Microgrids frequently experience a massive amount of faults, which compromise stable operation, disrupts the loads, and increases the grid recovery expenditures. The diagnosis of microgrid system faults is severely reliant on dimensionality reduction and requires complex data acquisition. To address these issues, machine learning-based methods are extensively implemented for fault diagnosis of microgrids providing robust features and handling a massive amount of data. However, the existing machine learning techniques use simplified models which are not capable of investigating diverse and implicit features and also are time-intensive. In this paper, a novel method based on a multiblock deep belief network (DBN) is suggested for fault diagnosis, underlying discrete wavelet transform (DWT), which allows the framework to discover the probabilistic reconstruction across its inputs. This approach equips a robust hierarchical generative model for exploiting features associated with faults, interprets highly variable functions, and needs lesser prior information. Moreover, the method instantaneously categorizes the fault modes, which eventually strengthens the adaptability of applying it to a variety of diagnostic problems in the microgrid domain. The proposed method is assessed using a substantial number of input signals at different sampling frequencies. A test model based on International Electrotechnical Commission (IEC) standard, was contemplated to assess the effectiveness of DBN. The system was also added with White Gaussian Noise (WGN) to verify the robustness of the proposed network. Results demonstrate the capability of the method for performing precise diagnosis operations.