The Domino Effect in Unbalanced Lines

In modern operations management, there is a justified obsession with OEE (Overall Equipment Effectiveness). However, an approach focused purely on the final result often masks the underlying pathology that prevents reaching operational excellence: stochastic variance. In sequential production lines, the problem is not only the average speed of each station but the variability of their cycle times. When this variability is not managed, what we call the "Domino Effect" is triggered—a physical phenomenon that destroys Throughput in a silent but exponential way.

This technical article breaks down the physics behind this phenomenon and proposes a solution based not on more sensors, but on rigorous statistical inference.

1. Factory Physics: Understanding variance propagation

For a Plant Engineer, assuming that a theoretically balanced line (where the sum of capacities exceeds demand) will run smoothly is a common mistake. Operational reality is governed by statistical, not arithmetic, laws.

1.1. Technical definition of the Domino Effect in sequential environments

In a production line where processes are dependent (step B cannot start until step A finishes), fluctuations do not cancel each other out; they accumulate. If Station A is delayed by a micro-stop, Station B loses unrecoverable capacity. However, if Station A finishes early, Station B cannot capitalize on that gained time if it is still processing the previous unit. This is asymmetric variance propagation.

1.2. Starving and Blocking: The pathologies of flow

The domino effect manifests clinically in two states that degrade OEE:

• Starving: The downstream station stops because it receives no material (the previous process was slow or stopped).
• Blocking: The upstream station must stop because it cannot discharge its piece (the next process is busy or broken down).

To monitor these states in real time and quantify their impact on global profitability, production-control tools such as Induly are essential, since they make it possible to visualize when OEE drops due to these systemic interruptions.

1.3. Little's Law and queuing models (M/M/1)

From a systems-engineering perspective, this behavior is explained by queuing theory. Kingman's formula for a G/G/1 queue (or the M/M/1 simplification) shows that the waiting time in queue ($W_q$) grows exponentially with utilization ($\rho$) and variability ($c_a^2 + c_s^2$). A linear increase in process variability (due to human or technical factors) results in an exponential degradation of Throughput.

2. The fallacy of OEE calculated without causal diagnosis

2.1. Why IoT sensors fail at identifying the "Root Cause"

Industry 4.0 has filled us with sensors. We know when the machine stops, but we rarely know why with qualitative precision. A sensor may report "Machine Stop," but it cannot tell whether the operator stopped because of fatigue, lack of clear instructions, or because they were waiting for a shared tool.

2.2. OEE vs. Wrench Time

There is a critical gap between machine availability and operator activity. Wrench Time (actual tool-in-hand time) usually ranges between 25% and 35% in non-optimized plants. An OEE of 60% can coexist with very low Wrench Time if the operator spends much of their time on internal logistics and travel ("looking for things").

2.3. Limitations of invasive hardware

Installing cameras or wearables to detect human micro-stops is costly, invasive, and complex from a labor-union perspective. Moreover, an operator may be in front of the machine (detected by the sensor) but mentally idle or blocked by an administrative process. This is where hardware fails and statistics wins.

3. WorkSamp Methodology: Diagnosis through Statistical Inference

To overcome sensor blindness, we turn to Work Sampling, a scientifically validated technique that diagnoses productivity without technological friction.

3.1. The legacy of L.H.C. Tippett

Developed in the British textile industry in the 1930s, Tippett's technique proved that taking random samples of a system allows us to infer the overall behavior with predictable mathematical precision. It is the foundation on which the WorkSamp platform is built, digitizing this process for modern industry.

3.2. Snap Reading and the elimination of the Hawthorne Effect

Traditional time study (continuous timing) alters the object of study: the operator changes pace when they feel observed (Hawthorne Effect). The Snap Reading method (instantaneous, random observation) mitigates this bias. By capturing "snapshots" of the plant at random moments, we obtain a faithful picture of operational reality without conditioning the worker.

Note: For those cases where a short-cycle, high-repetition time-and-motion analysis (MTM) is required, tools like Cronometras are the ideal complement to sampling, enabling granular analysis of the work method.

3.3. MECE Taxonomy

For the data to be robust, observation categories must be MECE (Mutually Exclusive, Collectively Exhaustive):

1. Productive (Value Added): The machine/operator is transforming the product.
2. Waiting (Starving): Idle due to lack of input.
3. Blocking: Idle due to inability to output.
4. Auxiliary: Maintenance, cleaning, transport.

4. Mathematical Rigor: Sample Size (N) and Confidence (Z)

The validity of a WorkSamp study does not depend on the engineer's opinion but on the Law of Large Numbers.

4.1. Binomial Distribution and Gaussian Curve

Each observation is a Bernoulli event (Occurs/Does Not Occur). The sum of these events converges to a Normal Distribution. This allows us to precisely calculate how many observations we need for the data to be statistically significant.

4.2. The critical formula

To design a study in WorkSamp, we use the following equation to determine sample size ($N$):

$N = \frac{Z^2 \cdot p \cdot (1-p)}{E^2}$

Where:

• $Z$: Statistical value for the Confidence Level (1.96 for 95%; 2.58 for 99%).
• $p$: Estimated probability of event occurrence (e.g., historical inactivity %).
• $E$: Tolerable absolute margin of error (precision).

4.3. Interpreting results to Management

If the calculation shows that, with a 95% confidence level ($Z=1.96$), the "Blocking" time is 12% $\pm$ 1%, we have an irrefutable figure. It is no longer "I think the line clogs up"; it is "we have statistical certainty that we lose 12% of capacity to downstream bottlenecks."

5. Industrial Context: Spain 2025 and Industry 5.0

The regulatory and competitive environment is evolving toward more sustainable, human-centered models.

5.1. From 4.0 to 5.0: Operational efficiency as an ESG vector

In Industry 5.0, efficiency is not only economic but ecological. Reducing cycle time and eliminating waits reduces energy consumption per unit produced. Integral control platforms such as Induly help companies align their OEE with their corporate sustainability goals.

5.2. Regulatory compliance and privacy

Constant worker monitoring via cameras skirts the limits of GDPR and the EU AI Act. Anonymous statistical sampling is a "Privacy-by-Design" alternative, enabling process optimization without invading individual privacy.

5.3. Human-Centric Manufacturing

Putting the worker at the center means eliminating tasks that add no value (waits, unnecessary travel) so their effort translates into results. A Wrench Time diagnosis is the first step toward dignifying work on the plant floor.

6. Conclusions and Roadmap

Variance is the silent enemy of OEE. The "Domino Effect" amplifies small local inefficiencies, turning them into large global losses.

6.1. Empirical correlation

The data suggests that stabilizing the process (reducing variance) has a greater impact on OEE than simply trying to increase the machine's maximum speed.

6.2. Investment strategy

Before investing millions in rigid automation, run a scientific diagnosis.

1. Use WorkSamp to identify—through statistical sampling—where the real (not theoretical) Blocking and Starving occur.
2. Once the bottleneck is identified, use Cronometras to standardize the work method and reduce the cycle's standard deviation.
3. Deploy Induly to maintain continuous control and ensure improvements are sustained over time.

Modern plant engineering is not based on intuition—it is based on statistical certainty.