Data-Driven Prognostics and Health Monitoring of Solder Joints
A Framework for Electronics Reliability
Published on 22. April 2026
Book
Paperback/Softback
125 pages
978-3-8396-2181-3 (ISBN)
Description
In today's fast-paced and highly interconnected world, digitalization is a significant driving force across various sectors, with electronic systems being the backbone of this. A record number of electronic devices are being manufactured which introduces significant reliability challenges. Faulty electronics can lead to catastrophic accidents if effective measures are not implemented. This dissertation provides a Prognostic and Health Monitoring (PHM) framework for solder joints as a use case.
The work introduces a tailored PHM framework that uses machine learning to predict creep-driven damage in solder joints. A validated finite element model generates synthetic temperature-load data from which accumulated creep strain serves as the primary damage indicator. A multilayer perceptron model learns the relationship between thermal half-cycle features and damage increments and thereby enables accurate creep strain prediction and subsequent remaining useful lifetime estimation. The approach is further trained and refined with real mission profile data from e-bike tests, incorporating field damage points to mitigate bias. As a next step, the methodology was demonstrated on resource-constrained edge hardware which reflects future implementation on real systems such as trams, automotives, and aircrafts.
The work introduces a tailored PHM framework that uses machine learning to predict creep-driven damage in solder joints. A validated finite element model generates synthetic temperature-load data from which accumulated creep strain serves as the primary damage indicator. A multilayer perceptron model learns the relationship between thermal half-cycle features and damage increments and thereby enables accurate creep strain prediction and subsequent remaining useful lifetime estimation. The approach is further trained and refined with real mission profile data from e-bike tests, incorporating field damage points to mitigate bias. As a next step, the methodology was demonstrated on resource-constrained edge hardware which reflects future implementation on real systems such as trams, automotives, and aircrafts.