13.2 Pull Systems, Kanban, Just-In-Time (JIT), and SMED Setup Reduction

Traditional manufacturing systems operate on a Push logic: a centralized Master Production Schedule pushes production orders onto the shop floor according to predetermined forecasts, routing batches into inventory queues regardless of whether the downstream workstation is ready to process them. In contrast, Lean manufacturing utilizes Pull logic governed by Just-In-Time (JIT) principles: parts are produced and moved only when authorized by an explicit signal from a downstream customer workstation.

---

1. Push vs. Pull Production Systems

Push System (Forecast-Driven / MRP)
 [Supplier] ──Push──► [Workstation 1] ──Push──► [Workstation 2] ──Push──► [Finished Goods]
   ▲                         ▲                        ▲
   └────── Central Master Production Schedule ────────┘ (Dispatched by schedule)

Pull System (Demand-Driven / Kanban)
 [Supplier] ◄──Pull── [Workstation 1] ◄──Pull── [Workstation 2] ◄──Pull── [Customer Demand]
             ──Parts──►               ──Parts──►               ──Parts──►
             (Visual signal triggers upstream production & movement)

---

2. Kanban Signaling Systems

A Kanban (Japanese for "signboard" or "visual card") is an authorization device that regulates the flow of materials in a pull production system. Without a Kanban card, no part may be produced, transferred, or stored.

Major Types of Kanban

1. Production Kanban (P-Kanban): Authorizes an upstream work center to manufacture a specific batch of parts to replenish a consumed container.
2. Withdrawal / Conveyance Kanban (C-Kanban / W-Kanban): Authorizes material handlers to transfer a standardized container of parts from an upstream storage buffer (supermarket) to a downstream point-of-use workstation.
3. Two-Bin System: A simplified mechanical pull mechanism where two identical bins of fasteners or hardware sit at a workstation. The operator draws components exclusively from Bin 1. When Bin 1 is empty, it is sent to the central storeroom as a replenishment signal; Bin 2 provides sufficient inventory to cover the replenishment lead time ($L$).

Dual-Card Kanban Loop Mechanics

 Downstream Assembly                 Supermarket Buffer                 Upstream Work Center
 ┌───────────────────┐               ┌───────────────────┐              ┌───────────────────┐
 │ 1. Operator pulls │               │ 2. C-Kanban card  │              │ 4. P-Kanban card  │
 │    parts from tote│               │    authorizes tote│              │    triggers batch │
 │    and detaches   │──────────────►│    retrieval from ├─────────────►│    fabrication at │
 │    C-Kanban card. │               │    supermarket.   │              │    machine.       │
 └───────────────────┘               └───────────────────┘              └───────────────────┘
           ▲                                   │                                  │
           │                                   ▼                                  ▼
           └───────────────────────────────────┴──────────────────────────────────┘
                      3. Tote replenished & returned to Point of Use

The NCEES Kanban Sizing Formula

To ensure that production is not starved while preventing excess inventory, the number of Kanban cards (or circulating containers) $k$ is calculated using the formula:

$k = \left\lceil \frac{D \cdot L \cdot (1 + \alpha)}{C} \right\rceil$

where:

• $k$ = number of Kanban card sets / containers (always rounded up to the next integer via the ceiling function $\lceil \dots \rceil$)
• $D$ = average demand rate per unit time (e.g., parts/hour, units/day)
• $L$ = replenishment lead time per container, expressed in identical time units as $D$ (sum of waiting, transit, setup, processing, and inspection durations)
• $\alpha$ = safety factor / policy buffer ($0 \le \alpha \le 1$, reflecting process variability and demand volatility; e.g., $\alpha = 0.20$ provides a 20% safety margin)
• $C$ = container capacity (the fixed number of parts held in each standardized tote/bin)

Maximum WIP Inventory: The absolute maximum inventory that can accumulate between two workstations is strictly bounded: $WIP_{\max} = k \times C$ Reducing lead time $L$ or reducing container size $C$ drives WIP down toward one-piece flow.

