Failure mode classification for grid-connected photovoltaic converters

Anomaly detection with machine learning techniques has been extensively used to assist the decision-making process during abnormal conditions. However, these approaches were majorly constrained in the fields of medical and image processing-based applications. This is because of the complexities due to the unavailability and uncertainty of the input data of real-world and mainly industrial applications. In light of the above observation, this chapter proposes a failure mode classification (FMC) for power electronics converters (PECs) that are associated with grid connected photovoltaic applications. Generally, the failure mechanisms are a critical aspect of determining the reliability of power electronic converters. The failure mechanisms deal with the physical, chemical, and electrical processes through which the failure occurs in the system. Based on the type of failure process, these failure mechanisms can be modeled when appropriate material and environmental information are available. Moreover, with the advancements in machine learning approaches, the failure data along with the modelled mechanisms can be used to identify the operating state of the PEC. This helps to monitor the operation of the system, performing risk analysis, estimating the reliability of the product, and reducing the probability that a customer is exposed to a potential product and or process problem. Further, the development of an FMC for a PEC can be discussed as follows: Initially, all the possible failure modes for PEC components can be identified. Further, the identified data is represented as a function of time and frequency domain to perform the feature extraction process. Further, the extracted feature data is minimized and subjected to classification using machine learning techniques. Finally, the trained machine learning classifier is implemented in a feed-forward loop to identify the operating state of a PEC.