Water pump cavitation fault early detection method based on acoustic emission signal and lightweight CNN

Abstract

The invention provides a water pump cavitation fault early detection method based on an acoustic emission signal and a lightweight CNN, belongs to the field of water pump fault detection, realizes early warning of cavitation in advance greater than or equal to 30 min, detection accuracy greater than or equal to 92%, misjudgment rate less than or equal to 5%, adapts to edge calculation deployment, and is suitable for real-time monitoring of industrial water pumps. The method comprises the following steps: (1) constructing an acoustic emission signal acquisition system, and synchronously acquiring water pump acoustic emission signals and working condition data such as inlet and outlet pressure and flow; (2) preprocessing the collected signals, eliminating interference and optimizing data quality; (3) extracting acoustic emission time-frequency characteristics at the early stage of cavitation through short-time Fourier transform, and converting the acoustic emission time-frequency characteristics into a two-dimensional grey-scale map; (4) identifying the features by using a CNN model with lightweight parameters, and outputting three state labels of normal, early cavitation and serious cavitation; (5) adaptively calibrating the judgment threshold based on the current working condition, and reducing the misjudgment rate; and (6) outputting graded early warning information and a processing instruction according to a judgment result.

Images