---

3. Single-Minute Exchange of Die (SMED)

Developed by industrial engineer Shigeo Shingo at Toyota, Single-Minute Exchange of Die (SMED) is a systematic methodology for reducing machine changeover and tooling setup times to less than 10 minutes (single digits of minutes).

Why Setup Reduction is Essential to Lean Flow

Under classical economic order quantity (EOQ) logic, large setup times mandate large production batches to amortize high fixed changeover costs ($Q^* = \sqrt{2DS/H}$). However, large batches create massive inventory piles, extended lead times, and scheduling inflexibility. Shingo demonstrated that setup time is not a fixed constraint, but an engineered variable. By slashing setup time toward zero, economical batch sizes collapse toward one-piece flow without incurring financial cost penalties:

$\text{As } S \to 0, \quad Q^* = \sqrt{\frac{2DS}{H}} \to 1$

The Three Sequential Stages of SMED

                      The SMED Implementation Hierarchy
                      
 [ Baseline Setup ] ──► Total Changeover Time (e.g., 90 minutes)
          │
          ▼
    Stage 1: Segregate Internal vs. External Operations
          │  - Identify steps done while machine runs
          │  - Result: 90 min down to 50 min internal
          ▼
    Stage 2: Convert Internal Operations to External Operations
          │  - Pre-heat dies, pre-stage tooling, standardize die shut heights
          │  - Result: 50 min down to 20 min internal
          ▼
    Stage 3: Streamline All Remaining Operations
          │  - One-turn clamps, hydraulic fasteners, parallel operators
          ▼
 [ Target State ] ──► Single-Minute Setup (< 10 minutes)

1. Stage 1: Segregate Internal from External Setup:
   • Internal Setup: Operations that can only be performed while the machine is completely stopped (e.g., physically removing the old stamping die, mounting the new die onto the platen, connecting hydraulic lines).
   • External Setup: Operations that can be performed while the machine is actively running and producing good parts on the prior job (e.g., retrieving the new die from storage, pre-heating the die, staging raw material coils, retrieving torque wrenches, reading work orders).
   • Initial Gain: Simply moving external operations outside the downtime window typically slashes machine stoppage by 30% to 50% immediately with zero capital expenditure.
2. Stage 2: Convert Internal Setup to External Setup:
   • Re-engineer the process so that operations previously requiring machine downtime are executed externally. Examples include: pre-heating molding dies off-line on an auxiliary hot-plate; utilizing standardized intermediate plates or pre-setter fixtures so tool alignment occurs on a setup bench while the machine runs; establishing fixed mechanical stops to eliminate in-machine trial-and-error measurement cuts.
3. Stage 3: Streamline All Remaining Operations:
   • Optimize all remaining internal and external tasks. Replace threaded bolts with quick-acting fasteners (pear-drop holes, U-slotted washers, quarter-turn cam clamps, pneumatic/hydraulic clamps); eliminate adjustments by standardizing die shut heights; implement two-technician parallel choreographies using standardized setup work combination sheets.

---

4. Cellular Manufacturing and Group Technology (GT)

Cellular Manufacturing is an application of Group Technology (GT), an engineering philosophy that identifies and exploits underlying physical and process similarities among parts.

Part Families and Composite Parts

• Part Family: A collection of parts that share similar geometric dimensions and shapes, or require similar manufacturing processing steps, machines, and tooling.
• Composite Part Concept: A hypothetical, all-inclusive master part containing every geometric design feature and operational step present across an entire part family. A manufacturing cell engineered to fabricate the composite part can manufacture any individual part in that family with minimal or zero changeover adjustment.

The Rank Order Clustering (ROC) Algorithm

Industrial engineers use King's Rank Order Clustering (ROC) algorithm to analyze a binary machine-part incidence matrix and reorganize disconnected equipment into dedicated manufacturing cells.

