| As a critical equipment in dredging operations,the safety and reliability of mud pumps(high-power electric centrifugal pumps)directly affect production efficiency.Effective fault detection of mud pumps can save maintenance costs and improve their reliability and operability.This thesis aims to address the challenges of fault detection,diagnosis,and predictive maintenance of mud pumps in dredging operations,and investigates the methods of signal processing and machine learning based on vibration signals for fault detection,diagnosis,and predictive maintenance of dredging vessel mud pumps.Supported by the Ministry of Industry and Information Technology project "Development of Intelligent Operation System for Dredging Vessels," the main research contents of this thesis are as follows:(1)As a rotating machinery,mud pumps are analyzed in terms of their vibration principles and basic characteristics.The impeller,as the main rotating component of the mud pump and a part of the rotor,is taken as an example to analyze the basic vibration features of rotating machinery.The fault mechanisms and the time-domain and frequency-domain characteristics of vibration signals are analyzed for common faults such as impeller unbalance,misalignment,bearing faults,and other common mud pump faults.(2)In the frequency domain,the vibration signals of mud pumps often exhibit similarities when faults occur,and the(Fast)Fourier Transform(FFT)analysis is not effective in distinguishing different faults,sometimes even unable to differentiate between different fault types.In this thesis,a signal processing method based on Wavelet Packet Analysis is employed to decompose and reconstruct the vibration signals of common mud pump faults.The signal is decomposed into different scales,enabling the visualization of fault characteristics at selected scales and the extraction of fault feature vectors for more accurate fault diagnosis.Wavelet Packet Analysis can also serve as a preprocessing step for the Support Vector Machine(SVM)algorithm.(3)To address the issue of automated fault detection,an improved Support Vector Machine algorithm based on Bacteria Foraging Optimization(BFO)algorithm,a machine learning algorithm,is attempted.The BFO algorithm optimizes the penalty factor and kernel parameters of the SVM through its optimization capability,thereby improving the fault detection performance of SVM.Based on the signal processing using Wavelet Packet Analysis,a comparative analysis is conducted through simulation and case studies to validate the superiority of this method.The results indicate that the BFO algorithm has better optimization capability compared to traditional SVM algorithms,and the SVM improved by the BFO algorithm exhibits better fault detection performance.(4)Regarding the prediction of the remaining useful life of mud pump bearings,some machine learning and deep learning methods face challenges such as difficulties in feature engineering,high computational complexity,and interference from data noise.In this thesis,a rolling bearing remaining useful life prediction method combining the Informer model is proposed.The model explores the temporal relationship between degradation indicators and remaining useful life,and verifies the accuracy of the Informer model and its feasibility in predicting the remaining useful life of mud pump bearings. |
