Abstract The identification of premature faults in Internal Combustion Engines has become determinant to guarantee suitable operation. Therefore, this study focuses on the implementation of fault diagnostic methodology by using advanced algorithms such as Back Propagation neural networks and Bayesian networks. Results indicated that the proposed methodology serves as a robust tool to identify different fault conditions in a wide operational spectrum with an reliability of nearly 73%. Moreover, the Backpropagation network diagnostic methodology presented an reliability of 18%, which is 3% higher than Bayesian networks. Overall, the implemented methodology counterbalanced interference conditions and noise signals while providing versatility to operate for different types of engines. In conclusion, this study can be extrapolated to different fields of physics to assist in identifying flaws in experimental test benches.