Accurately forecasting the energy consumption of industrial equipment and linking these forecasts to equipment health has become essential in modern manufacturing. This capability is crucial for advancing predictive maintenance strategies to reduce energy consumption and greenhouse gas emissions. In this study, we propose a hybrid model that combines Long Short-Term Memory (LSTM) for energy consumption prediction with a statistical change-point detection algorithm to identify significant shifts in consumption patterns. These shifts are then correlated with the equipment’s health status, providing a comprehensive overview of energy usage and potential failure points. In our case study, we began by evaluating the prediction model to confirm the performance of LSTM, comparing it with several machine learning models commonly used in the literature, such as Random Forest, Support Vector Machines (SVM), and GRU. After assessing different loss functions, the LSTM model achieved the strongest prediction accuracy, with an RMSE of 0.07, an MAE of 0.0188, and an R² of 92.7%. The second part of the model, which focuses on detecting change points in consumption patterns, was evaluated by testing several cost functions combined with binary segmentation and dynamic programming. Applied to a real-world case, it successfully detected a change point two months before equipment failure, offering the potential to reduce energy consumption by 27,052 kWh. This framework not only clarifies the relationship between equipment health and CO2 emissions but also provides actionable insights into emission reduction, contributing to both economic and environmental sustainability.

Achieving high fault-classification accuracy alone does not guarantee reliable autonomous operation under varying operating conditions, raising the need to assess prediction reliability and deployment readiness. This work proposes a resilience-oriented framework for fault diagnosis under varying operating conditions, characterizing diagnostic behavior under operating-condition shifts and providing complementary information on confidence, deployability, and supervision requirements. The framework fuses multi-sensor vibration and motor current signals within a Multi-Stage architecture combining a data-driven branch (DD-MSCNN) and a physics-aware branch (PA-MSCNN) integrating order-tracking descriptors, augmented by a confidence-aware deployability assessment layer. Evaluated on the Paderborn KAT dataset across six bidirectional shifts involving speed, torque, and radial force, the results reveal that operating-condition shifts are not equivalent and that their impact is strongly direction-dependent. Physical knowledge does not systematically guarantee superior performance, highlighting the complementary roles of the two representations. To formalize these observations, the Physics Contribution Index (PCI), the Shift Directionality Index (SDI), and a four-level deployability classification are introduced, providing quantitative insights into prediction reliability and autonomous operation readiness in dynamic industrial environments.

The transition toward Industry 5.0 (I5.0) is reshaping industrial systems around sustainability, Human-Centricity (HC), and resilience, yet most Predictive Maintenance (PdM) approaches remain focused on technical objectives inherited from Industry 4.0. To characterize the maturity of I5.0-aligned PdM research, a scoping review of the Scopus database was conducted, screening 238 articles and retaining 123 based on substantive-engagement criteria, followed by a weighted scoring protocol assessing the depth of engagement with the three I5.0 pillars. The review reveals a pronounced skew toward shallow integration, with only a minority of studies operationalizing sustainability, HC, and resilience jointly. Building on these findings, this paper proposes an implementation-oriented Predictive Maintenance 5.0 (PdM 5.0) framework structured around three interconnected layers. The framework introduces a structured Large Language Model-mediated human-in-the-loop interaction, an adaptive learning mechanism, and an energy and CO₂ estimation pathway linking equipment degradation to sustainability indicators. A preliminary industrial case study using operational data from a recycled-paper mill demonstrates the practical execution of three interconnected pathways of the framework: equipment health assessment, energy and CO₂ impact estimation, and human-in-the-loop interaction supported by LLM-mediated explanations. Layer-specific evaluation metrics are further proposed to support broader operationalization. This work contributes a scoping-review-grounded architecture that advances PdM from a purely technical function toward a holistic, I5.0-aligned paradigm, while providing initial empirical evidence of its operational feasibility and a structured basis for broader industrial validation.