Application of lean manufacturing principles to improve a conceptual 238Pu supply process

Journal of Manufacturing Systems 46 (2018) 1–12

Contents lists available at ScienceDirect

Journal of Manufacturing Systems journal homepage: www.elsevier.com/locate/jmansys

Full Length Article

Application of lean manufacturing principles to improve a conceptual 238 Pu supply process夽 Tomcy Thomas a , Steven R. Sherman b,∗ , Rapinder S. Sawhney a a

Department of Industrial and Systems Engineering, University of Tennessee, Knoxville, TN, United States Radiochemical Science and Engineering Group, Nuclear Security and Isotope Technology Division, Oak Ridge National Laboratory, Oak Ridge, TN, United States b

a r t i c l e

i n f o

Article history: Received 18 April 2017 Received in revised form 9 August 2017 Accepted 29 October 2017 Keywords: Plutonium Neptunium RTG Reverse engineering Lean manufacturing Little’s law

a b s t r a c t The mission of the United States (U.S.) Department of Energy’s Pu-238 Supply Project is to rebuild capability to produce 238 Pu at the kilogram scale in the U.S. This radioisotope is used by the National Aeronautics and Space Administration (NASA) to power deep space probes, and the supply is dwindling. It was last produced in the U.S. in 1988. A conceptual design of a 238 Pu supply process is described that uses existing processes and facilities at Oak Ridge National Laboratory’s Radiochemical Engineering Development Center. The rate-limiting section of the conceptual process was analyzed using discrete-event system simulation to determine expected production rates, bottlenecks, and the effects of time delays on the production rate. Process alternatives were generated based on Lean Manufacturing principles, and those were examined and compared to the original process using simulation to identify better operating strategies. © 2017 Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers.

1. Introduction The Pu-238 Supply Project was initiated in 2011 at Oak Ridge National Laboratory’s (ORNL) Radiochemical Engineering Development Center (REDC) in Oak Ridge, Tennessee. The purpose of

Abbreviations: Am, americium; ATR, advanced test reactor; Bk, berkelium; Cf, californium; Cm, curium; CONWIP, constant work in process; CT, cycle time; DOE, US Department of Energy; Es, einsteinium; Fm, fermium; HFIR, high flux isotope reactor; HNO3 , nitric acid; HS-PuO2 , heat source plutonium oxide; INL, Idaho National Laboratory; LANL, Los Alamos National Laboratory; NaOH, sodium hydroxide; NASA, National Aeronautics and Space Administration; Np, neptunium; 237 Np, neptunium237; 237 NpO2 , neptunium-237 oxide; 238 Np, neptunium-238; ORNL, Oak Ridge National Laboratory; 233 Pa, protactinium-233; pH, negative log10 of hydrogen ion concentration; Pu, plutonium; 238 Pu, plutonium-238; Pu-238, plutonium-238; REDC, Radiochemical Engineering Development Center; RPS, radioisotope power source; SRS, Savannah River Site; TH, throughput; VSM, value stream map; WIP, work in process. 夽 This manuscript has been authored by UT-Battelle, LLC and the University of Tennessee under Contract DE-AC05-OR222725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally funded research in accordance with the DOE Public Access Plan (http://energy.gov/ downloads/doe-public-access-plan). ∗ Corresponding author. E-mail address: [email protected] (S.R. Sherman).

the project is to reestablish capabilities to produce plutonium-238 oxide (238 PuO2 ) in quantities useful for fabrication of plutoniumpowered radioisotope power systems (RPS) [1] by the early 2020 s at ORNL. RPSs are used by the National Aeronautics and Space Administration (NASA) to power deep space probes and planetary rovers. 238 Pu is produced by irradiating neptunium-237 (237 Np) with neutrons in a nuclear reactor. 238 Pu production began in the U.S. in the early 1960s and continued until 1988 at the Savannah River Site (SRS) [2,3] near Aiken, South Carolina. After 238 Pu production ceased at SRS and facilities were decommissioned at that location, 238 Pu was purchased from Russia from 1992 until 2009. In 2009, Russia halted further sale of 238 Pu to the U.S. and attempted to negotiate a better contract. A new contract was never achieved, and the U.S. has not been able to purchase 238 Pu from Russia since that time. At present, there is no other 238 Pu supplier and the U.S. has only enough 238 Pu in storage [4] to power a handful of future space missions. 238 Pu production must be restarted if NASA is to continue deep space exploration using plutonium-powered RPSs. The U.S. Department of Energy (DOE) and NASA have established expectations for a modern 238 Pu supply process [5]: • The process must use existing infrastructure and facilities at ORNL rather than build new nuclear facilities to reduce setup cost, but equipment and support services may be modified, as needed. • The process must have capability to produce 1500 g heat source PuO2 (HS-PuO2 )/year on average. HS-PuO2 is defined as pluto-

https://doi.org/10.1016/j.jmsy.2017.10.007 0278-6125/© 2017 Published by Elsevier Ltd on behalf of The Society of Manufacturing Engineers.

2

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

“Fresh” 237NpO2 Remove 233Pa Upstream Make pellets Make targets

Fig. 1, Remove 233 Pa). 237 Np recycled from the processing of irradiated targets is also treated by this method. The 237 NpO2 powder is mixed with aluminum powder and compressed into pellets (in Fig. 1, Make pellets). The pellets are loaded into aluminum tubes and fabricated into targets using target designs [12] approved for use in the High Flux Isotope Reactor (HFIR) at ORNL or the Advanced Test Reactor (ATR) at INL (in Fig. 1, Make targets). In the nuclear reactor, 237 Np reacts with neutrons to form 238 Np, which decays to 238 Pu by the emission of a beta particle [7] (see Eq. (1); in Fig. 1, Irradiate targets).

