From past few years, fault detection and diagnosis had been widely applied across various fields and this process involves in identifying and locating faults within a system. Traditional approaches for fault detection had faced several challenges which include difficulty in handling complex systems, reliance on manual intervention, and inability to adapt to dynamic operating conditions. Therefore, this research proposes Deep Variational Mode Decomposition (Deep VMD) for detecting the faults in industrial systems. Initially, the data is collected from NASA bearing dataset which consists of vibration signals which is captured by accelerometers placed at fixed positions on test bearings. Then, this data is preprocessed by using Wavelet denoising and Independent Component Analysis (ICA) which effectively removes high-quality noise. After that, features are extracted by using Stockwell Transform (ST) which effectively extracts features for fault detection in mechanical systems. Next, extracted features are selected by using Recursive Feature Elimination (RFE) which effectively selects the features by iteratively eliminating unwanted features. Finally, detection is done by using Deep VMD which effectively detected the faults in industrial systems. The proposed Deep VMD achieved better results in terms of accuracy (98.01%), precision (97.76%), recall (98.52%) when compared with existing Recurrent Neural Network (RNN).