What question did this study set out to answer?

The research aims to develop a hybrid methodology that integrates physics-based models with AI for efficient health management of complex systems under low data conditions.

February 12, 2026Open Access

A physics-informed AI methodology for Prognostics and Health Management of complex systems under low data constraints

Key Points

The research aims to develop a hybrid methodology that integrates physics-based models with AI for efficient health management of complex systems under low data conditions.
Utilized Bond Graph (BG) modelling for simulating system behavior and anomaly detection.
Constructed a Fault Signature Matrix (FSM) to classify faults using physics-based methods.
Employed Modified Hamming Distance for fault classification in the FSM.
Applied Dictionary Learning (DL) to refine fault diagnosis and forecast progression.
Collaborated with maintenance experts to embed domain knowledge into model design.
Successfully integrated human expertise in the AI-driven model, providing interpretable diagnostics.
Demonstrated effective fault classification and prognosis on a machine tool's linear axis.
Enhanced maintenance decision-making through the deployment of the developed methodology.

Abstract

Abstract Prognostics and Health Management has become a cornerstone of industrial efficiency, yet its deployment for complex systems in low-data environments remains limited. Data-driven approaches often fail due to data scarcity, while purely physics-based models are hindered by system complexity. Hybrid modelling offers a way to overcome these limitations. Moreover, the rise of Industry 4.0 and its focus on fully autonomous AI has not consistently delivered higher factory efficiency, underscoring the need to keep human expertise at the center of decision-making. In this context, a hybrid methodology is proposed that combines Bond Graph (BG) modelling with Dictionary Learning (DL) while embedding human knowledge into the process. BG is used to simulate system behaviour and detect anomalies through its Linear-Fractional-Transformation form, leading to the construction of a Fault Signature Matrix (FSM). The FSM is then analysed using the Modified Hamming Distance for physics-based fault classification, after which DL refines the diagnosis by isolating specific failures within each group and forecasting their progression. Designed in collaboration with maintenance experts, this model integrates domain knowledge into its physical layer and provides interpretable outputs to support justifiable maintenance decisions. The effectiveness of the approach is demonstrated on the linear axis of a machine tool, a critical component within a complex industrial system.

Ask AI

Helpful

Bookmark

View Full Paper