Irradiate targets

ˇ 237 93 Np (n, )

→ 238 94 Pu

(1)

2.1d

Declad targets (NaOH) Dissolve targets (HNO3)

Downstream

Condion the soluon Np/Pu Separaon by solvent extracon Np Purificaon (ion exchange)

→ 238 93 Np ↑

Pu Purificaon (ion exchange) Oxalate Precipitaon Calcine the precipitate Package and Ship

Fig. 1. Block diagram of baseline 238 Pu supply process.

nium oxide containing sufficient 238 Pu isotopic content to meet NASA RPS specifications [6]. • The product must be a drop-in replacement for HS-PuO2 feed material used currently in Los Alamos National Laboratory’s (LANL) RPS pellet-making process. The Pu-238 Supply Project identified a conceptual process based on the original SRS process [7] and more recent modifications [8] suggested by Idaho National Laboratory (INL) in Idaho Falls, Idaho, and is working to demonstrate, scale up, and optimize that process. In the process, 237 Np feedstock at INL is transported to ORNL where it is fashioned into 237 Np pellets and targets, and then the targets are irradiated in a nuclear reactor to form 238 Pu. The 238 Pu in the targets is separated, purified, converted into powder, packaged, and shipped to LANL to be made into 238 Pu oxide pellets for RPSs. Residual 237 Np is recycled to be made again into targets. ORNL proposed the process block flow diagram shown in Fig. 1. The process shown is the preferred alternative among many process options considered [9,10]. At INL, 237 Np oxide (237 NpO2 ) is removed from storage, re-packaged, and shipped to ORNL (in Fig. 1, “Fresh” 237 NpO2 ). After arrival, the 237 NpO2 is processed to remove the protactinium decay daughter (233 Pa) by dissolving the material in nitric acid (HNO3 ), purifying it by ion exchange, and heating it in a high-temperature furnace to remake the 237 NpO2 as a powder free of 233 Pa, a method called Modified Direct Denitration [11] (in

After irradiation, the targets are stored for several months at the reactor sites to allow short-lived fission products to decay. When sufficiently decayed, the irradiated targets are transported from the reactor facilities to the chemical processing facilities [13,14] (i.e., radiation shielded hot cells) at REDC where they are chemically processed to make a Pu product stream, a recovered Np stream, and waste streams containing unwanted fission products and residual actinides. In the hot cells, the aluminum target bodies and the aluminum powder in the irradiated pellets are dissolved (in Fig. 1, Declad targets (NaOH)) using a combination of sodium hydroxide (NaOH) and sodium nitrate. The remaining oxide material is dissolved in a subsequent step using HNO3 [15] (in Fig. 1, Dissolve targets (HNO3 )). Next, the acidic solution is concentrated by evaporation, and chemical additives are applied to adjust the oxidation states of Np and Pu, and to adjust the solution pH (in Fig. 1, Condition the solution). Solvent extraction is used to perform the Np/Pu separation [8] (in Fig. 1, Np/Pu Separation by solvent extraction). The Np and Pu streams are further purified using ion exchange [16] (in Fig. 1, Np purification (ion exchange), Pu purification (ion exchange)). The purified Np stream is recycled, and the Pu stream is converted into an oxide using oxalate precipitation (in Fig. 1, Oxalate precipitation) and calcining methods [17] (in Fig. 1, Calcine the precipitate). The final product, HS-PuO2 , is then packaged and shipped to LANL (in Fig. 1, Package and ship). Ideally, the Pu-238 Supply Project would have defined the production requirements, and then designed the equipment and facilities to meet the requirements. Upstream of the nuclear reactor, this is the case. The automated equipment needed to make pellets and targets were not available, and new equipment is being designed and fabricated to meet the process requirements. However, the nuclear reactors (ATR and HFIR) and the downstream processing equipment are already in place, and the existing equipment and facilities must be adapted to meet the process requirements. This is especially true for the downstream equipment, which was designed and built for production and separation of isotopes other than 238 Pu (i.e., Am, Cm, Bk, Cf, Es, and Fm [14,15]); it is not specifically sized or optimized for 238 Pu production. Simple schedule analysis indicates the desired production rate can be achieved under nominal conditions with this equipment, but the outcome is uncertain when nuclear safety constraints and potential process variabilities are recognized. From informal study of equipment capacities and throughput rates, it is suspected the rate-limiting processing step(s) lies somewhere within the downstream processing section (see Fig. 1, downstream from “Irradiate targets”). Collectively, the Pu-238 Supply Project calls those downstream processing steps the chemical processing section. The automated pellet and target fabrication equipment are new, and are being designed and fabricated to generate an excess number of targets per year. The nuclear reactors (i.e., ATR and HFIR) are large user facilities with excess capacity for multiple simultaneous irradiation activities. The downstream

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

equipment, however, is not adapted and optimized for 238 Pu production. The facility does not have excess capacity, and it is not obvious to the engineers working to develop the 238 Pu supply process that the downstream section can meet the desired supply rate. Some of the equipment can be modified, but basic functions must be preserved because the existing equipment is still needed for other ongoing radioisotope production missions when the 238 Pu supply process activities are not underway. The chemical processing facility is expected to be available 300 days per calendar year for 238 Pu production operations. Assuming the rate-limiting step(s) lies within the chemical processing section, discrete-event simulation and analysis methods were used to determine processing rates, identify process bottlenecks, and test potential process improvements in that process section. The specific objectives of this work are: (1) Express the chemical processing section in the form of a value stream map [18] (VSM). (2) Incorporate the VSM into a discrete-event simulation model. (3) Run simulations of the chemical processing section to predict throughput rates, identify process bottlenecks, and examine how increasing processing times for individual steps affects overall production rate. (4) Propose process alternatives generated by using concepts from the Theory of Constraints [19], Lean Manufacturing [20], and Reverse Engineering [21]. (5) Compare process alternatives to the baseline conceptual process to identify potential process improvements. This work seeks to determine whether the baseline conceptual process can meet the production goal, and to identify better ways of performing the process. Although it is possible the global constraint may lie within the upstream or irradiation processes, it is not likely given the excess capacity of those systems. Modeling of the downstream section by itself allows for search of constraints in the process section most likely to contain them without waiting for the upstream processes to become more defined. If the downstream section can meet the production goal, then the overall process may also meet the goal if the upstream and irradiation sections perform as expected. If the downstream section cannot meet the goal, then no improvements to the upstream sections will alter or change the outcome, and efforts can be focused immediately on improving the process bottlenecks in the downstream section. Unlike the nuclear reactors, the upstream and downstream sections will only be used to produce 238 Pu when the process is active (i.e., exclusive use). When the 238 Pu supply process is not active, the upstream equipment is idled, and the equipment in the downstream section is made available (i.e., flushed and cleaned) for other missions. 2. Baseline process simulation and analysis 2.1. Value stream map (VSM) A VSM is a tool used in Lean Manufacturing to depict the flow of inventory and information in a manufacturing process. A VSM of the chemical processing section was constructed from: • The block flow diagram in Fig. 1 • Processing step duration data collected from process log books for research operations • Equipment capacity limits, and • Subject matter expert opinion (see Fig. 2). Initial data concerning process step durations was collected by speaking to process technicians and system engineers about how long it typically takes to set up, perform, and do post-run cleanup work for each of the downstream steps in Fig. 1. Then, the expert data were compared to process log books to determine how

