Lifecycle Model Survivorship Bias Index (LSBI)

GMI-MIN-020MiningInfrastructure-Dependent

Definition

Degree to which the AI lifecycle model training data underrepresents catastrophic failure outcomes relative to their actual occurrence rate.

The Measurement Gap

Lifecycle optimization models recommend replacement intervals and overhaul timing based on patterns learned from historical data. The critical vulnerability is that the training data itself may be incomplete. When catastrophic failures result in equipment being scrapped, written off to insurance, or managed through incident investigation processes separate from routine maintenance, those records often reside in different systems than the CMMS/EAM work order history that feeds model training. Legacy metrics such as mean time between failure and component life analysis use the data that is available, inheriting whatever gaps exist without detecting them. Availability and utilization metrics capture the consequences of catastrophic failures in production terms but do not feed back into lifecycle model calibration. The result is a model that appears to validate itself: it recommends longer intervals, fewer catastrophic failures appear in its evaluation dataset (because they were always underrepresented), and the model's confidence grows on a foundation of missing data. No conventional maintenance KPI quantifies this data completeness problem.

Decision Supported: Is the lifecycle model recommending longer intervals because it genuinely knows the components last longer at this site, or because it was trained on a biased dataset that excluded the worst outcomes?

Formula

Plain Language

Calculate the proportion of catastrophic failures in the model's training dataset and the proportion in the complete fleet history (including records from scrap logs, insurance claims, and incident reports that may not be in the CMMS/EAM). Divide the training rate by the actual rate and subtract from 1. A result of 0 means the training data perfectly represents catastrophic outcomes. A result approaching 1.0 means the training data is missing nearly all catastrophic events.

Notation

LSBI = 1 - (Catastrophic_rate_training / Catastrophic_rate_actual)

Variables

Variable Definition Unit
Catastrophic_rate_training Catastrophic failures in the training dataset divided by total removals in the training dataset Ratio
Catastrophic_rate_actual Catastrophic failures in the full fleet history divided by total removals in the full fleet history Ratio
Catastrophic failure A component removal where the component was scrapped due to in-service failure with secondary damage to adjacent systems or structures Binary classification
Training dataset The specific dataset used to train or retrain the lifecycle optimization model, as provided or documented by the vendor Dataset
Full fleet history Complete removal history across all record systems, including CMMS/EAM, scrap records, insurance claims, and incident investigation reports Dataset

Proposed formula, pending validation pilots.

At a Glance

Unit
Index (0 to 1)
Trend Direction
Lower; approaching 0
Reporting Frequency
Annually
Category
Predictive Model Integrity & Calibration
← Override Value Index - Geology (OVI-G)Premature Replacement Rate (PRR-MIN) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment