There is an increasing interest both in academic or industry for health monitoring of photovoltaic (PV) power plants. The main reasons are safety issues and the loss of income due to faults or failures. In a PV power plant, on the DC side, the fault can affect a single cell, a module or a string. The fault effect or signature can be detectable or not, depending on the available information, the fault severity and the fault diagnosis method. From the abundant literature, there is a diversity of approaches based on different input data (array I-V characteristic , array or string maximum power point, module level power point, infrared images, etc.. .), different techniques (image processing, neural network, etc.. .) depending on fault types (mismatch, short-circuit, open-circuit, etc.. .). From the application point of view, it is not obvious to identify what would be the most efficient method to implement a condition-based maintenance that is now recognised as the most cost effective method. Therefore, we propose in this work from the analysis of the publications in 2017 to classify the fault detection and diagnosis methods through a framework defined in 4 steps: modelling, pre-processing, features extraction and features analysis.