Flow, Pull and Production System Design

Introduction

Lean manufacturing is fundamentally a system-design discipline. It seeks to deliver customer value with the least necessary lead time, inventory, motion, waiting, rework and managerial complexity. Flow, pull and production leveling are the mechanisms through which lean principles become an operating system rather than an isolated collection of improvement tools.

A flow-oriented system links processes so that work moves at a stable rate and defects or shortages become visible quickly. A pull system authorizes production in response to downstream consumption rather than speculative forecasts. Production-system design integrates these ideas with takt-based planning, cellular layouts, supermarkets, Kanban, FIFO lanes, heijunka and bottleneck management.

The central managerial challenge is not to maximize the utilization of every individual resource. It is to design a system that reliably converts material and information into customer output. Local efficiency can increase queues and lead time when it creates excess batch production. Conversely, a deliberately protected bottleneck may appear less efficient locally but improve total throughput.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain the relationships among takt time, cycle time, lead time, throughput and work-in-process.
2. Design continuous-flow and cellular systems under realistic process constraints.
3. Select and size Kanban, supermarkets and FIFO lanes.
4. Apply heijunka and one-piece-flow principles to demand and product-mix variability.
5. Identify bottlenecks and use Theory of Constraints insights in lean system design.
6. Evaluate trade-offs involving capacity, changeover, inventory, variability and resilience.
7. Critically assess the managerial, technological and governance implications of pull systems.

1. From Process Improvement to System Design

A production system has at least three coupled flows:

• Material flow: the physical movement and transformation of parts.
• Information flow: signals that authorize, sequence and coordinate work.
• Cash and value flow: the economic consequences of inventory, throughput, quality and delivery.

A lean design attempts to make these flows visible and synchronized. The ideal condition is continuous one-piece flow: each item moves immediately to the next operation, without waiting in a queue. This ideal is not universally economical or technically feasible. Product variety, long setup times, unreliable equipment, shared resources and safety requirements may justify controlled decoupling through supermarkets or FIFO lanes.

A useful design hierarchy is:

1. Establish customer demand and takt.
2. Examine process capacity and identify constraints.
3. Group compatible products and processes into flow or cells.
4. Reduce changeover and stabilize operations.
5. Choose the appropriate control mechanism: direct flow, FIFO, supermarket pull or a hybrid.
6. Level the sequence and continuously improve the constraint and the whole system.

This hierarchy prevents a common error: installing Kanban cards before the process is stable enough to respond to them.

2. Takt-Based Planning and Capacity

Takt time

Takt time is the available production time per period divided by required customer quantity for that period.

Available production time must exclude planned breaks, meetings and other non-production intervals. If a shift provides 25,200 seconds of net available time and demand is 420 units, then:

The result means that, on average, one completed unit must leave the system every 60 seconds to satisfy demand. Takt is a demand-based pacing requirement; it is not the same as the actual cycle time of a machine.

Cycle time and capacity

Cycle time is the elapsed processing time between completed units at a particular operation under stated conditions. For a stable single-process operation, approximate capacity per period is:

The effective cycle time should include relevant losses such as minor stoppages, or capacity should be adjusted separately for availability, performance and quality. If an operation has a 52-second cycle time and 25,200 available seconds, its ideal output is approximately 484.6 units per shift. This does not guarantee that output because quality losses, downtime, material shortages and labor constraints may reduce effective capacity.

For a process to support takt under the stated assumptions:

A process with a 68-second effective cycle time against a 60-second takt is a capacity constraint unless additional parallel capacity, overtime, work redistribution or process improvement is provided.

Line balancing

For a product requiring total manual work content W, with takt time T, the theoretical minimum number of stations is:

Since stations are indivisible in practice, the required number is at least the next whole number, subject to precedence constraints and practical balancing. Line-balance efficiency is commonly expressed as:

where N is the number of stations. This measure is useful but incomplete: a high percentage does not guarantee smooth flow if the line has poor ergonomics, unreliable equipment or uneven demand.

