Correcting Left‐Truncation Bias in Weibull Reliability Analysis: Synthetic Data Validation for Electrical Utility Assets

ABSTRACT

Incomplete historical records in utility asset datasets frequently introduce left‐truncation bias, leading to distorted reliability estimates and suboptimal asset management decisions. This study presents a systematic framework for validating left‐truncation correction in Weibull reliability analysis using synthetic data representative of electrical utility assets. A controlled dataset of 300 lifetimes (η = 60 years, β = 3.75) was progressively conditioned to reflect staggered entry, right censoring, and left truncation. Standard maximum likelihood estimation (MLE) produced substantial bias ( = 71.96 years, +19.9%), whereas conditional likelihood methods recovered accurate estimates ( = 60.26 years, +0.44%). A Monte Carlo study across 432 parameter combinations spanning decreasing, constant, and increasing hazard regimes confirms that truncation bias arises from the sampling mechanism and is primarily governed by the Weibull shape parameter β. Left‐truncation and right‐censoring (LTRC) correction removes 83% to 99.9% of material bias where the data can support estimation, with recovery accuracy governed by the observed‐failure count; under decreasing hazard the naive estimator exhibits runaway bias while LTRC remains accurate. Results are further validated using anonymised transformer datasets, highlighting material implications for regulatory replacement planning. The proposed framework provides a reproducible benchmark for evaluating reliability estimators and supports the routine adoption of truncation‐aware methods in utility asset management.