Prognostic health monitoring technologies, which can evaluate performance degradation, load history, and degree of fatigue, have the potential to improve the effective maintenance, reliability design methods and availability under improper use conditions of electronics systems. We propose a method to assess the thermal fatigue life and thermal load history under asymmetric cycles of a power electronics module by using finite element method-based Lagrangian neural network. We apply this method to an insulated-gate bipolar transistor power module with a temperature history monitoring function.Power modules are subject to temperature variations resulting from power switching, load factor variations and ambient temperature changes. Because of differences in the thermal expansion coefficients of the various materials in power modules, thermal fatigue failure may occur under thermal cycling. Reliability prediction for thermal fatigue life under field conditions is a key issue for improving reliability. Utilizing field load and health-monitoring information for the case of performance degradation during use, maintenance, and field failure of the product in order to reduce the associated downtime is also important. Whether the load history waveform is symmetric or asymmetric affects the fatigue life of solder joints. In this paper, we propose a novel health-monitoring method for thermal fatigue failure corresponding to time-dependent inelastic strain response such as asymmetric cycles as described following.&nbsp;Step 1: Formulate the physics-of-failure model to estimate physical variables related to failure from monitoring variables through a surrogate model such as Lagrangian neural network model.Step 2: Identify the surrogate model by generating the learning data based on physical phenomena simulation such as FEM analyses, and measurement of loading experiments.Step 3: Assess the load history and the damage and failure based on the physics-of-failure model through the surrogate model.&nbsp;The surrogate model can transform the monitoring and structure variables, boundary conditions, and material properties to physical variables of damage and failure, for example, by using a novel finite element method-based Lagrangian neural networks, in which a Lagrangian neural network is applied to FEM analysis. The physics-of-failure model, such as the fatigue damage law, can assess the damage and failure based on the estimated physical variables of failure.The assessment model for estimating the fatigue damage through physical variables from monitoring variables is formulated by using both measurement of loading experiments and physical phenomena simulations. The grain boundary sliding model for assessing creep-fatigue life was proposed by Sakane et al. for fatigue damage assessment in the case of asymmetric load waveforms. We introduced this model to our health-monitoring method. The grain boundary sliding model focuses on the creep-fatigue damage of solder materials corresponding to the amount of grain boundary sliding, and failure occurs when the cumulative damage value reaches the limit.It is shown that this method can estimate the inelastic strain cycles and thermal fatigue life distribution of the solder joints from temperature monitoring history. It is also confirmed that probabilistic load assessment of thermal cyclic load distributions can be conducted using the estimated inelastic strain cycles. It is concluded that the proposed method could be effective for assessing thermal load history and thermal fatigue life estimation in prognostic health monitoring.

Prognostic Health Monitoring Method for Thermal Fatigue Failure of Power Modules Based on Finite Element Method-Based Lagr Health Monitoring Method for Thermal Fatigue Failure of Power Module Based on Finite Element Method-Based Lagrangian Neural Networks

Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2021/volume/85543

ASME 2021 International Mechanical Engineering Congress and Exposition (IMECE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.