Worked example. A product has 276 seconds of manual work content and a 60-second takt. The theoretical minimum is 4.6 stations, so at least five stations are required. If precedence constraints require six stations, efficiency is:

The 23.3% balance loss is not automatically waste. Some loss may be necessary because tasks cannot be split, must follow a sequence, or require specialized equipment.

Staffing and flexible work

Takt-based planning also supports staffing decisions. If demand changes by time period, managers should distinguish permanent capacity from flexible capacity. Cross-training, standardized work and modular cells allow labor to be reassigned without destabilizing quality. However, flexible staffing must not become a substitute for safe workload design or reasonable employment conditions.

3. Continuous Flow, One-Piece Flow and Cellular Manufacturing

Continuous flow

Continuous flow exists when products move through successive operations with minimal waiting and a controlled sequence. It is strongest when demand is relatively stable, product families are similar and equipment can be arranged according to process sequence.

One-piece flow does not necessarily mean that every machine processes only one item at a time. It means that the transfer batch between operations is one, or as close to one as technically and economically feasible. The design objective is to expose problems rapidly and reduce waiting, not to pursue a ritual number regardless of context.

Benefits include shorter throughput time, lower work-in-process, rapid defect detection and simpler visual control. Risks include greater sensitivity to downtime and quality interruptions. Therefore, one-piece flow requires reliable equipment, quick response to abnormalities, balanced work and disciplined standardized work.

Cellular manufacturing

A cell groups machines, people and materials according to a product family or value stream rather than a functional department. A U-shaped cell may reduce walking and support flexible staffing, but physical shape alone does not create flow. The cell must have compatible routing, appropriate equipment, clear standard work and a replenishment method.

Product-family analysis can use routing data, similarity matrices or production-flow analysis. A cell is appropriate when products share enough processing requirements to justify dedicated or semi-dedicated flow. If product families are wrongly grouped, the result may be excess movement, frequent setups and hidden queues.

A cellular design should answer five questions:

1. What products and demand volumes belong to the family?
2. What is the takt and required staffing pattern?
3. Which tasks can be combined or redistributed?
4. Where are quality checks and abnormality responses located?
5. What happens when demand, mix or equipment availability changes?

4. Pull Control: Kanban, Supermarkets and FIFO

Push and pull

In a push system, upstream operations produce according to a schedule or forecast and send output forward. In a pull system, downstream consumption triggers replenishment. Pull does not eliminate planning; rather, it separates strategic planning from local replenishment authorization.

Pull is especially valuable where demand is repetitive and the process can replenish within an agreed response time. It is less suitable when demand is highly erratic, products are engineered to order, or supply and process lead times are too long to protect customer service without substantial decoupling inventory.

Kanban

Kanban is a signal that authorizes production or movement of a defined quantity. A card, electronic message, container, barcode or radio-frequency signal can serve this function. A valid Kanban normally specifies part identity, quantity, source, destination and handling information.

The basic control rule is: consume first, then replenish the consumed quantity. Upstream production must not exceed the number of authorized signals. This creates a limit on work-in-process and makes shortages, downtime and quality problems visible.

A common approximate sizing relationship is:

where N is the number of Kanban containers or cards, D is demand rate in units per time, L is replenishment lead time in the same time unit, S is the safety factor expressed as a decimal, and C is units per container. The result must be rounded up to a whole number because partial containers are not normally practical.

Assumptions: demand is reasonably represented by an average rate; lead time is defined consistently; the container quantity is fixed; the safety factor reflects variability or desired protection; and each signal represents one replenishment quantity.

Example. Demand is 480 units per day, replenishment lead time is 0.25 day, safety factor is 20%, and each container holds 30 units:

Therefore, at least five containers are required under the stated assumptions. The calculation is a starting point, not a substitute for observing actual shortages, lead-time distribution and container-return discipline.

Supermarkets

A supermarket is a controlled inventory point from which a downstream process withdraws material and an upstream process replenishes what was consumed. It decouples processes with different production rhythms while preserving a pull relationship. It is appropriate when direct flow is not feasible because of distance, different batch sizes, changeovers or process reliability.

