The evolving automotive industry with the rapid growth of using fuel resources causes a shortfall of fuel availability. The engine maintenance has great importance and it is essential to develop a fault detection system and condition monitoring is done to reduce the damage-causing circumstances to improve engine safety. The engine tested at the rated speed of 1800 rpm by varying the loads. Captured vibration data is analyzed by using feature extraction and classification algorithms to obtain the best performance and minimum vibrations of blends. The best classification accuracy blend results are selected and used to develop a neural network model and to predict the vibration signatures. The neural network model is developed with a feedforward backpropagation algorithm and using Levenberg–Marquardt as training function. The developed ANN model is trained and vibration signatures are predicted for the selected blend B25 with the classification accuracy of 97%. © Springer Nature Singapore Pte Ltd. 2020.