Lifetime and service life forecast

Electronic components used in safety-relevant systems must meet particularly high requirements in terms of reliability. In addition to fixed test or replacement intervals or the use of redundant systems, it can be worthwhile to rely on a failure prognosis. This means that the service life already consumed by the system is calculated based on condition monitoring. By comparing this with a service life model that applies to the system, it is then possible to forecast the future from the historical data on service life consumption. The result is the so-called residual ultimate lifetime (RUL) or remaining useful life.

Classic service life models for electronics are based on the fatigue of the solder contacts. The increase in creep strain due to the thermal mismatch between substrate and component represents the damage increment. This cannot be measured directly but is calculated numerically using a virtual twin. A service life prediction on this basis is very computationally intensive, so it must be carried out on central servers. This also requires a communication infrastructure. Artificial intelligence (AI) and machine learning (ML) approaches, on the other hand, enable a much more efficient evaluation of existing load scenarios. Fraunhofer IKTS was able to show that a microcontroller (MCU), such as an Arduino Nano or an STM32 Nucleo, has sufficient resources to process the chain from condition detection using suitable sensors, data pre-processing and evaluation of a pre-trained ML approach through to the output of results. This means that the RUL can be determined directly on the system to be monitored. There is no need for a communication chain.