3

well the subject matter expert estimates compared to the process histories in the log books. Corrections were made to the process durations when there were wide discrepancies and the process experts agreed their estimates needed correction. Batch sizes were determined using inputs from system engineers and were verified by checking the system engineers’ information against the facility manager’s equipment batch size estimates. No data were available on a full production run because scale-up work was still in progress and no full production runs have yet been performed. The VSM in Fig. 2 contains much of the process data needed to simulate the logistical operations of the chemical processing section. The VSM simplifies the process into four blocks – dissolve targets, prepare the dissolved solution for solvent extraction (conditioning), solvent extraction and Np/Pu purification, and HS-PuO2 powder production and packaging. The analysis in this paper studies the process only up to the packaging of 238 Pu into cans, and shipping methods are not included in the VSM. On the left, the VSM shows the movement of targets from irradiated target storage to the target dissolver. From mass balance information, it is necessary to process 432 irradiated targets to produce 1500 g HS-PuO2 /year [see Eq. (2)]). This number of targets compensates for small residual losses of 237 Np and 238 Pu to waste (i.e., 0.1% of the oxide material is assumed to be lost to the waste stream). In Eq. (2), conversion is the fraction of 237 Np transmuted to 238 Pu when a target is irradiated in the nuclear reactor and it also contains a correction for HS-PuO2 lost to the waste stream. #targets=

desired mass HS-PuO2 #pellets target

×

pellet mass pellet

×

initial mass Np pellet mass

× conversion

(2)

The existing target dissolution vessel can accommodate no more than 60 targets per dissolution activity. This is a physical limit because attempting to dissolve more than 60 targets would result in a solution too viscous to filter easily. This limit could be improved by using a larger dissolution vessel with a higher solvent capacity, but a larger vessel cannot be substituted due to space restrictions in the hot cells. Given a 60-target upper limit, at least eight target deliveries (of equal size) must occur per year to provide 432 irradiated targets to the chemical processing section. The target delivery activity is shown as a pull process using a thin circular arrow, and a delivery of a target batch does not occur until the target dissolver is ready to receive the target batch. The target dissolution process includes an aluminum dissolution step and an oxide dissolution step. The duration of the activity (i.e., machine cycle time or machine CT) is estimated at 17 days, which includes the time needed to load the targets in the dissolver, prepare reagents, dissolve the aluminum, filter and drain the aluminum waste solution, dissolve the oxide material, and filter and drain the dissolved oxide solution. The duration of the target dissolution activity is assumed to be constant regardless of the number of targets to be dissolved because the processing steps follow a scripted procedure with fixed hold times and the amount of solvent used does not vary with the number of targets to be dissolved. After the target dissolution step, the liquid solution is collected in a storage tank. This storage tank is depicted by the triangle in between the target dissolution and stream conditioning boxes in Fig. 2. The storage tank has sufficient volume to store the output of between one and several target dissolution batches. The stream conditioning step has a lead time of 1 day and a process duration of 13 days. Once the previous step is complete, this step takes over from the previous step. The process duration for this activity includes all work associated with filling the stream conditioning vessel, performing reagent and volume adjustments, and performance of analytical measurements. The duration of the stream conditioning step is assumed to be constant regardless of the liquid volume to be conditioned because the conditioning pro-

4

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

Fig. 2. Value stream map (VSM) of chemical processing section.

cess follows a procedural script with fixed hold times, and the script does not vary with the volume of liquid in the conditioning vessel. There is no tank large enough downstream of the dissolution vessel to hold the conditioned volume, so the stream conditioning vessel is used as the feed tank to the downstream solvent extraction step. It does not become available for new material until the downstream solvent extraction step is completed and the stream conditioning vessel is empty. The solvent extraction step has a lead time of 1 day and an assumed process duration of 44 days. The process duration estimate includes setup time, production of a Pu product stream, production of a Np recycle stream, production of waste aqueous and organic waste streams, further purification of the Pu product stream and the Np recycle stream either by solvent extraction or ion exchange, and analytical measurements. The duration of the solvent extraction step is assumed to be constant and does not vary with the volume of liquid to be extracted because the actual chemical separation time is relatively short in comparison to the time needed to set up and clean out the equipment. The Pu product stream is collected in a storage tank, and this collection step is indicated by the triangle symbol following the solvent extraction box in Fig. 2. The Pu powder production and packaging step occurs when 50 g of dissolved Pu product is available, and a shipment to LANL occurs when 300 g of powder is available. The production and packaging of 300 g of HS-PuO2 powder has a lead time of 1 day and an assumed duration of 24 days, and there are six powder production runs for every shipment (4 days per 50-g can). This process duration includes the work needed to set up the powder production process, produce an oxalate precipitate, calcine the precipitate to produce an oxide, and package the oxide for shipment. At the end of the last campaign, the remaining 238 Pu powder will be placed into a can even if the quantity is less than 50 g 238 Pu powder cans are moved from the hot cell area to a dedicated glove box to prepare

for shipment, and this frees the hot cell area for processing of the next batch of 238 Pu. The VSM is useful for defining some terms used in describing the simulation model. A batch refers to a batch of irradiated targets. A product package is a 300-g package of HS-PuO2 ready to be shipped to LANL. A campaign is defined as the processing of one or more batches from start to finish to produce one or more product packages. Multiple campaigns per year must be performed to meet the production goal. The time needed to process one unit of material from start to finish is called the total product cycle time (total product CT), and is minimally the sum of the process step durations. In Fig. 2, this sum is shown in the bottom right corner and is equal to 84 d + 17 d × n, where n is the number of batches dissolved per campaign. Since batches are delivered and dissolved whenever the target dissolver is available, campaigns can overlap, and processing occurs in waves until the yearly quota of targets is dissolved and processed into product. The facility allows 300 days per year for operational work, and all processing activities (i.e., campaigns) must be completed within that time. Completion of a campaign is achieved when all assigned material has been processed, and the system has zero inventory of material at the end of the year (i.e., all Np recycled, all Pu produced, all waste streams dispositioned, all tanks and process vessels empty). 2.2. Simulation setup When a process consists of discrete, successive steps having start-to-finish dependencies, a discrete-event system simulation model is useful for modeling its operational behaviors [22–24]. The 238 Pu supply process is such a process. The information from the VSM is codified in the process simulator, and the process simulation is run repeatedly using a random number generator to influence the stochastic elements of the model. The random number generator is