• Incidence Matrix Definition: An $m \times n$ binary matrix where rows represent machines ($i = 1, \dots, m$), columns represent parts ($j = 1, \dots, n$), and cell entry $a_{ij} = 1$ if part $j$ requires processing on machine $i$, and $a_{ij} = 0$ otherwise.
• Algorithmic Steps:
  1. Row Weighting: Assign binary weights $2^{n-j}$ to each column $j$ (from left to right, decreasing power of 2). Compute the decimal equivalent value for each row: $\text{Row Weight}i = \sum{j=1}^n a_{ij} \cdot 2^{n-j}$. Sort rows in descending order of their decimal weights.
  2. Column Weighting: Assign binary weights $2^{m-i}$ to each row $i$ (from top to bottom). Compute the decimal equivalent value for each column: $\text{Column Weight}j = \sum{i=1}^m a_{ij} \cdot 2^{m-i}$. Sort columns in descending order of their decimal weights.
  3. Convergence Check: Repeat alternating row and column sorting until the ordering no longer changes. The resulting matrix reveals distinct block-diagonal clusters, each defining a self-contained manufacturing cell.
  4. Exceptional Elements: Any isolated $1$s located outside the primary diagonal blocks indicate inter-cell transfers (parts that require processing on a machine outside their designated home cell), requiring duplicate tooling or routing redesign.

U-Shaped Manufacturing Cell Layouts

In cellular layouts, workstations are arranged in a U-shape rather than a linear straight line:

                  U-Shaped Manufacturing Cell Architecture
                  
                         Raw Material Infeed
                                  │
                                  ▼
                        ┌───────────────────┐
                        │  Machine 1 (Cut)  │
                        └─────────┬─────────┘
                                  │
      ┌───────────────────┐       │       ┌───────────────────┐
      │  Machine 6 (Pack) │       │       │ Machine 2 (Mill)  │
      └─────────▲─────────┘       │       └─────────┬─────────┘
                │                 │                 │
                │          Operator Patrol          │
                │               Path                │
                │                 │                 │
      ┌─────────┴─────────┐       │       ┌─────────▼─────────┐
      │  Machine 5 (Test) │       │       │ Machine 3 (Drill) │
      └─────────▲─────────┘       │       └─────────┬─────────┘
                │                 │                 │
                │       ┌─────────┴─────────┐       │
                └───────┤  Machine 4 (Debur)├───────┘
                        └───────────────────┘
                                  │
                                  ▼
                        Finished Goods Outflow

• Close Proximity of Start and End: The raw material entry point and finished goods exit point are adjacent, allowing a single operator to load raw stock and unload finished parts simultaneously.
• Minimized Travel Distance: Eliminates long walks down linear conveyor belts.
• Flexible Capacity Rebalancing: By cross-training operators, management can dynamically scale cell staffing based on demand: one multi-skilled operator can patrol all six machines during slow demand periods; two operators split the cell during medium demand; three operators man adjacent stations during peak demand, maintaining perfect takt synchronization.

---

5. Poka-Yoke (Mistake-Proofing)

Conceived by Shigeo Shingo, Poka-Yoke (Japanese for "avoiding inadvertent errors") is the design of devices, mechanisms, and fixtures that make it impossible for an operator to make a mistake or guarantee that defects cannot proceed downstream.

• Philosophical Foundation: Human beings are inherently fallible, but operational errors need not cause defective products. Quality must be built into the process rather than inspected into the finished lot.
• Classification by Mechanism:
  • Control / Shutdown Mechanism: The device physically stops the machine, blocks operation, or prevents part insertion if an abnormal condition occurs. Examples: asymmetric locating pins on an assembly jig that physically prevent inserting a printed circuit board upside down; an automated interlock switch that cuts power to a stamping press if the operator's hands breach an infrared light curtain.
  • Warning / Signaling Mechanism: The device detects an anomaly and triggers an alert (buzzing alarm, flashing strobe, chime), but does not mechanically stop the operation. Examples: a beeping alarm if a torque wrench fails to reach specified torque; a sensor flashing yellow if an operator reaches into the wrong parts bin.
• Regulatory Rule: Control mechanisms are far superior to warning mechanisms because they eliminate human judgment and enforce absolute zero-defect compliance.

