Developing Tools and Techniques to Implement Prognostics and Health Management in Electronic
Products and Systems

Electronics are critical parts of modern engineered systems and turn out to be the largest cause of failures of these systems. Although industry recognizes the need for reliable products to gain and maintain market share, global competitive pressure, sourcing of parts from external sources, outsourcing of manufacturing, loss of faith in outdated and ineffective reliability standards, and uncertainties about the actual usage conditions have resulted in organizations taking short cuts in the reliability assessment and improvement steps. Meanwhile, unexpected failures of critical systems can have catastrophic consequences in terms of human life, property damage and economic ramifications. Current reliability design, assessment and validation methods for electronic products and systems lack the fidelity to prevent these failures because of the constant evolution in technological diversity and product changes at all levels in the supply chain where the product integrator does not have the time and resources to know about and evaluate the effects of changes in the sub component parts and raw materials.

Prognostics is the process of predicting the future reliability of a product by assessing the product’s extent of deviation or degradation from its expected normal operating conditions. Prognostics and health monitoring (PHM) provides advance warning of failure, prevents catastrophic failure, assesses reliability in a totally new manner, reduces unscheduled maintenance, increases availability, identifies faults efficiently, and improves both qualification methods and the design of future products. PHM enables users, maintainers and manufacturers to dynamically understand the state of health of a system or product and thereby helps them make informed and timely life cycle management decisions.

CALCE’s PHM group focuses on the application of PHM to complex electronic products and systems, as well as to systems-of-systems. The research is aimed at integrating CALCE’s expertise in reliability and physics of failure (PoF) of electronic products into data driven (feature extraction, data trending and mapping) models for prognostics. Through this process, we seek to learn from the prognostics data, detect changes in real-time, and predict the future performance of electronic systems. Our current PHM research focuses on computational algorithms, advanced sensors and data collection techniques, condition-based maintenance, prognostics and health management for the application of in-situ diagnostics and prognostics. We use physics based models along with empirical models for prognostics. CALCE is pioneering the use of the Fusion approach, which combines physics of failure and data driven methods for accurate prognostics and diagnostics. We are working in the areas of reliability modeling and prediction, pattern recognition, time series forecasting, machine learning, and fusion technologies. The prognostics group is evaluating the use of intelligent reasoning technologies to model and manage the life cycle of electronic products. In addition optimal maintenance planning and business case development to assess the return on investment associated with the application of PHM to systems are being performed at CALCE.

Physics of failures (PoF) based Approach

PoF based prognostics utilize knowledge of a product's life cycle loading conditions, geometry, material properties, and failure mechanisms to estimate its remaining useful life. Assessment of remaining life is a key part of PHM, and is based upon the science of characterizing the evolution rate of damage evolution. There are many dominant failure mechanisms that are commonly encountered in electronic systems, starting from mechanisms that dominate at the semiconductor device level to mechanisms that dominate at the system level. These can be broadly grouped in mechanical, thermal, electrical and chemical categories, depending on the nature of the primary driving stress. Many failure mechanisms span across multiple of these categories. The evolution of these mechanisms is not fully understood and CALCE is involved in systematic research to identify and characterize new mechanisms as well as new multi-scale, multi-physics material properties for existing mechanisms.

Data-Driven Approach

Data-driven prognostics use statistics and probability for analyzing current and historical data to estimate remaining useful life. The data-driven effort has so far been approached in two stages: a) anomaly detection and b) failure prediction. The current focus is on learning approaches with models, which rely on training data, and inference of health on statistical comparisons. A large part of the learning framework involves machine learning augmented with stochastic processes, time series analysis as well as other statistical techniques from reliability and survival analysis literature. Anomaly detection has been approached using classification, clustering, density estimation, statistical hypothesis testing and other state space driven approaches.

Fusion Approach

The Fusion approach pioneered by CALCE combines the PoF and the data-driven approaches to provide online diagnosis and estimation of remaining useful life of a product. The approach helps identify precursors to failure, provide information such as the failure mechanisms and site where the failure is precipitated in the product. Fusion PHM strongly motivates the use of models based on stochastic processes because they allow flexibility to incorporate time series information about not only covariates but also events that are characteristic to the process and in turn to the PoF modes and mechanisms of the product. The correlation between the precursor data and product failure helps provide an estimation of the RUL and signal product failure.

Return on Investment Tool

The CALCE PHM ROI (return-on-investment) tool is a stochastic discrete-event simulator that can follow the life history of a population of systems containing one or more line replaceable units (LRUs) and determine the effective life cycle costs, availability, and failures avoided for sockets (a socket is a unique instance of an installation location for an LRU). The PHM ROI tool is available for the CALCE PHM Consortium members and includes the following functionalities:

• Availability penalty model (modeled in two different ways: cumulative and non-cumulative model)
• Inventory On/Off switch–previous versions did not provide the option of considering (On) or ignoring (Off) the inventory model while running the simulation.
• Spares inventory (including initial spares, spare replenishment policy, and lead time on spares)
• ROI analysis of one PHM approach relative to another
• ROI plot including plotting negative values and displaying probability of negative ROIs.
• Conversion from static to stochastic ROI calculation
• Investment cost plotting
• Detailed implementation cost calculations (including recurring, non-recurring, and infrastructural costs)
• Operational profile specification
• Cost of money