Abstract

Purpose: The study critically interrogates the techno-economic and mathematical foundations of IoT-enabled predictive maintenance (PdM) in industrial manufacturing, questioning whether current data-driven reliability models genuinely optimize maintenance decisions or merely shift uncertainty into algorithmic opacity.

Methodology: A purely quantitative framework is developed by integrating stochastic degradation modelling, Remaining Useful Life (RUL) estimation, and multi-objective cost-reliability optimization. A simulated but methodologically valid industrial dataset is analysed using Weibull hazard functions, proportional hazards modelling, and deep learning-based prognostics. Model performance is evaluated through RMSE, precision-recall, availability, and lifecycle cost functions.

Findings: Results show that IoT-driven PdM improves system availability by 18.7% and reduces expected lifecycle maintenance cost by 23.4% compared with preventive maintenance. However, accuracy gains are non-linearly constrained by sensor data entropy, class imbalance, and degradation non-stationarity. The study demonstrates that hybrid physics-informed/data-driven models outperform purely data-driven architectures in RUL prediction stability.

Value: Rather than celebrating PdM as an Industry 4.0 inevitability, the paper exposes unresolved mathematical, architectural, and decision-theoretic contradictionsparticularly the tension between predictive accuracy, interpretability, and economic optimality. It provides a unified reliability-optimization model linking IoT data streams to maintenance policy selection.

Keywords: Predictive maintenance; Industrial IoT; Remaining useful life; Reliability engineering; Smart manufacturing; Cost optimization