Introduction: The Hidden Cost of Bottlenecks
Investigate capacity constraints using observations, flow, queues and comparable production data.
Updated:
A bottleneck limits system throughput under particular demand, mix and resource conditions. It is not necessarily the station with the most waiting, highest direct-work proportion or widest confidence interval. Sampling detects patterns; confirmation requires flow data.
Combine three evidence layers
- Activity: operation, setup, blocking, material starvation, failure and support shares.
- Flow: good units per period, cycles, queues, inventory and entry/departure sequence.
- Context: mix, staffing, changeovers, quality and supply availability.
A machine waiting for material may be affected by an upstream constraint. A busy resource may still have spare capacity and not constrain output. Investigate throughput changes when a condition changes while maintaining safety and quality.
Design observations
Define the cell and period; distinguish person and machine observations. Cover possible states with rules for support and breaks. Do not label a wait a supply failure before its cause is verified.
Distribute instants with known probabilities across shifts and workloads. A week may be an initial campaign, not a universal optimum. Randomisation does not guarantee independence: cell resources may share blocks.
The simple reference n=ceiling[z²p₀(1−p₀)/e²] has assumptions. With p₀=0.30, z=1.96 and e=0.05, plan 323 independent records. This does not establish representativeness or the ability to detect every improvement.
Simulated controller-cell case
Assume 280 good units per shift versus nominal capacity of 400: a 30% difference from nominal, pending confirmation that conditions match.
A campaign obtains 400 readings:
| State | Count | Exact proportion |
|---|---|---|
| Defined operation | 180 | 45% |
| Component waiting | 95 | 23.75% |
| Setup | 70 | 17.5% |
| Transport | 40 | 10% |
| Rework | 15 | 3.75% |
The proportions sum to 100%. The approximate normal 95% operation interval is 40.1–49.9%; waiting is approximately 19.6–27.9%, assuming independent records. The interval measures estimation uncertainty, not temporal process variability.
Waiting warrants investigation but establishes neither the bottleneck nor logistics responsibility. Check component shortages against arrivals, queues and production sequence. Setup improvements require frequency and duration data.
There is no universal optimum above 70% direct work, and these data cannot promise a 25% OEE improvement.
From hypothesis to trial
Test a bounded change: replenishment signals, external tool preparation or sequencing. Measure good output, waiting and quality before and after under comparable conditions; report uncertainty and other changes. An observed difference does not itself establish causality.
Do not add a buffer solely because proportions differ: review space, safety, inventory and the risk of hiding failures. OEE requires planned and running time, ideal cycle time and total/good units. Active-person proportion is not OEE.
Tool and questions
WorkSamp supports planning and state recording. Timing may complement it for specific operations.
Can a robotic cell be observed? Yes, with visibility and an authorised safe route; the method does not guarantee absence of hazards.
Does a wide interval identify a bottleneck? No: it may reflect a small sample, dependence or a proportion near 50%.
Is all transport waste? No. Assess necessity before proposing removal.