligible.
Model Maker Utilization is the same as Workstation Utilization
In the early versions of the model, no resources were assigned to the workstation location. It was assumed that whenever the workstation was active, one model maker was needed to run it. Therefore, the utilization of the model maker was assumed to be exactly that of the workstation. As the model was refined it and the cleaning process was added, it was necessary to define model makers as actual resources.
Number of Parts Requested per Work Order is not Handled Explicitly
Although in reality, the number of parts fabricated in a given work order can vary from one to ten, no attribute for this was assigned to the work order entities. It was assumed that the effect of the number of parts requested in a given work order is handled implicitly in the variation of processing times. In other words, the variation in processing times is as large as it is due partly to the variation in number of parts requested in the work orders.
SLA Machines Can Handle Two Work Orders Simultaneously
In reality, as many as five work orders may be fit into an SLA machine, if the part sizes and quantities are sufficiently small. Conversely, only one work order may fit it requires larger parts or many parts. This assumption is based on the fact that most of the time the work order requirements are such that two jobs can fit, and that it is rare in practice to see more than two jobs combined in one SLA cycle.
No Downtime
The author can recall only two times during a period of four years that the model shop maker announced that an SLA machine was down (for a two to three day period). Therefore downtime has been assumed to be negligible. At one time, there was only one SLA in the cell and significant downtimes occurred when the machine had to be flushed so that it could be switched to the other resin.
SLA Service Time = In-Process Time – Model Maker Service Time
This assumption came out of a limitation of the data. The data did not explicitly include the actual service time for the SLA’s. The data did however, give starting and ending dates for jobs once they left the queue, and it also gave the hours spent on the job by the model maker.
Model Makers Do Not Leave an SLA Job to Work on Another Job
Essentially, it was assumed that once a job left the queue and went into process, a model maker did not start working on another non-SLA job until the SLA job was completed.
Data
A database of input and performance data has been growing since the implementation of the electronic work form system about two years ago. A subset of the entire database, representing all work orders submitted during the 12 months starting 9/1/1999 and ending 8/31/2000, was extracted from the database.
Each record has the following fields that were used to develop the model:
Table 1: Relevant Model Shop Work Order Fields
Field / Type / Descriptionstart_date / System-generated date/time / date/time electronic work order was created
end_date / System-generated date/time / date/time electronic work order was updated to complete (only populated for completed work orders)
Material / User-generated text / material of which part is to be made (plastics, steels, etc.)
Sla / Y/N attribute / is the part to be grown in an SLA?
Startdate / User-generated text / date model maker started
completion_date / User-generated text / date model maker finished
total_hours / User-generated number / model maker hours
Input Analysis
These figures (3 and 4) show the interarrival time distribution for all SLA work orders (those requesting the 5170 resin and those requesting the SOMOS resin):
Figure 3
Figure 4
Mean interarrival time for the 5170 jobs was 4.85 hours and 14.10 hours for the SOMOS jobs.
Figure 5 shows the distribution of model maker service times. No acceptable analytic distribution was found to model this distribution, so a user-defined distribution was used. Note that this data was subjective in that it was provided by the model maker as an estimate and was not computer generated.
Figure 5
Performance Analysis
The primary performance measure will be the time in system (mean and distribution). It is not apparent that throughput rates, or work in process numbers, or any other measure is of direct concern to model shop management, model makers or R&D engineers (customers). The primary concern of all parties seems to be only how long it takes to get jobs through the system.
These figures (6 and 7) show the current time in system distribution for the 5170 jobs and SOMOS jobs:
Figure 6
Figure 7
The mean time in system for 5170 jobs was 339 hours (14.13 calendar days) and that for the SOMOS jobs was 315 hours (13.15 calendar days). It is interesting to note that the time in system for each type of job is essentially the same, despite the fact that 5170 jobs are requested about four times more than SOMOS jobs. Since each SLA machine has the same capacity and processing speed, it might be expected that the SOMOS jobs would be handled much more quickly.
Verification, Validation and Resulting Refinements
The verification and validation activities were conducted as an integral part of the development of the model. Some of the activities served to both verify and validate, as in the case of the inter-arrival times. Refinements were made to the model as verification and validation processes indicated deficiencies in the model. Among there were:
Shifts
Work orders arrive only during the first shift and the workstations are used to process the jobs during the first shift. However, once the SLA machine is started on a job it can be (and is) left unattended for overnight and weekend operation. The model was therefore created to run through 24-hour days and shifts were added to those locations and resources that were available only during shifts. As an example, without employing shifts, jobs arrived 24 hours a day, even though in reality, the engineers were only submitting jobs during their workday. This resulted in approximately three times too many jobs being submitted, even though the interarrival time distribution was correct.