---

6. Step-by-Step Worked Engineering Calculations

Worked Example 13.2.1: Kanban Card Sizing and WIP Analysis

Problem: An automotive subassembly line operates 16 hours per day (two 8-hour shifts). Demand for a specific brake bracket is 1,200 units per day. The replenishment lead time between the stamping department and the subassembly line averages 3.2 hours. Each standard tote holds 40 brackets. Plant engineering mandates a safety factor of 15% ($\alpha = 0.15$) to account for transit congestion and die change delays.

1. Calculate the hourly demand rate ($D$).
2. Determine the required number of Kanban containers ($k$).
3. Calculate the maximum allowable WIP inventory of brake brackets in this pull loop.
4. If an industrial engineering Kaizen event successfully applies SMED to reduce replenishment lead time to 1.2 hours, recalculate the required number of Kanbans and determine the percentage reduction in maximum WIP inventory.

Solution:

Step 1: Calculate Hourly Demand Rate $D = \frac{1,200\text{ units/day}}{16\text{ operating hours/day}} = 75.0\text{ units/hour}$

Step 2: Determine Baseline Kanban Containers ($k$) Using the Kanban sizing formula: $k = \left\lceil \frac{D \cdot L \cdot (1 + \alpha)}{C} \right\rceil$

• Lead time demand: $D \cdot L = 75.0\text{ units/hour} \times 3.2\text{ hours} = 240.0\text{ units}$
• Demand with safety buffer: $240.0 \times (1 + 0.15) = 240.0 \times 1.15 = 276.0\text{ units}$
• Divide by container capacity $C = 40$: $\frac{276.0}{40} = 6.90$
• Applying the ceiling function: $k_1 = \lceil 6.90 \rceil = 7\text{ Kanban containers}$

Step 3: Calculate Baseline Maximum WIP $WIP_{\max, 1} = k_1 \times C = 7 \times 40 = 280\text{ units}$

Step 4: Recalculate Post-SMED Kanban Sizing ($L = 1.2\text{ hours}$)

• New lead time demand with buffer: $75.0 \times 1.2 \times 1.15 = 90.0 \times 1.15 = 103.5\text{ units}$
• Divide by container capacity $C = 40$: $\frac{103.5}{40} = 2.5875$
• Applying the ceiling function: $k_2 = \lceil 2.5875 \rceil = 3\text{ Kanban containers}$
• New maximum WIP inventory: $WIP_{\max, 2} = 3 \times 40 = 120\text{ units}$
• Percentage reduction in WIP: $\Delta WIP = \frac{280 - 120}{280} \times 100\% = \frac{160}{280} \times 100\% = 57.14\%$
• Takeaway: Reducing lead time via SMED eliminated 4 circulating containers, driving a 57.1% reduction in shop-floor WIP.

---

Worked Example 13.2.2: Rank Order Clustering for Cellular Design

Problem: A machine shop manufactures four distinct components (P1, P2, P3, P4) across four machine tools (M1 = Lathe, M2 = Shaper, M3 = Drill, M4 = Mill). The initial machine-part incidence matrix is provided below:

$\begin{array}{c|cccc} & \text{P1} & \text{P2} & \text{P3} & \text{P4} \\ \hline \text{M1} & 1 & 0 & 1 & 0 \\ \text{M2} & 0 & 1 & 0 & 1 \\ \text{M3} & 1 & 0 & 1 & 0 \\ \text{M4} & 0 & 1 & 0 & 1 \end{array}$

Reorganize the matrix using Rank Order Clustering to identify distinct manufacturing cells.

