• Home
• PHM Team
• Awards
• Research
• Contracts
• PHM Library
• Education
• Tutorials
• ROI Tool
• Members & Partners

Tutorial on Data-Driven Prognostics

Data-driven approach is one of the many approaches to implementing prognostics for a system. The data-driven prognostics methods use current and historical data to statistically and probabilistically derive decisions, estimates, and predictions about the health and reliability of products. DD approaches are useful to monitor the health of large multivariate systems and are capable of intelligently detecting and assessing correlated trends in the system dynamics to estimate the current and future health of the system. Areas of interest for data-driven approaches include anomaly detection, fault identification, fault isolation and prediction of remaining useful life (prognostics). Machine learning is highly used in the data-driven approach since it incorporates statistical and probability theory in addition to data preprocessing, dimensionality reduction by compression and transformations, feature extraction, and cleaning (de-noising) of data. The PHM group within CALCE conducted a one-day tutorial on data-driven techniques for prognostics and health management on March 18, 2009.

• Prognostic metrics
• Anomaly detection techniques
• Covariance estimation techniques
• Mahalanobis distance and symbolic time series techniques
• Multivariate state estimation technique and sequential probability ratio test
• Principal component analysis
• Support vector machines
• Self-organizing maps
• Neural networks

For more information contact Prof. Michael Pecht

Tutorial on Physics-of-Failure-Based Prognostics

Physics-of-failure (PoF) based approach for prognostics utilizes knowledge of a product's life cycle loading conditions, geometry, material properties, and failure mechanisms to estimate its remaining useful life. PoF methodology is based on the identification of potential failure mechanisms and failure sites for a device, product, or system. The methodology proactively assesses reliability by establishing a scientific basis for evaluating new materials, structures, and technologies. Failures can be broadly categorized by the nature of the loads—mechanical, thermal, electrical, radiation, or chemical—that triggers or accelerates the mechanism. Failure mechanisms are categorized as either overstress or wear-out mechanisms.

PoF-based prognostics permit the assessment and prediction of system reliability under its actual application conditions. Based on the stress level and severity determined by stress analysis, the architecture, the material properties, and the life-cycle profile of the products, reliability assessment is conducted by calculating the time to failure for dominant failure mechanisms at particular sites based on PoF models. The approach integrates sensor data with models that enable in situ identification of the deviation or degradation of a product from an expected normal operating condition (i.e., the system’s “health”) and the prediction of the future state of reliability.  
The PHM group within CALCE conducted a one-day tutorial on physics-of-failure-based prognostics on October 15, 2009. This event was free for CALCE Prognostics and Health management (PHM) consortium. CALCE Electronic Products and Systems (EPS) consortium members were given a 20% discount on the registration fee per person.
The tutorial covered the following topics:

For more information contact Sony Mathew

CALCE | University of Maryland. All Rights Reserved.