Troubleshooting in vertical sieve tray using acoustic detection

In the chemical production process, real-time monitoring of the operating status of production equipment is essential to ensure the stability and safety of the production process. In this paper, a SECA-ResNet50D model that can be used to diagnose the operating condition of vertical sieve plate towers using acoustic signals is proposed. This method accurately predicts the operating conditions in a vertical screen plate tower, including five types of faults such as downcomer, entrained water flooding, gas phase remixing, and weeping. First, the various fault stages related auditory signals are obtained. Next, the duration, rate, and cepstrum areas are utilized to collect and merge the unique attributes linked to these acoustic signals, including the speech spectrogram, the Mel frequency cepstrum coefficient (MFCC), the short-time average energy (STAE), and the short- time over zero rate (STOZCR). Finally, the features of the processed signals are input to the SECA-ResNet50D model for five-state classification. From the experimental results, it can be seen that the prediction accuracy of the SECAResNet50D network is generally higher than that of the ResNet50D network, and the recognition accuracy based on the STOZCR + MFCC + Spectrogram features is the highest among all the features (99.67 %).