Supermarkets should have defined maximum and minimum levels, clear locations, standard replenishment rules and ownership. Uncontrolled inventory placed between processes is not a supermarket; it is an unmanaged queue.

FIFO lanes

A first-in, first-out lane preserves sequence between two processes. It is useful when the upstream and downstream processes can remain closely linked but operate at different speeds or have short-term interruptions. A FIFO lane needs a capacity limit. When the lane is full, upstream production stops or changes to an approved response. Without a limit, FIFO becomes a disguised accumulation of work-in-process.

5. Heijunka and Demand Variability

Heijunka means leveling production by volume and mix over a planning interval. Suppose daily demand is 100, 80 and 120 units across three days. Producing exactly the daily demand creates a varying workload. If inventory, staffing and customer agreements permit, a leveled plan may produce approximately 100 units each day, with mix sequenced in a repeating pattern.

A heijunka sequence should consider:

• total volume over the leveling interval;
• product mix and pitch;
• setup time and changeover capability;
• material availability;
• bottleneck capacity; and
• customer service requirements.

Pitch is a practical scheduling interval associated with a standard quantity, often linked to container size and takt. If takt is 60 seconds and a container holds 20 units, the pitch is 1,200 seconds, or 20 minutes. A visual schedule may then authorize one container’s worth of production every 20 minutes, while actual work proceeds continuously.

Leveling reduces demand amplification, uneven workloads and excessive finished-goods inventory. It cannot remove genuine demand variability. If customer demand is volatile, the system may require strategic finished-goods inventory, flexible capacity, postponement or differentiated service policies.

6. Bottlenecks and Theory of Constraints

A bottleneck is a resource whose available capacity is insufficient relative to the flow demanded of it. Theory of Constraints frames improvement through five focusing steps:

1. Identify the system constraint.
2. Exploit it using existing resources and better operating discipline.
3. Subordinate other activities to the constraint’s requirements.
4. Elevate the constraint through added capacity or process improvement.
5. Repeat when the constraint moves.

The implications for lean design are significant. A non-bottleneck should not produce merely to maximize its own utilization. Doing so creates inventory in front of the constraint and increases lead time. Instead, upstream processes should be synchronized with the constraint, often through a controlled buffer or a drum-buffer-rope arrangement.

The drum is the constraint’s operating rhythm; the buffer protects it from disruption; and the rope communicates release authorization upstream. This logic is compatible with lean pull when the release signal is tied to actual system need rather than local utilization targets.

A practical bottleneck analysis examines:

• demand load versus effective capacity;
• downtime and setup losses;
• quality losses at or before the constraint;
• labor and material availability;
• queue length and waiting time; and
• whether apparent congestion is caused by scheduling rather than true capacity shortage.

Little’s Law provides a useful system-level relationship:

WIP is average work-in-process, throughput rate is average completed units per time, and flow time is average time in the system. Units must be consistent. For example, 40 units per hour and 0.5 hour average flow time imply 20 units of average WIP. Little’s Law describes a stable system over a suitable observation period; it does not by itself explain causality or guarantee that reducing WIP will preserve throughput.

7. Variability, Capacity and Resilience

Flow systems are vulnerable to variability because queues form when arrivals temporarily exceed service capacity. Important sources include demand variation, processing-time variation, machine failures, quality problems, absenteeism and supplier unreliability.

Increasing utilization toward 100% generally reduces available recovery capacity and can cause disproportionate waiting, especially when variability is high. Consequently, a design may require capacity buffers, time buffers, inventory buffers or flexible labor. The correct buffer depends on the failure mode and the cost of protection.

Managers should avoid two simplistic claims: first, that all inventory is waste; and second, that more capacity always solves flow problems. Inventory can protect customers from unreliable supply, but excessive inventory hides abnormalities. Additional capacity can protect a bottleneck, but capacity added at a non-constraint may increase WIP without increasing throughput.