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

built into the simulation system and isn’t needed for systems with deterministic processes and behaviors, but those systems are rare in the physical world, and in most cases process step durations, batch arrival times, and the outcomes of processes such as quality assurance tests are subject to random variations (i.e., they conform to probability distributions). Discrete-event simulations are useful when it becomes too difficult or unreliable to determine total processing times and other process measures using deterministic schedules. There are many discrete-event simulation modeling tools on the market [25]. For this work, Process Simulator [26] (2014 Professional Version 9.2.4) by ProModel Corporation Inc. was selected. Process Simulator has a relative cost advantage compared to other simulators and is relatively easy to use. It is an add-on to Microsoft Visio [27] and is capable of translating Visio-based block flow diagrams into simulation models. Fig. 3 shows an example of how blocks in the process diagram and VSM information are converted into Process Simulator model activities. The values shown in the figure are for illustration purposes only and may not be the actual values. In the figure, this process is shown for the “declad targets (NaOH)” step in the downstream section. In the translation process, the process structure described in the VSM is used to construct a more detailed block flow diagram in Microsoft Visio. Then process information from the VSM is entered into dialog boxes for each processing unit in the diagram to govern the operational characteristics of individual streams and blocks. Batching rules for each specific process block can be set differently in many ways such as: the batch size can be a fixed number or a predefined variable quantified by reading the input before the process, and other ways. The simulation is performed by pressing the play button on the simulation engine. For the baseline process, the following assumptions were included in the simulation model. • 432 irradiated targets are processed per year (a number divisible by 36, 48, and 54). • Four campaigns are performed per year. • Three irradiated target batches are processed per campaign (twelve batches/year, 36 targets/batch). • Pu purification is performed when 50 g of purified 238 Pu is available and the powder is placed into cans. Shipment normally occurs when 300 g of powder are available for shipment. • A successive campaign begins as soon as stream conditioning is underway for the previous campaign and the receiving tank following the target dissolution step is empty. • The system tanks and vessels contain zero inventories at the start of the simulation. • The system tanks and vessels must contain zero inventory at the end of the simulation (i.e., at the end of the 300th day) for the simulation to be successful. • Human resources are assumed to be unlimited and the organization will hire enough qualified people to perform the schedule of operations. • Machine CTs are constant and do not conform to a probability distribution.

5

Table 1 Simulation Mass Balance under Assumed Operating Conditions.

Input Pu + Np + Al + other Outputs HS-PuO2 NpO2 Al Total

Mean (g)

Std. Dev. (g)

95% C.I. Range (g)

28,080 1546 13,652 12,882 28,080

N/A 70 132 62

N/A 1476–1615 13,520–13,784 12,720–13,045

also be expressed as stochastic elements using probability distribution functions, but process data are not yet available to determine how much the machine CTs vary for each step because initial process testing is still underway. With the number of targets fixed at 432 targets per year, the numbers 36, 48, and 54 become significant for the dissolver. If 36 targets are dissolved per batch, then twelve batches of targets per year must be dissolved to dissolve 432 targets. If 48 targets are dissolved per batch, then nine batches of targets must be processed to dissolve 432 targets. If 54 targets are dissolved per batch, then eight batches of targets must be processed to dissolve 432 targets. Physically, the maximum number of targets that can be dissolved in a single batch is 60, but processing targets in quantities other than 36, 48, or 54 targets would have made the model more complex than necessary for this work. 2.3. Simulation results An initial check of the model was to determine whether the mass inputs equal the mass outputs, and there is no creation or destruction of mass in the system. Table 1 shows the simulation mass balance of process inputs and outputs. Simulation of the conceptual process over six runs produced 1546 g Pu/year, 13,652 Np/year, and 12,882 g Al/year given a feed of 432 targets per year and an estimated but imprecise conversion factor in Eq. (2). The simulation runs were completed in 290–295 days. The input mass and output masses balance, and there is no loss or gain of mass in the model. The 95% confidence interval on the mean HS-PuO2 produced was 1476 to 1615 g, which indicates a significant probability that the process will meet or exceed the required production rate of 1500 g HS-PuO2 /year. Machine utilization rate is defined as the time a processing step is active divided by the total processing time. Utilization rates vary between 0% and 100%. When a processing step has a 100% utilization rate, the rate of the processing step controls the rate of the process, and the step is called a processing bottleneck [19]. If the utilization rate is less than 100%, control of the overall processing rate is shared between steps, and the steps with the highest utilization rates have the greatest influence on production rate. Fig. 4 shows the calculated utilization rates for the chemical processing section. In Fig. 4, the target dissolution step and the solvent extraction and Np/Pu purification step have the highest utilization rates at 68% and 60%, respectively, and these steps have the greatest influence on the production rate. 2.4. Sensitivity to process delays

Although the machine CTs are fixed in the model, the model contains a few stochastic elements. The material streams are split in some processes based on its probability. This occurs in the split between Al and Np + Pu streams in the target dissolution step, the split between Np and Pu in the solvent extraction step, and each gram of oxide in the target has a low probability of going to the waste stream instead of downstream purification in the target dissolution process. These stochastic elements cause the relative distribution of mass to the Np, Pu, and waste streams to fluctuate when simulations are repeated. Naturally, the machine CTs could

Operational detractors are anything that can negatively affect the performance of the process including: variability in process step durations, human resource limitations, interferences, unscheduled downtime, or interruptions in target deliveries. The Law of Variability [28] says operational detractors always degrade the performance of a manufacturing system. The effects of process delays on a manufacturing system can be measured by observing trends in three process characteristics: work-in-process (WIP), throughput (TH), and total product CT. WIP is the inventory of components

6

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

Fig. 3. Illustration of how Process Simulator model is prepared by building a block flow diagram in Microsoft Visio from the process structure described in the VSM and then populating the dialog boxes with process information from the VSM. Additional process logic is then added to the dialog boxes to complete the model.

