Industries are looking for advanced maintenance techniques to reduce the cost of maintenance and increase the productivity of plants. Fault diagnostics is the focus area to reduce the downtime of the machine and ensure its durability. Bearing condition monitoring is important because bearings are the key component of rotary machines. Bearing failure detection relies heavily on vibration signal analysis. A defect in a rolling element causes the bearing to generate an impulsively responsive signal. This work contributes to the development of the framework to generate and acquire bearing vibration data. The data is collected for healthy bearing, bearings with cage damage, and ball damage. It generates a unique pattern of vibration signals for each bearing defect at constant, increasing, increasing and decreasing, and decreasing and increasing speeds. The data is post-processed using deep learning techniques capable of diagnosing and categorizing the different failure conditions. The paper proposes feature extraction and deep learning algorithms for diagnosing bearing faults. Feature extraction of signals is the traditional method of fault diagnosis, which involves expert knowledge and time. The 1D and 2D convolutional neural network (CNN) algorithms give an accuracy of 99%, which is far better than the feature extraction and artificial neural network (ANN) techniques, with an accuracy of 55%. Such techniques will help in the adaption of smart manufacturing in India.