Research On Boiler Pipeline Leakage Fault Identification Based On VMD Sample Entropy Integrated Learning

The technology of acoustic emission detection system can realize the real-time detection of cracks and defects in the furnace wall of large boilers and furnace wall pipe components.Through the detection,it can achieve early maintenance and prevent unnecessary greater losses.The detection method of boiler pipe leakage based on acoustic emission is expected to provide a systematic solution.Due to the complex environment of the boiler working site and the complex and changeable noise sources that affect the measurement signal,it is the key to research and design the leakage detection system of the boiler pipeline and study the processing,identification and positioning methods of the leakage acoustic emission signal.The time-frequency analysis method represented by wavelet analysis and empirical mode decomposition(EMD)has been paid more and more attention to for the strong noise,non-linear and non-stationary features of acoustic emission signals.Various methods for extracting the effective features of complex signals and intelligent recognition and location methods represented by machine learning have been studied and applied in depth.In this paper,three kinds of acoustic emission signals,metal rod knocking,sandpaper and lead breaking,are obtained by simulation experiments on retired boilers.EEMD is used to decompose the experimental data,and approximate entropy is used as the eigenvector.BP and PNN are used for signal recognition,and the recognition rates based on EEMD approximate entropy and BP fault recognition and based on EEMD approximate entropy and PNN fault recognition are compared.Based on the characteristics of VMD adaptive multi-resolution analysis and sample entropy,which are sensitive to the change of physical process of acoustic emission signal,a fault feature information extraction method based on VMD adaptive multi-resolution and sample entropy is proposed.Using VMD to decompose data,comparing the overlapping phenomenon of EEMD and VMD to decompose signals,obtaining approximate entropy and sample entropy as eigenvectors respectively,applying BP and PNN to identify signals,comparing the recognition rate of BP faultrecognition based on VMD approximate entropy and PNN fault recognition based on VMD approximate entropy,comparing the recognition rate of BP fault recognition based on VMD sample entropy and PNN fault recognition based on VMD sample entropy.In view of the low accuracy and generalization ability of single BP neural network fault recognition,parallel BP and BP ? Ada Boost based leakage fault signal recognition methods are proposed.Compared with EEMD multi-resolution approximate entropy and VMD multi-resolution approximate entropy fault feature information methods.After the above experiments,it is found that the recognition rate of sample entropy eigenvector based on VMD decomposition is higher by using BP parallel network and BP Ada Boost.