[Paper Review] A Review of Physics-Informed Machine Learning Methods with Applications to Condition Monitoring and Anomaly Detection

Yuandi Wu, Brett Sicard+1 more|arXiv (Cornell University)|Jan 22, 2024

Non-Destructive Testing Techniques6 citations

TL;DR

This survey compiles physics-informed machine learning (PIML) approaches for condition monitoring and anomaly detection, detailing how physical laws are embedded into ML models and outlining common architectures, benefits, and case studies.

ABSTRACT

This study presents a comprehensive overview of PIML techniques in the context of condition monitoring. The central concept driving PIML is the incorporation of known physical laws and constraints into machine learning algorithms, enabling them to learn from available data while remaining consistent with physical principles. Through fusing domain knowledge with data-driven learning, PIML methods offer enhanced accuracy and interpretability in comparison to purely data-driven approaches. In this comprehensive survey, detailed examinations are performed with regard to the methodology by which known physical principles are integrated within machine learning frameworks, as well as their suitability for specific tasks within condition monitoring. Incorporation of physical knowledge into the ML model may be realized in a variety of methods, with each having its unique advantages and drawbacks. The distinct advantages and limitations of each methodology for the integration of physics within data-driven models are detailed, considering factors such as computational efficiency, model interpretability, and generalizability to different systems in condition monitoring and fault detection. Several case studies and works of literature utilizing this emerging concept are presented to demonstrate the efficacy of PIML in condition monitoring applications. From the literature reviewed, the versatility and potential of PIML in condition monitoring may be demonstrated. Novel PIML methods offer an innovative solution for addressing the complexities of condition monitoring and associated challenges. This comprehensive survey helps form the foundation for future work in the field. As the technology continues to advance, PIML is expected to play a crucial role in enhancing maintenance strategies, system reliability, and overall operational efficiency in engineering systems.

Motivation & Objective

• Motivate the use of PIML to address data scarcity and the need for physically consistent predictions in engineering systems.
• Catalogue methodologies for embedding physical knowledge into ML models (features, regularization, architectures).
• Evaluate advantages and limitations of PIML methods in terms of accuracy, interpretability, computational efficiency, and generalizability.
• Highlight representative case studies demonstrating PIML efficacy in condition monitoring and anomaly detection.

Proposed method

• Classify PIML integration frameworks: physics embedded in feature space, data-enhanced refinement of physical models, physics-informed regularization, and physics-guided architecture design.
• Describe how physical laws are incorporated into model inputs, loss functions, and network architectures (e.g., PINNs, physics-based feature augmentation).
• Discuss synthetic data generation via physics-based simulations and semi-supervised or transfer learning to augment training data.
• Summarize literature through tables and figures outlining architectures and application domains.
• Present examples where finite element models and other physics tools generate training data or guide feature formation.
• Explain the trade-offs of each approach in terms of interpretability, computation, and generalizability.

Experimental results

Research questions

• RQ1 What are the main methodologies for integrating physics with data-driven models in condition monitoring and anomaly detection?
• RQ2 What are the advantages, limitations, and practical considerations (computational cost, interpretability, data requirements) of each PIML framework?
• RQ3 How do PIML methods perform across different engineering applications and fault scenarios?
• RQ4 What trends emerge in recent literature regarding architecture design and data synthesis for PIML?

Key findings

• PIML provides improved interpretability and robustness by enforcing physical principles in learning.
• Feature-space physics embedding and physics-guided architectures can reduce data requirements and improve generalization.
• Synthetic data generation using physics-based models and transfer learning enhances training when real labeled data are scarce.
• A variety of architectures (PINNs, physics-informed CNN/RNN/GNN, and physics-guided regularization) are used across condition monitoring tasks.
• Finite element models and other physics simulators frequently serve as data sources or guides for feature formation and model calibration.