Spark plug fault recognition based on sensor fusion and classifier combination using Dempster–Shafer evidence theory

• Acoustic and vibration signals were used for fault detection of engine spark plug.
• Bior1.3 and Coif2 were selected as the best wavelets for de-noising the signals.
• The classification accuracy increased at least 20% using sensor fusion method.
• Sensor fusion with classifier combination improved the accuracy at least 31%.

A proper intelligent approach was developed for fault diagnosis of spark plug in an IC engine based on acoustic and vibration signals using sensor fusion and classifier combination. Wavelet de-nosing technique was used for removing the signal noises. ANN and LS-SVM were employed in classification stage. D–S evidence theory was applied to increase the fault detection accuracy. The results showed that the classification accuracies of ANN were 67.46% and 65.08% based on the acoustic and vibration signals. For LS-SVM, the classification accuracies of 65.08% and 57.94% were achieved based on the acoustic and vibration signals. By employing D–S theory, the classification accuracy reached a high level of 98.56%. The results indicated that the data fusion method improved significantly the performance of the intelligent approach in spark plug fault detection. The simultaneous use of acoustic and vibration signals increased the effectiveness of diagnostic system in engine condition monitoring. Moreover, the results demonstrated that the proposed procedure had great potential in spark plug fault recognition.

Figure optionsDownload as PowerPoint slide