Fig. 4. Utilization rates of steps in the chemical processing section.

or materials in the manufacturing system at any moment in time. TH is the material output of the manufacturing system over time. Total product CT was previously defined in Section 2.1. When the

process is at steady state, the process characteristics are related by Little’s Law [29] (see Eq. (3)). WIP = TH × CT

(3)

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

7

Fig. 5. Average daily WIP versus processing days under limited and unlimited feed conditions.

Little’s Law may be utilized as a basis for understanding the effects of different operating conditions and in-process inventory policies on the 238 Pu supply process. Little’s Law implies the throughput of any system can be increased by increasing WIP, reducing CT, or both. In the simulation model, WIP is the total mass of material in the system not including the reagent and solvent masses. Fig. 5 shows a plot of the daily average WIP for the baseline process under two conditions. In the first condition, the feed was limited to 432 irradiated targets. In the second condition, the feed was unlimited to allow the system to reach and maintain a steady state during the simulation. In practice, the system will not operate with unlimited feed, but running the simulation with unlimited feed reveals the steady state pattern in WIP. Both curves achieve a repeating pattern at about day 100, and then the curves deviate from each other at about day 185. The established pattern continues indefinitely when the feed is unlimited, indicating the pattern is a steady-state behavior. Any examination of Little’s Law statistics must use WIP, TH, and CT metrics calculated during this steady state behavior. A likely non-ideal condition is an increase in process step durations (i.e., machine CTs) due to unintended delays or process inefficiencies. Machine CT is the time a machine (or chemical processing step) requires to complete all operations on one piece or batch. The machine CT estimates used for the nominal case may be optimistic, and the observed machine CTs may be longer once the process is operating in an integrated fashion due to resource limitations or interferences. Under steady state conditions, increasing CT generally causes WIP to increase if TH is held constant, or TH to decrease if WIP is held constant. Table 2 shows the simulation results for the average HS-PuO2 output for the first 300 days of the simulation under the limited feed condition (432 targets/year) and different levels of increased machine CTs. At the 0% level, the machine CT estimates were set equal to nominal process step durations. For the 5% to 30% levels, the machine CTs for all process steps were increased by the defined percentages (e.g., at 0%, target dissolution lasts 17 days; at +5%, target dissolution lasts 17.9 days; at +10%, target dissolution lasts 18.7 days, etc.). The simulation results for the limited feed condition show the system does not complete processing of the required number of targets in 300 days when the machine CTs are increased for all processing steps by +5% or higher.

Increasing the machine CTs for the processing steps affects the process utilization rates. Fig. 6 shows the change in utilization rates as the machine CTs are increased during the first 300 days of operation. As the step durations increase, so do the utilization rates. The target dissolution step becomes the near bottleneck at the higher levels, and the process rate of the solvent extraction and Np/Pu purification step also has increasing effect on the overall processing rate. For the solvent extraction and Np/Pu purification step, a utilization rate is not available at the +30% level because processing of all feed material is not completed in 300 days. For the HS-PuO2 powder production and packaging step, a utilization rate is only available at the 0% level because processing of all feed material through that step is not completed in 300 days at the higher levels. Table 2 also shows Little’s Law statistics for the system under steady state conditions. The steady state production rate (TH) is the highest rate the system can achieve in the long term, and the rate cannot be increased unless changes are made to the process configuration that reduce machine CTs or increase efficiencies. All outputs of the system – Al, NpO2 , and HS-PuO2 – are considered as part of TH. At steady state and nominal machine CTs, TH is 35,100 g/year, which is equivalent to a production rate of ∼1900 g HS-PuO2 /year for a 365-d production year. When machine CTs were increased by between +5% and +20%, TH was reduced to 28,080 g/year, which is equivalent to a production rate of ∼1500 g HS-PuO2 /year for a 365d production year. When the machine CTs were increased by +25% and +30%, TH was reduced still further to 21,060 g/year, which is equivalent to production rate of ∼1100 g HS-PuO2 /year for a 365-d production year. As machine CTs were increased, the total product CT also increased. At the 0% level, the total product CT was 0.384 years, while at the +30% level, the total product CT climbed to 0.499 years. Little’s Law was used to calculate WIP given TH and total product CT at each level in Table 2. There were expected trends in WIP. When TH was higher at similar CTs, WIP was higher, and when WIP was fixed, CTs increased as TH decreased. There was also an unexpected difference. At the 0% level, the WIP calculated using Little’s Law (13,478 g) was different and higher than the average daily WIP observed in Fig. 5, which varied between ∼6500 and ∼9500 g. This difference occurred because aluminum, neptunium, and plutonium have different residence times in the process. Aluminum, which composes 46% of the mass input, leaves

8

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

Table 2 Simulation Results and Little’s Law Metrics for Conceptual Process. Simulation Outputs

Little’s Law Metrics

Increase in Machine CTs

Avg. HS-PuO2 Output (g) in First 300 days

Time Needed to Complete Processing (days)

TH (g total mass/year)

Total Product CT (year)

Calculated Average WIP (g total mass)

0% +5% +10% +15% +20% +25% +30%

1546 1452 1398 1242 1161 1139 1052

295 309a 324a 339a 354a 367a 381a

35,100 28,080 28,080 28,080 28,080 21,060 21,060

0.384 0.403 0.422 0.442 0.461 0.480 0.499

13,478 11,316 11,849 12,411 12,944 10,109 10,509

a

Processing not completed in 300 days; prohibited condition.

Fig. 6. Process utilization rates as machine CTs are increased from 0% to +30%.

the process within days after entering it. Neptunium, which composes another ∼48% of the mass input, leaves the process before plutonium. Only plutonium remains in the system for the observed total product CT. Little’s Law does not recognize differences in residence times between components. Since this study was focused on the final product, HS-PuO2 , the Little’s Law metrics were calculated using the CT of that output.

3. Alternative operating strategies The conceptual process appeared capable of meeting the production goal when there were no operational detractors, but throughput became unacceptable when time delays were introduced. Process alternatives must be sought that improve process efficiency and/or increase the processing rate. An ideal production process is one that consistently provides a quality product, produces at a constant rate (i.e., little variation in CT), suffers few disruptions, and the throughput rate matches customer demand (in Lean Manufacturing terminology, the production rate required to just meet customer demand without exceeding it is called Takt time). These conditions are met in a one-piece flow system where WIP is minimized [30]. In a physical system, variations in CT and process disruptions will occur, but the system must be designed to tolerate expected variations and disruptions without significantly affecting the ability of the process to meet customer demand.