Useful measures include:

Metrics should be balanced. Rewarding only utilization encourages overproduction; rewarding only inventory reduction can create stockouts; rewarding only output can encourage defects.

8. Integrated Design Method

A graduate-level production-system design can follow this sequence:

Step 1: Define the demand boundary. Establish product families, customer demand, operating days, service requirements and demand variability.

Step 2: Calculate takt by relevant period. Use net available time and distinguish average demand from peak or contractual demand.

Step 3: Map the current state. Record processing, waiting, inventory, information signals, quality losses and changeovers.

Step 4: Determine effective capacity. Use observed performance rather than nameplate rates. Identify the constraint and test whether it changes by product mix.

Step 5: Select the flow pattern. Use direct flow where routing and rhythm are compatible; cells for coherent product families; FIFO where short-term synchronization is possible; and supermarkets where decoupling is necessary.

Step 6: Design the pull signal. Define container quantity, replenishment lead time, signal ownership, maximum WIP and abnormality rules.

Step 7: Level the sequence. Use heijunka where demand and setup conditions permit, while protecting the constraint.

Step 8: Validate through a pilot. Measure lead time, WIP, service, quality and operator burden. Revise standards using evidence.

Step 9: Institutionalize improvement. Audit signal discipline, review buffer sizes, solve recurring abnormalities and periodically recalculate capacity.

Applied Case: Illustrative Medical-Device Assembly Cell

Background

A fictional medical-device manufacturer assembles a disposable pump in three major stages: component preparation, electronic testing and final packaging. Daily demand is 420 units. Net available time is 25,200 seconds per shift, giving a 60-second takt.

Problem

The functional departments operate in batches. Preparation releases 200-unit lots, testing waits for full lots, and packaging frequently experiences shortages followed by overtime. Average WIP is 620 units and average flow time is approximately 1.48 shifts. Defects discovered during testing require extensive sorting because preparation and testing are physically separated.

Analysis

Observation shows that preparation can complete a unit in 48 seconds, testing requires 72 seconds, and packaging requires 55 seconds. Testing is therefore above the 60-second takt and is the immediate constraint. Changeover consumes 45 minutes twice per shift because products are sequenced in large batches. The system is not suffering simply from insufficient labor; it is suffering from batching, an over-takt constraint and delayed feedback.

Approach and solution

Management first separates product families and introduces standardized work. A focused improvement reduces testing cycle time to 58 seconds by relocating a measurement step and eliminating repeated data entry. A small supermarket of tested units is retained because testing and packaging require different staffing rhythms. The supermarket is capped at 90 units and replenished through two-container Kanban, with 30 units per container. A FIFO lane of 20 units links preparation to testing so that defects are identified quickly without creating unlimited accumulation.

A heijunka sequence releases three product variants in a repeating pattern. Changeover work is moved externally where possible, reducing each changeover to 18 minutes. Electronic Kanban is used for traceability, but the physical container and visual location remain the primary controls for operators.

Outcome

After a controlled pilot, average WIP falls from 620 to 250 units, and average flow time falls correspondingly when throughput remains approximately 420 units per shift. Packaging shortages decline, and defects are detected closer to their source. Testing remains the constraint, so its availability is monitored closely; management does not attempt to increase preparation output merely to improve preparation utilization.

Learning points

The case illustrates that pull is not simply a card system. It requires capacity analysis, a consciously sized decoupling point, leveled sequencing, quality at the source and a constraint-centered management logic. The supermarket is useful because it decouples mismatched rhythms, but its cap prevents inventory from becoming invisible.

Technology, Governance and Contemporary Issues

Digital Kanban, real-time location systems, machine data and artificial-intelligence forecasting can improve visibility and detect abnormal patterns. However, technology does not create pull automatically. Poor master data, inaccurate cycle times or automated releases can amplify errors at higher speed.

AI may support demand sensing, anomaly detection, predictive maintenance and sequencing. Its use should be governed by transparent decision rules, human review for abnormal conditions, cybersecurity controls and protection of worker data. A model that recommends production without explaining its assumptions may undermine the visual management and learning required by lean systems.

