Root Cause Analysis with Hierarchical Taxonomy

The Failure of Flat Data Architectures in Industrial Failure Analysis

Category Overlap and Invalidity in the Binomial Distribution

The state of the art in reliability engineering reveals that approximately 68% of industrial plants present systemic failures in their failure analysis due to a flat and ambiguous data architecture. When an analyst or a system tries to measure inefficiencies, the use of generic and interconnected categories causes fatal statistical overlaps.

Mathematically, if the inactivity categories overlap, the calculation of the underlying probabilities in a Binomial Distribution (where $p$ is the probability of occurrence and $q=1-p$ is its counterpart) loses its validity. For the variance of a sample, defined by the equation $\frac{p(1-p)}{N}$, to be statistically sound, the event must be binary and unambiguous at the instant of observation.

The Bias of Invasive Hardware and the Emergence of the Hawthorne Effect

In an attempt to solve this lack of precision, the industry has massively turned to invasive telemetry: biometrics, algorithmic surveillance cameras, and constant IoT sensors. As W. Edwards Deming warned, the observer invariably impacts the system.

By subjecting operators to suffocating monitoring, the Hawthorne Effect is triggered, a psychological bias where the worker artificially alters their behavior upon knowing they are continuously observed. This introduces a "special variation" into the system, corrupting the captured data, which stops representing the plant's natural stochastic behavior and only reflects operational stress under surveillance.

Foundations of Root Cause Analysis (RCA) Based on MECE Taxonomy

Transition from Macro Categories to Hierarchical Trees of Sublevels

To overcome flat architectures, it is imperative to structure data under the MECE principle (Mutually Exclusive, Collectively Exhaustive). This implies the transition from macro categories (e.g., "Machine Down") to precise hierarchical trees of sublevels. If an operator does not contribute value, the taxonomy must systematically break down whether the cause is intralogistics, maintenance, quality, or organization, branching the event down to its ultimate origin.

Principle of Exclusivity and Mathematical Rigor in Reliable Systems (ISO 14224)

Based on the data collection guidelines of the ISO 14224 standard for industrial reliability, a MECE taxonomy ensures the Principle of Exclusivity. This guarantees that the sum of the probabilities of all inactivity categories of a system is exactly the unit ($\sum p_i = 1$). If this axiom is broken because an observation can fit into two categories simultaneously, probabilistic inference about failure causes collapses abruptly.

Extrapolation of Observations toward the Gauss Curve without Inflating the Margin of Error

If the taxonomy is not strictly MECE, classifying the same event into multiple root causes will artificially inflate the margin of error. By guaranteeing hierarchical exclusivity, random observations remain pure. This allows accurate extrapolation of the sample data toward the population's Gauss Curve, ensuring that management decisions are made on the plant's statistical reality and not on a mathematical artifact.

WorkSamp and Statistical Inference as an Alternative to Telemetry

Test Load Equation: Calculation of Sample Size (N) and Confidence Level (Z)

To execute a non-invasive diagnosis with empirical rigor, the methodology of WorkSamp redesigns Work Sampling under strict foundations. The validity of the entire study depends on calculating the required Sample Size ($N$). The fundamental inference equation used is:

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

Where:

• $Z$: Desired confidence level referencing the Gauss Curve. Typically 1.96 is used for 95%, although for hypercritical processes we recommend validation of the 99% confidence level in studies ($Z=2.576$).
• $p$: Probability of occurrence of the activity (estimated by pre-sampling).
• $e$: Tolerated margin of error (generally $\pm 3\%$ to $\pm 5\%$).

The Snap Reading Method: Random Observations via the Tippett Technique

Founded by L.H.C. Tippett in 1934, the Ratio Delay technique demonstrated that structured sampling is asymptotically equivalent to continuous timekeeping. WorkSamp digitizes this approach through the Snap Reading method.

The analyst does not remain static watching the operator, but captures random statistical instants. Being ephemeral and unpredictable observations, the Hawthorne Effect is completely neutralized.

Reconstructing Availability and Performance: How to Calculate an OEE without Sensors

By classifying each rapid observation in the MECE tree, we can reconstruct Availability and Performance, the pillars of Overall Equipment Effectiveness (OEE). If in a statistically valid sample ($N=2000$), 20% of the pure observations fall into "Mechanical Failure", intrinsic availability has been diminished by 20% with a known margin of error.