The conceptual process addressed in this simplified analysis operates nearly as a push process. A push process is one in which WIP is not constrained [31]. In such a system, the flow of work and material from one processing unit to the next is not coordinated, and each processing unit operates independently at its own processing rate. No limits are place on the accumulation of inventory between processing steps. In an actual radiochemical production facility, inventory limits are applied to avoid nuclear criticality, reduce radiation dose, and to stay within other safety limits. Alternatively, Lean Manufacturing principles encourage the design of a pull process. A pull process is one where WIP is constrained [32]. In such a process, the flow rate of work and material from one step to the next is coordinated, and inventory only moves downstream when: (1) there is space or equipment available to work on it, (2) only if the downstream unit requests it, and (3) system inventory limits allow it. A pull process is more resilient to process variability, and throughput rates in such a system are less sensitive to increased variability. Pull systems establish a WIP upper limit that decreases flow time while maintaining similar throughput levels. A pull inventory control strategy is appropriate for a radiochemical process where inventory safety limits are in effect. There are two methods by which pull production can be implemented [31,32], Kanban and CONWIP (constant WIP). In a Kanban method, specific intermediate inventory limits are set in between each processing stage and the flow of materials is determined by the inventory level at the downstream processing unit. In a CON-

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

9

Table 3 Hybrid Operating Strategy with Conceptual Process and Increased Machine CTs. Simulation Outputs Increase in Machine CTs

Avg. HS-PuO2 Output (g) in First 300 days

Time Needed to Complete Processing (days)

0% +5% +10% +15% +20% +25% +30%

1527 1433 1276 1189 1139 1243 1005

302a 317a 332a 347a 362a 378a 393a

a

Processing not completed in 300 days; prohibited condition.

WIP method, the total inventory of the system is controlled, but specific inventory limits between processing stages are not set and inventory can accumulate without practical limit in waiting or storage areas between processing steps. Push systems are used by make-to-stock companies whereas pull systems are used in maketo-order situations. No physical production system is pure pull. All production systems are a combination of push and pull (hybrid); the degree of push versus pull depends on the way the system is designed and on the type of industry. Production systems that are a hybrid may use either Kanban or CONWIP for parts of the system that require it, and push for other parts of the system. 3.1. Hybrid operating strategy A hybrid operating strategy was simulated to determine whether it would make the system more resistant to time delays [32–34]. In the simulation model, materials were pushed into the system until forward movement became blocked, and then elements of Kanban were introducing by releasing material from one processing step to the next only when the downstream unit was available to receive it (i.e., no storage of excess intermediate materials between processing steps). Irradiated targets were still made available every 17 days to feed the target dissolver, but they were not pushed into the dissolver if the dissolver were waiting on a downstream process to be completed. In the simulation of the baseline process, irradiated targets were introduced to the dissolver as soon as the dissolver completed a dissolution, and dissolved inventory was maintained in a virtual storage area if downstream processing steps were not available. In the physical process, there is no storage tank where dissolved target solutions can be held indefinitely, and the hybrid operating strategy more closely resembles the actual system. Each process block forwards its output to the successive process block when that particular process block has finished its processes and the next process block is empty. Table 3 shows simulation statistics for the process operating with the hybrid strategy as machine CTs were increased from 0% to +30%. Unfortunately, changing the process into a pull system negatively affected performance. Even under nominal conditions, the process was not capable of meeting the production goal within 300 days. Looking deeper into the problem, the failure of the hybrid operating strategy was due to process characteristics. The processing steps with the highest utilization rates are the target dissolution step and the solvent extraction and Np/Pu purification step. Any delay in starting these steps due to unavailability of equipment slows down production. The process is sensitive to delays, and the hybrid strategy introduces delays as a normal operating strategy. 3.2. Modified hybrid operating strategy If throughput is to be maintained when variability increases or time delays occur, then either the productivity of the ratecontrolling steps must be increased, or the processing rate of the

rate-controlling steps must be increased (or both) to compensate for variability and delays. A search for alternative operating strategies was performed. Combinations of differing numbers of dissolution batches, the number of dissolution batches per conditioning event, the number of solvent extractions performed per campaign, and the number of campaigns performed per year were varied in the simulation model while holding constant the number of targets to be processed at 432, taking care to avoid processing a non-integer number of targets in any batch or non-constant batch sizes (e.g., 36 targets in Batch #1, 48 targets in batch #2, etc.). The model results were evaluated and ranked, and the combinations producing the lowest times needed to complete processing at the 0% to 30% increased machine CTs were ranked higher on the list. Best results (i.e., least number of days to complete processing) were achieved with the following combination. If the number of targets dissolved per batch is increased from 36 targets to 54 targets, and the number of target batches processed per campaign is increased from three to four, then the same number of targets (432) can be processed in two campaigns instead of four. This reduces the number of target dissolutions performed per year from twelve to eight, and the number of solvent extractions performed per year from four to two. Table 4 shows the simulation results for a process deploying two campaigns, four target batches per campaign, and 54 targets per batch (432 targets/year) using the same process inventory controls discussed previously in Section 3.1 for different levels of increased machine CTs. The table shows the desired HS-PuO2 output was maintained up to +25% increase in machine CTs. Fig. 7 shows a plot of the average daily WIP versus processing time for the conceptual process with a modified hybrid operating strategy under two conditions. The first condition assumed the feed was limited to 432 targets and two campaigns were performed, and the second condition assumed unlimited feed to allow the system to reach and maintain a steady state. In practice, the system would not operate with unlimited feed, but running the simulation with unlimited feed revealed the steady state pattern in WIP. The average daily WIP for the model under both feed conditions overlapped until day 110. After that, the average daily WIP for the limited feed condition leveled out, and then fell to zero when the last gram of HS-PuO2 was produced. The steady state pattern in WIP is not fully established until about day 180, and the system never reaches steady state for the limited feed condition. Table 4 also shows Little’s Law statistics for the steady state condition. The Little’s Law statistics were calculated using the simulation data for the unlimited feed condition because steady state was not reached during simulation of the limited feed condition. At steady state, up to the +25% level, TH was maintained at 42,120 g/year, which is equivalent to a production rate of ∼2200 g HS-PuO2 /year for a 365-d production year. When the machine CTs were increased by +30%, TH was reduced to 28,080 g/year, which

