This study aimed to diagnose healthy and misfire conditions in vehicle engines using vibration data recorded by a low-cost ADXL1002 accelerometer interfaced with a BeagleBone Black controller. Vibration signals were acquired from a vehicle engine using a cost-effective microelectromechanical system (MEMS) accelerometer. The data was analyzed using artificial neural network (ANN) and convolutional neural networks (CNN) models. The experimental results demonstrated that the two-dimensional (2D) CNN and 2D DCNN models significantly outperformed the ANN and one-dimensional (1D) CNN models in terms of prediction accuracy. Specifically, the validation accuracies were 90.12%, 92.42%, 96.52%, and 98.21% for ANN, 1D CNN, 2D CNN, and 2D DCNN, respectively. Furthermore, detailed accuracy analysis revealed that the 2D DCNN model achieved the highest prediction accuracy of 99% with healthy and 97% with misfire conditions. The findings indicate that transforming 1D vibration signals into 2D grayscale images enhances the model’s ability to distinguish between different engine conditions. Thus, employing 2D neural network architectures in conjunction with low-cost ADXL1002 accelerometer proves to be a highly effective approach for diagnosing complex systems such as vehicle engines.