Multi-variable classification model for valve internal leakage based on acoustic emission time-frequency domain characteristics and random forest.

To use acoustic-emission technology to detect leaks inside valves, the necessary first step is to model the valve-internal-leakage acoustic-emission signal (VILAES) mathematically. A multi-variable classification model that relates the VILAES characteristics and the leakage rate under varying pressure is built by combining time-frequency domain characteristics and the random-forest method. A Butterworth bandpass filter is used to preprocess the VILAES from a liquid medium, and the best frequency band for filtering is determined as being 140 kHz-180 kHz. Then, (i) the standard deviation, (ii) root mean square, (iii) wavelet packet entropy, (iv) peak standard-deviation probability density, and (v) spectrum area are calculated as the VILAES characteristics, and six parameters-the pressure and the five VILAES characteristics-are used as the inputs for the random-forest classification model. Analysis shows that the five VILAES characteristics increase with an increase in the leakage rate. The multi-variable classification model is established by random forest to determine whether the valve leakage is small, medium, or large. The random forest uses many decision trees to predict the final result. For the same experimental data, the accuracy and operating time of the multi-variable classification model are compared with those of a support-vector-machine classification method for the bandpass and wavelet packet filtering preprocessing methods. The results show that the modeling method based on the combination of time-frequency characteristics and random forest has shorter operating time and higher accuracy.