10

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

Table 4 Hybrid Operating Strategy with Modified Process and Increased CT Variability. Simulation Outputs

Little’s Law Metrics

Increase in Machine CTs

Avg. HS-PuO2 Output (g) in First 300 days

Time Needed to Complete Processing (days)

TH (g total mass/year)

Total Product CT (year)

Calculated Average WIP (g total mass)

0% +5% +10% +15% +20% +25% +30%

1521 1524 1524 1524 1534 1534 1127

251 258 264 273 285 296 307a

42,120 42,120 42,120 42,120 42,120 42,120 28,080

0.496 0.512 0.529 0.553 0.581 0.614 0.627

20,892 21,565 22,281 23,292 24,472 25,862 17,606

a

Processing not completed in 300 days; prohibited condition.

Fig. 7. Average daily WIP versus processing days under limited and unlimited feed conditions for process using modified hybrid operating strategy.

is equivalent to a production rate of ∼1500 g HS-PuO2 /year for a 365-d production year. As machine CTs were increased, the total product CT increased. At the 0% level, the total product CT was 0.496 years, while at the +30% level, the total product CT was 0.627 years. These values are higher than for the baseline process (see Table 2). This difference is due to the number of targets batches dissolved per campaign. In the baseline process, three batches of targets are dissolved in series per campaign, while for the modified process, four batches of targets are dissolved in series per campaign. Little’s Law was used to calculate WIP given TH and total product CT at each level in Table 4. Like the Little’s Law metrics calculated for the baseline process, the WIP calculated using Little’s Law for the modified hybrid process exceeds the observed average daily WIP shown in Fig. 7. At the 0% level, the calculated WIP is 20,892 g, while the average daily WIP varies between ∼8200 and ∼15,000 g. This difference is due to the residence times of aluminum, neptunium, and plutonium in the system as was discussed previously in Section 2.4. A direct comparison of operating strategies can be made by plotting the average daily production rate as a function of machine CT increase. The average daily production rate is determined by dividing the HS-PuO2 output in the first 300 days by the time needed to complete processing, or 300 days, whichever is shorter. In the actual system, HS-PuO2 is not produced daily, but the production metric provides a convenient way to compare relative production rates for competing processes. Fig. 8 was created using the data provided in Tables 2 and 4 for the limited feed condition. In Fig. 8, the minimum

required average daily production rate is 5 g HS-PuO2 /day, which is equivalent to producing 1500 g HS-PuO2 in 300 days if the system were to operate at a constant rate for a full production year. When the original operating strategy was used, the average daily production rate varied between 5.3 and 3.5 g HS-PuO2 /day. When the modified hybrid strategy was used, the average daily production rate varied between 6.1 and 3.7 g HS-PuO2 /day. The modified hybrid operating strategy has 15% higher productivity than the original operating strategy at the 0% level, and it has enough margin to allow the system to exceed the minimum average daily production rate requirement up to the +25% level. 4. Conclusions and future work The conceptual 238 Pu supply process is a batch process subject to operational detractors. Simulations show the conceptual process appears capable of achieving the production goal – 1500 g HS-PuO2 /year – without operational detractors, but the simulated process failed to reach its production goal when process time increases were introduced. Alternative operating strategies were identified based on concepts drawn from Theory of Constraints and Lean Manufacturing, and discrete-event simulations were performed to study those alternatives. Simulations results indicated inventory management alone was insufficient to improve the process, and best results were obtained by increasing the number of targets dissolved per campaign, and by reducing the number of campaigns performed per year (i.e., fewer solvent extraction runs per year).

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

11

Fig. 8. Average daily HS-PuO2 production rate versus% increase in machine CTs for the baseline and modified hybrid processes.

The development of the 238 Pu supply process is in its infancy, and much more needs to be done to determine whether the process might still achieve its goal as the fidelity of the process model is increased. So far, only increases in process step durations were simulated as potential non-idealities. Otherwise, no restrictions were placed on the total global inventory of specific nuclear isotopes in the system, the availability of human resources, storage area/tank capacities, and other constraints. Also, statistical variations in process step durations were unexamined. As constraints are identified and quantified, these will be incorporated into the model. Going forward, the Pu-238 Supply Project wants to use the model to identify which processing steps need improvement and to quantify what improvements would be meaningful in terms of reducing the machine CTs. In Figs. 4 and 6 for example, the target dissolution step and the solvent extraction steps are identified as having the highest utilization rates. These steps have not been longpracticed and optimized, and it is very likely these steps are not performed efficiently. With more detailed observation and deliberate study of these steps, the machine CTs for these steps could be reduced, resulting in a faster overall process. Another potential improvement to the target dissolution process might be to increase the 237 Np content of each target. This would increase the 238 Pu productivity of each target upon irradiation, and fewer targets would need to be dissolved per year to achieve the same HS-PuO2 output. Hypothetically, it may be possible to increase the 237 Np content in targets by 50% without weakening or melting the target cladding during irradiation due to excess heating. If such a change were allowable within the reactor safety envelope, then the number of targets dissolved per year could be reduced from 432 to 288 with a commensurate reduction in process time to completion. Taking this idea to the limit, a target design could be conceived using a pure NpO2 pellet with no aluminum content. Such a change would allow the number of targets dissolved per year to be reduced to a minimum of 96 targets. However, such a change would compel the use of zirconium alloy as a target cladding material instead of aluminum to prevent melting of the target cladding during irradiation, and the safety case for use of such a target in ATR or HFIR has yet to be developed.