This empirical approach is vital as a baseline diagnosis. In fact, many plants use it as a preliminary phase before implementing definitive software. It is extremely useful to perform a SCADA systems audit through manual WorkSamp observations to validate whether the software is measuring correctly, before consolidating production in a continuous system like Induly, which handles production control and OEE calculation in real time for daily operations.

Regulatory Impact in Spain 2025: The Urgency of a Non-Invasive Diagnosis

Adaptation to the 37.5-Hour Workday: The Criticality of Measuring Wrench Time

The regulatory reduction of the working day in Spain to 37.5 hours in 2025 represents a critical contraction of the available time in the plant. Competitive survival requires radically optimizing Wrench Time ("tool in hand" time or net value contribution). Increasing Wrench Time is the only way to absorb the hour reduction without triggering wage costs from overtime. For very specific operational descents after this macro-diagnosis, the use of video time and motion analyzers such as Cronometras is an ideal complement.

Compliance with the AEPD and Privacy Directives in Industry 5.0

The legal siege on workplace privacy is increasingly tight. The Spanish Data Protection Agency (AEPD) and Industry 5.0 directives explicitly prohibit the disproportionate use of invasive individual monitoring technologies. This is where Work Sampling shines: it is anonymous by design, evaluates stochastic processes instead of specific individuals, and supplements the need for constant telemetry with high-level mathematical inference, mitigating risks of union conflict.

Industry and Sustainability Law: Auditable Justification of Operational Inefficiencies

The new Industry and Sustainability Law requires plants to justify their inefficiencies and outline decarbonization and environmental impact plans. A Root Cause Analysis with statistical and taxonomic rigor is consolidated as the only auditable protocol capable of demonstrating due diligence in the technical allocation of operational resources.

Empirical Evidence: Practical Case of Structured Work Sampling

Below, we present the raw data derived from a technical investigation with WorkSamp in a discrete manufacturing environment, demonstrating the power of taxonomy.

Statistical Parameters of the Study (Z=1.96, Margin of Error $\pm$ 3.5%)

• Z (Confidence Level): 1.96 (95% Distribution).
• Tolerable Margin of Error ($e$): $\pm 3.5\%$
• N (Valid Random Observations): 1,250
  (Technical Note: To monitor the evolution of these ratios over productive time, we recommend subsequent use of P-Charts (P-Charts Control Charts) for analysts).

Breakdown of Raw Data in a 3-Level Hierarchical Taxonomy

Level 1 (Macro State)	Level 2 (Operational Classification)	Level 3 (Root Cause Analysis)	Absolute Frequency	Proportion ($p$)
Value Added	Direct Wrench Time	Operation in progress	562	45.0%
Value Added	Support to Wrench Time	Tooling preparation	113	9.0%
Loss (Inactive)	Supply Failure	Lack of material on line	188	15.0%
Loss (Inactive)	Supply Failure	Forklift / AGV wait	87	7.0%
Loss (Inactive)	Failure Analysis	Mechanical Failure - Main Shaft	125	10.0%
Loss (Inactive)	Failure Analysis	Electrical / Panel Failure	63	5.0%
Loss (Inactive)	Personal Delays	Standardized Fatigue / Rest	112	9.0%
TOTAL	-	-	1,250	100.0%

Operational Inference: Identifying the True Root of Losses (Intralogistics vs. Maintenance)

A look at the data reveals a critical finding. In a plant without MECE taxonomy, the flat report would have yielded 37% of technical inactivity ("Operational Downtime"). Any Plant Engineer's instinct would have been to audit the Maintenance department.

However, the Level 3 breakdown mathematically proves that maintenance inactivity barely sums 15% (Mechanical Failure + Electrical Failure). The real black hole of Wrench Time is in intralogistics, summing 22% (15% Lack of material + 7% AGV Wait). The statistical diagnosis successfully diverted a multi-million investment in machinery reliability toward the simple reprogramming of forklift routes.

The New Standard for Operations Directors and Plant Engineers

The Synergy Between Mathematical Rigor, Workplace Privacy, and Empirical Efficiency

Modern Operations Management cannot rely on intuition or intrusive surveillance systems that deteriorate the work climate and violate 2025 regulations. Work Sampling supported by an expert application like WorkSamp offers the squaring of the circle.

Implementing this standard based on a MECE hierarchical taxonomy and the Snap Reading technique provides Plant Engineers with the definitive tool: the ability to perform an auditable RCA, map the reality of Wrench Time with high fidelity, and calculate an OEE without sensors. It is about aligning, once and for all, mathematical excellence with legal and operational industrial viability.