Solution:

Step 1: Calculate Row Decimal Weights and Sort Rows Assign binary column weights $2^{4-j}$, meaning Column 1 weight $= 2^3 = 8$, Column 2 weight $= 2^2 = 4$, Column 3 weight $= 2^1 = 2$, Column 4 weight $= 2^0 = 1$.

• Row M1: $(1 \times 8) + (0 \times 4) + (1 \times 2) + (0 \times 1) = 8 + 2 = 10$
• Row M2: $(0 \times 8) + (1 \times 4) + (0 \times 2) + (1 \times 1) = 4 + 1 = 5$
• Row M3: $(1 \times 8) + (0 \times 4) + (1 \times 2) + (0 \times 1) = 8 + 2 = 10$
• Row M4: $(0 \times 8) + (1 \times 4) + (0 \times 2) + (1 \times 1) = 4 + 1 = 5$
• Row ordering: M1 (10), M3 (10), M2 (5), M4 (5). (Already partially grouped).

Step 2: Calculate Column Decimal Weights and Sort Columns Assign binary row weights $2^{4-i}$ based on the row sequence [M1, M3, M2, M4]: Row M1 weight $= 2^3 = 8$, Row M3 weight $= 2^2 = 4$, Row M2 weight $= 2^1 = 2$, Row M4 weight $= 2^0 = 1$.

• Column P1: $(1 \times 8) + (1 \times 4) + (0 \times 2) + (0 \times 1) = 12$
• Column P2: $(0 \times 8) + (0 \times 4) + (1 \times 2) + (1 \times 1) = 3$
• Column P3: $(1 \times 8) + (1 \times 4) + (0 \times 2) + (0 \times 1) = 12$
• Column P4: $(0 \times 8) + (0 \times 4) + (1 \times 2) + (1 \times 1) = 3$
• Column ordering: [P1, P3] (weight 12), [P2, P4] (weight 3).

Step 3: Construct the Block-Diagonal Matrix $\begin{array}{c|cc|cc} & \text{P1} & \text{P3} & \text{P2} & \text{P4} \\ \hline \text{M1} & 1 & 1 & 0 & 0 \\ \text{M3} & 1 & 1 & 0 & 0 \\ \hline \text{M2} & 0 & 0 & 1 & 1 \\ \text{M4} & 0 & 0 & 1 & 1 \end{array}$

• Result: The algorithm cleanly separates the system into two completely decoupled manufacturing cells with zero exceptional elements:
  • Cell 1: Machines {M1, M3} processing Part Family {P1, P3}.
  • Cell 2: Machines {M2, M4} processing Part Family {P2, P4}.

---

7. NCEES Reference Handbook Tips & Realistic Exam Traps

• The Ceiling Function in Kanban Sizing: Never round down or use standard arithmetic rounding on Kanban sizing problems! If the formula yields $k = 4.12$, rounding down to 4 containers will produce a systemic inventory shortage, causing line starvation during peak lead times. You must always round UP to $k = 5$.
• Internal vs. External Setup Traps: NCEES questions frequently describe an operator action during changeover and ask how to classify it under SMED. Remember: if the action CAN be executed safely while the machine is actively spinning, cutting, or stamping (e.g., retrieving tooling, pre-heating dies, staging coils), it is External. If and only if the spindle must be stopped to perform it (e.g., mounting bolts into the machine bed), it is Internal.
• Control vs. Warning Poka-Yoke: A device that only sounds a horn or illuminates a light is a warning mechanism. To be a control poka-yoke, the device must physically prevent the mistake or automatically shut down the line.
• Time Unit Consistency: In $k = \lceil D \cdot L \cdot (1 + \alpha) / C \rceil$, verify that the time dimensions of $D$ and $L$ match. If $D$ is in units per day and $L$ is in hours, divide $D$ by operating hours per day, or convert $L$ into days.