The exchange of information between the laboratory testing effort and the modeling effort comprises a two-way exchange. Data measured in the laboratory will be used to improve the process model, and simulations of the process model will be used to identify process alternatives to be tested in the laboratory. Although the target dissolution step and the solvent extraction and Np/Pu purification step were identified as the process steps with the greatest influence on production rate, future process changes and inclusion of greater process details in the model may shift the constraint to other parts of the process. Should this occur, the methods employed here may be used to identify those process points and to develop alternatives to alleviate the bottlenecks. The machine cycle times will be varied based on probability once there are enough data on distribution functions. Acknowledgements The U.S. Department of Energy Office of Space and Defense Power Systems Office of Nuclear Energy (NE-75) and the National Aeronautics and Space Administration sponsored this work. This manuscript was authored by the University of Tennessee and UTBattelle, LLC, under contract no. DE-AC05-00OR22725 with the U.S. Department of Energy. The authors thank R Chris Wright for collecting the initial data, for developing the initial VSM, and for starting the work to define the model for the baseline system. References [1] Rosenberg KE, Johnson SG. Assembly and testing of a radioisotope power system for the new horizons spacecraft. In: 4th Int. Energy Convers. Eng. Conf. Exhib. 2006., http://dx.doi.org/10.2514/6.2006-4031. p. 4192. [2] Groh HJ, Poe WL, Porter JA. Development and performance of processes and equipment to recover neptunium-237 and plutonium-238. In: 50 Years Excell. Sci. Eng. Savannah River Site Proc. Symp. 2000. p. 165–78. [3] Rankin DT, Kanne Jr WR, Louthan Jr MR, Bickford DF, Congdon JW. Production of Pu-238 oxide fuel for space exploration. In: 50 Years Excell. Sci. Eng. Savannah River Site Proc. Symp. 2000. p. 179–86. [4] Witze A. Nuclear power: desperately seeking plutonium. Nature 2014;515(11):484–6, http://dx.doi.org/10.1038/515484a. [5] Miller Jr WF. Start-up Plan for Plutonium-238 Production for Radioisotope Power Systems; 2010. Report to Congress. [6] Wong AS. Chemical analysis of plutonium-238 for space applications. In: AIP Conf. Proc. Albuquerque. 2001. p. 753–7.

12

T. Thomas et al. / Journal of Manufacturing Systems 46 (2018) 1–12

[7] Vondy DR, Lane JA, Gretsky AT. Production of Np237 and Pu238 in thermal power reactors. I&EC Proc Des Dev 1964;3(4):293–6. [8] Todd T, Herbst R, Mincher B, Frank S, Law J, Swanson J. Plutonium-238 recovery from irradiated targets using solvent extraction. In: Proc. 15th Pacific Basin Nucl. Conf. 2006. [9] Summary of Plutonium-238 Production Alternatives Analysis Final Report. Idaho National Laboratory; 2013. [10] Supplemental Analysis for the Nuclear Infrastructure Programmatic Environmental Impact Statement for Plutonium-238 Production for Radioisotope Power Systems; 2013. Washington D.C. [11] Kitts FG. Pilot-Scale Demonstration of the Modified Direct Denitration Process to Prepare Uranium Oxide for Fuel Fabrication Evaluation. Oak Ridge, Tennessee: Oak Ridge National Laboratory; 1994. [12] Chapman TG. HFIR Target Design Study. Oak Ridge, Tennessee: Oak Ridge National Laboratory; 1965. [13] King LJ, Mathern JL. Containment of radioactive material in the transuranium processing plant. In: Proc. 14th Conf. Remote Syst. Technol. 1966. [14] King LJ. Current status of the transuranium processing plant. In: Proc. 27th Conf. Remote Syst. Technol. 1979. [15] King LJ, Bigelow JE, Collins ED. Experience in the separation and purification of transplutonium elements in the transuranium processing plant at Oak Ridge National Laboratory. In: Proc. Symp. Ind. Prod. Transplutonium Elem. 1980. [16] Burney GA. Ion exchange process for recovery of 238 Pu from irradiated 237 Np. I&EC Proc Des Dev 1964;3(4):328–32. [17] Mew DA, Krikorian OH, Dodson KE, Schmitz JA. Purification of 238 Pu Oxide using the Pu Oxalate Process. Livermore, California: Lawrence Livermore National Laboratory; 2001. [18] Rother M, Shook J. Learning to See: value-stream mapping to create value and eliminate MUDA. Cambridge, Massachusetts: The Lean Enterprise Institute; 2003. [19] Goldratt EM. The Theory of Constraints and how should it be implemented? Great Barrington. Massachusetts: North River Press; 1990. [20] Sayer NJ, Williams B. Lean for Dummies. 2nd ed. Hoboken, New Jersey: John Wiley & Sons; 2012.

[21] Raja V, Fernandes KJ, editors. Reverse Engineering: An Industrial Perspective. Cardiff, United Kingdom: Springer-Verlag London Limited; 2008. [22] Pyzdek T, Keller P. The Six Sigma Handbook. 4th ed. New York: McGraw Hill Education; 2014. [23] Law AM, Kelton WD. Simulation Modeling and Analysis. 2nd ed. New York: McGraw-Hill; 1991. [24] Banks J, Carson II JS, Nelson BL, Nicol DM. Discrete-Event System Simulation. 5th ed. Upper Saddle River, New Jersey: Pearson; 2009. [25] Simulation software survey, 2015. http://www.orms-today.org/surveys/ Simulation/Simulation.html [Accessed 21 April 2016]. [26] Process Simulator, 2016. https://www.promodel.com/products/ processsimulator/ [Accessed 21 April 2016]. [27] Microsoft Visio; 2016, https://products.office.com/en/Visio/visio-standardflowcharts-and-diagrams [Accessed 27 April 2016]. [28] Hopp WJ, Spearman ML. Factor Physics. 3rd ed. Long Grove, Illinois: Waveland Press; 2008. [29] Little JDC, Graves SC. Building intuition: Insights from Basic Operations Management Models and Principles. New York: Springer Science + Business Media LLC; 2008. [30] Protzman C, McNamara J, Protzman D. One-Piece Flow vs. Batching: A Guide to Understanding How Continuous Flow Maximizes Productivity and Customer Value. Boca Raton, Florida: CRC Press; 2016. [31] Hopp WJ, Spearman ML. To pull or not to pull: what is the question? Manuf Serv Oper Manage 2004;6(2):133–48. [32] Gaury EGA, Pierreval H, Kleijnen JPC. An evolutionary approach to select a pull system among Kanban, Conwip, and Hybrid. J Intell Manuf 2000;11:157–67. [33] Askin RG, Goldberg JB. Design and Analysis of Lean Production Systems. John Wiley & Sons; 2001. [34] Mondon Y. Toyota Production System: An Integrated Approach to Just in Time. Boca Raton, Florida: CRC Press; 2001.