Ethically, system design must consider ergonomic exposure, psychological pressure from takt, equitable allocation of overtime and the effect of monitoring technologies on privacy. Regulatory environments such as medical-device or aerospace manufacturing may require complete traceability, validated processes and controlled changes. In such settings, lean flow must strengthen—not bypass—quality and compliance controls.

Challenges and Limitations

Flow and pull designs can fail when demand is too irregular, product variety is too high, suppliers are unreliable or equipment requires long batch runs. Kanban calculations based only on average demand can understate protection needed for high variability. Heijunka can be misapplied when management levels production without considering customer commitments or constraint capacity. One-piece flow can increase disruption if equipment reliability and problem-solving capability are weak.

Lean systems also risk transferring pressure to employees when takt is treated as an unchanging performance quota rather than a planning signal. A mature system distinguishes normal work from abnormal conditions, stops for safety and quality, and uses deviations as learning opportunities.

Best-Practice Principles

1. Design from customer demand, but validate demand assumptions.
2. Use effective capacity rather than nameplate capacity.
3. Protect the constraint and subordinate non-constraints.
4. Prefer direct flow when technically and economically feasible.
5. Use FIFO or supermarkets deliberately, with explicit limits.
6. Size Kanban from demand, lead time, container quantity and variability.
7. Level volume and mix where setup and demand conditions permit.
8. Treat visual management and standardized work as control infrastructure.
9. Balance throughput, quality, delivery, cost, safety and employee well-being.
10. Recalculate and improve the system when demand, mix or process capability changes.

Chapter Summary

Flow, pull and production-system design connect lean principles to operating decisions. Takt translates customer demand into a required pace; cycle time and effective capacity determine whether that pace is feasible. Continuous flow and cellular manufacturing reduce waiting when product routing and process capability are compatible. Where direct flow is not practical, FIFO lanes and supermarkets provide controlled decoupling.

Kanban authorizes replenishment and limits work-in-process, while heijunka reduces unevenness in volume and mix. One-piece flow improves visibility but requires stability, rapid problem response and reliable equipment. Theory of Constraints adds a system perspective: throughput is governed by the constraint, so upstream activity must be synchronized rather than optimized locally. Little’s Law links WIP, throughput and flow time, but must be applied to stable systems with consistent units.

Effective design therefore combines quantitative analysis with practical judgment. It considers variability, changeover, quality, labor, resilience, safety, regulation and technology. The goal is not minimum inventory or maximum utilization in isolation; it is a capable, transparent and continuously improving system that delivers customer value.

Key Terms

Takt time: Available net production time divided by required customer demand.

Cycle time: Time between completed units at an operation under defined conditions.

Effective capacity: Realistic output capability after relevant losses and constraints.

Continuous flow: Linked processing with minimal waiting and controlled movement.

One-piece flow: Transfer of units individually or in very small quantities between operations.

Cellular manufacturing: Arrangement of resources around a product family or value stream.

Pull system: Production and replenishment authorized by downstream consumption.

Kanban: A signal authorizing production or movement of a defined quantity.

Supermarket: Controlled inventory point replenished in response to withdrawal.

FIFO lane: Capacity-limited queue that preserves first-in, first-out sequence.

Heijunka: Leveling of production volume and product mix.

Pitch: Scheduling interval associated with a standard quantity and takt.

Bottleneck: Resource whose effective capacity limits system flow.

Constraint: The factor that most restricts achievement of the system goal.

Drum-buffer-rope: Constraint-based method linking pace, protection and release.

Work-in-process: Items that have entered production but are not complete.

Flow time: Time an item spends in the production system.

Line-balance efficiency: Work content divided by available station time, expressed as a percentage.

Review Questions

Short Answer Questions

1. Distinguish takt time from cycle time.
2. What is the purpose of a supermarket in a pull system?
3. Why must a FIFO lane have a capacity limit?
4. State the five focusing steps of the Theory of Constraints.
5. What information does a Kanban signal normally contain?

