Picid: A Modular Evaluation Infrastructure for Reproducible PHM Across Tasks and Domains

Researchers introduce Picid, a standardized evaluation infrastructure for Prognostics and Health Management (PHM) that addresses the reproducibility crisis in predictive maintenance across industries. The framework formalizes dataset construction, preprocessing, and evaluation metrics to enable fair comparisons of fault detection, diagnostics, and prognostics models across diverse domains like batteries, bearings, and engines.

The PHM field faces a critical methodological fragmentation where inconsistent evaluation protocols make it nearly impossible to compare research results or reproduce published findings. Picid addresses this infrastructure gap by creating a unified, executable protocol that abstracts away ad hoc choices around data splitting, temporal windowing, label alignment, and metric selection—areas where subtle implementation differences can dramatically affect reported performance. This standardization problem mirrors challenges in machine learning broadly, where reproducibility failures have undermined research progress and inflated performance claims.

The framework's design enables deterministic, leakage-free dataset construction while maintaining flexibility across heterogeneous PHM applications. By enforcing consistent data contracts and evaluation boundaries, Picid allows researchers to conduct fair cross-domain comparisons between classification-based diagnostics and regression-based prognostics using identical model families. The empirical validation across thirteen models and twelve datasets spanning batteries, bearings, turbofan engines, hydraulics, filtration systems, and buildings demonstrates practical applicability across industrial domains.

For the predictive maintenance industry, this work reduces friction in model development and deployment validation. Engineers can now reference standardized benchmarks rather than implementing custom evaluation pipelines, accelerating adoption of advanced PHM techniques in manufacturing and infrastructure monitoring. The infrastructure becomes particularly valuable as organizations invest in condition-based maintenance to reduce downtime and operational costs. However, widespread adoption requires community engagement and integration into existing research workflows.

• →Picid standardizes PHM evaluation protocols to address reproducibility and comparison challenges across datasets and domains.
• →The framework enforces deterministic dataset construction and prevents data leakage while remaining flexible across fault detection, diagnostics, and prognostics tasks.
• →Cross-domain comparisons between classification and regression models become possible through unified evaluation boundaries and consistent data contracts.
• →Empirical validation covers thirteen models across twelve industrial datasets including batteries, bearings, engines, and hydraulic systems.
• →Standardized infrastructure reduces implementation friction and enables fair benchmarking for predictive maintenance research and deployment.