Symmetrized Dot Patterns and CNN-Based Acoustic Signal Analysis for Fault Diagnosis in Internal Combustion Engines.

Early detection of faults in internal combustion engines (ICEs) remains an important challenge for improving operational reliability and supporting condition assessment. Conventional approaches based on On-Board Diagnostics II OBD-II data, vibration, or exhaust gas analysis face practical limitations, motivating the exploration of non-invasive acoustic diagnostics. This study proposes and validates a pipeline that combines acoustic signals, Symmetrized Dot Pattern (SDP) transformations, and convolutional neural networks (CNNs) to identify injector and ignition failures in multicylinder ICEs. Acoustic measurements were acquired in a purpose-built semi-anechoic chamber using a smartphone equipped with Micro-Electro-Mechanical Systems (MEMS) microphones, ensuring high-signal-to-noise-ratio recordings under controlled fault conditions. A total of 540 independent acoustic recordings were transformed into 540 grayscale SDP images, which were used to train and test a custom CNN architecture. Results demonstrated an overall accuracy of 81.11% in independent verification, with outstanding performance in Spark Plug 3 and Spark Plug 4 (F1 = 1.000), as well as Injector 4 and normal operation (F1 = 0.947). More challenging cases included Injector 1, Injector 3, and Spark Plug 1 (F1 ≈ 0.70-0.78), where spectral overlap led to cross-misclassifications. These findings are consistent with recent studies reporting the robustness of SDP for representing transient acoustic phenomena and the superior classification capability of CNNs for pattern recognition tasks. The study extends SDP applications to internal combustion engines, demonstrating the feasibility of detecting injection and ignition faults using acoustic measurements. The proposed SDP-CNN pipeline represents a non-invasive, low-cost approach to acoustic fault analysis using consumer-grade sensing devices. Future research should focus on expanding datasets, incorporating domain adaptation techniques, and validating performance under real-world operating conditions to improve generalization and support practical acoustic monitoring applications.