Long Answer Questions

1. Explain how takt-based planning, line balancing and effective capacity analysis should be integrated when designing a production cell.
2. Compare direct flow, FIFO control and supermarket pull, including the conditions under which each is appropriate.
3. Discuss how heijunka reduces instability while recognizing the limits imposed by demand variability and changeover requirements.
4. Explain why maximizing local utilization can reduce total system performance in a constrained production system.
5. Critically evaluate the benefits and risks of digital Kanban and AI-supported production control.

Analytical / Application Questions

1. A plant has 28,800 seconds of net time per shift and demand of 480 units. One operation has an effective cycle time of 70 seconds. Calculate takt time, determine whether the operation can meet demand, and propose two system-level responses.
2. Demand is 360 units per day, replenishment lead time is 0.4 day, safety factor is 25%, and container size is 24 units. Calculate the required number of Kanban containers, rounding up, and discuss what observations would justify revising the result.
3. A value stream has throughput of 50 units per hour and average WIP of 175 units. Use Little’s Law to calculate average flow time, then explain why reducing WIP without protecting the bottleneck could damage customer service.

Multiple Choice Questions with Answers

Question 1

A plant has 25,200 seconds of net available time and customer demand of 420 units per shift. What is the takt time?

A. 50 seconds per unit

B. 60 seconds per unit

C. 70 seconds per unit

D. 80 seconds per unit

Correct Answer: B. 60 seconds per unit

Question 2

Which statement best distinguishes takt time from cycle time?

A. Takt is machine processing time, while cycle time is customer demand.

B. Takt is demand-based required pace, while cycle time is actual operating time between outputs.

C. Takt applies only to bottlenecks, while cycle time applies only to cells.

D. Takt measures inventory, while cycle time measures quality.

Correct Answer: B. Takt is demand-based required pace, while cycle time is actual operating time between outputs.

Question 3

Demand is 480 units per day, replenishment lead time is 0.25 day, safety factor is 20%, and container size is 30 units. How many Kanban containers are required?

A. 3

B. 4

C. 5

D. 6

Correct Answer: C. 5

Question 4

What is the primary function of a supermarket in a lean production system?

A. To maximize upstream utilization

B. To provide uncontrolled excess inventory

C. To decouple processes while replenishing what downstream consumes

D. To eliminate the need for demand planning

Correct Answer: C. To decouple processes while replenishing what downstream consumes

Question 5

Which condition most directly supports one-piece flow?

A. Unreliable equipment and long response times

B. Stable processes, balanced work and rapid abnormality response

C. Large economic batches and hidden queues

D. High product variety with incompatible routings

Correct Answer: B. Stable processes, balanced work and rapid abnormality response

Question 6

Under Theory of Constraints, what should management do first?

A. Increase every department’s utilization

B. Identify the system constraint

C. Add inventory to every process

D. Reduce staffing at the constraint

Correct Answer: B. Identify the system constraint

Question 7

If throughput is 40 units per hour and average flow time is 0.5 hour, what is average WIP according to Little’s Law?

A. 20 units

B. 40 units

C. 80 units

D. 120 units

Correct Answer: A. 20 units

Question 8

What is the main purpose of heijunka?

A. To increase batch size without limit

B. To level production volume and mix

C. To eliminate all finished-goods inventory

D. To maximize the utilization of non-constraints

Correct Answer: B. To level production volume and mix

Question 9

A FIFO lane becomes full. In a properly designed system, what should normally occur?

A. Upstream production continues indefinitely

B. The lane capacity is increased automatically

C. Upstream production stops or follows a defined abnormality response

D. The downstream process is bypassed without authorization

Correct Answer: C. Upstream production stops or follows a defined abnormality response

Question 10

Why can maximizing a non-bottleneck’s utilization harm total performance?

A. It always reduces quality at the bottleneck

B. It can create excess WIP and queues without increasing system throughput

C. It prevents any use of standardized work

D. It automatically reduces customer demand

Correct Answer: B. It can create excess WIP and queues without increasing system throughput