Research

End-of-Life Detection of Power Electronic Converters by Exploiting an Application-Level Health Precursor

Read the article

Abstract

Extensive studies have been conducted in regards to estimating the useful life of power electronic converters in order to prevent failure related downtime. The state-of-the-art of converter reliability is based on the physics-of-failure approach, which involves some concerns such as the uncertainty introduced by extrapolating the accelerated test results to other usage conditions, which results in failure probability distributions inhibiting a variance of several years. In order to overcome this issue, the state-of-the-art is currently seeing an increase in research output, which are in favour of basing the reliability studies on physics-of-degradation based methods. The existing methods are restricted to only consider a single component and due to the impracticability of monitoring each single component, there is an essential need for methods, which can monitor the health state of the entire converter. This paper proposes a method which uses the converter operating efficiency to monitor parameter shifts of multiple components, which can be used to gain precision when stating the converter lifetime. A supervised classifier is used to detect the end-of-life while simultaneously coping with the fundamental issue of the loading influencing the efficiency. The method proves to have high accuracy even when measuring inaccuracies are taken into account.

Power Converter Reliability Condition Monitoring Degradation End-of-life detection Health precursor Loading Maximal margin classifier Monitoring Power electronics Reliability Renewable energy sources Supervised learning Support vector machine Voltage supervised learning power electronics health precursor Condition monitoring end-of-life detection support vector machine maximal margin classifier