Research On Structure Optimization Of Twin Model For Fault Diagnosis Of High-pressure Diaphragm Pump In Slurry Pipelin

Slurry pipeline transportation,emerging as a progressive method for the conveyance of solid materials,is increasingly adopted owing to its attributes of environmental sustainability,heightened efficiency,and cost-effectiveness.The pivotal role of key motor components and high-pressure diaphragm pumps in ensuring the operational integrity and transport efficacy of slurry pipeline systems cannot be overstated.Within these systems,the one-way valve of high-pressure diaphragm pumps emerges as a critical yet susceptible component,where its consistent performance is integral to the overall safe and efficient operation of the pump.Failures in one-way valves stem from a spectrum of causes,ranging from physical wear and tear,such as abrasion,perforation,and valve jamming,to more complex factors like the rheological properties of the slurry being transported and varying production conditions.In light of ensuring operational safety,the prevalent practice of routinely replacing vulnerable components represents a cautious yet arguably excessive maintenance approach.This strategy,while aiming to mitigate risks,inadvertently leads to resource wastage and does not unequivocally forestall the occurrence of malfunctions.Hence,the monitoring of the operational state of diaphragm pump one-way valves assumes significant theoretical and practical relevance.This paper concentrates on surmounting challenges such as environmental noise interference and the impact of unknown multifarious disturbances on vibration signal integrity.Furthermore,it addresses the issue of imbalanced data samples regarding equipment state and endeavors to enhance the structural design of fault diagnosis models.The principal contributions of this study are delineated as follows:(1)Single features often inadequately represent fault states.To enhance the comprehensiveness of feature extraction from one-way valve vibration signals,a multifeature extraction method was employed.The one-way valve vibration signal is first divided into several non-overlapping samples.Then,16 time-domain features,13frequency-domain features,16 wavelet packet energy and energy entropy features of each sample are extracted to construct a multi-feature set to characterize the operation state of the one-way valve.The dimension of the 45-dimensional multi-feature data is reduced by Kernel Partial Least Squares(KPLS),eliminating redundant information from the signal.These features were then reduced using Kernel Partial Least Squares(KPLS)to eliminate redundant information.A fault diagnosis model based on Kernel Extreme Learning Machine(KELM)was developed,achieving a diagnostic accuracy of 96.88%.(2)The operating conditions of the high-pressure diaphragm pump’s one-way valve are complex,with vibration signals frequently exhibiting non-stationarity and nonlinearity.To extract non-linear dynamic parameters from vibration signals,the Multiscale Weighted Permutation Entropy(MWPE)was utilized.Based on MWPE,a Structure Optimization Regularized Twin Extreme Learning Machine(SO-RTELM)fault diagnosis model was introduced.This model demonstrated the ability to effectively diagnose oneway valve fault states with an accuracy rate of 97.833%.(3)A function norm-constrained model construction method was proposed to enhance the generalizability of the fault model.The Regularized Least Squares Twin Extreme Learning Machine(LS-RTELM)was introduced as a diagnostic model.To improve the optimization speed of the model,inequality constraints are converted to equality constraints.Based on TELM,the Least Squares Twin Extreme Learning Machine(LS-TELM)is proposed.This paper uses MWPE to extract features from vibration signals.The feature extraction method based on MWPE has better adaptability and flexibility.Experiments show that the proposed method can effectively extract vibration signal features.The fault diagnosis model can effectively diagnose the fault state of the one-way valve.Simultaneously,the model training speed is faster,and the fault diagnosis result time is greatly improved,with an average accuracy rate of 97.824%.(4)A fault diagnosis model using the Regularized Twin Random Vector Functional Link(RTRVFL)was proposed.The Random Vector Functional Link(RVFL)network model captures linear and non-linear features of input data better through direct connections,enhancing the model’s expressive power.Similar to TELM,the Twin Random Vector Functional Link(TRVFL)uses a random feature mapping mechanism to build a network.It then learns two non-parallel separation hyperplanes in the random feature space.By solving two smaller programming problems,two non-parallel hyperplanes are obtained.Regularization conditions are introduced based on standard TRVFL to construct the regularized RTRVFL model.Through the fractional Fourier transform,MWPE features under different order values P are extracted.The optimal P value vibration signal feature is selected experimentally for fault classification.The RTRVFL fault diagnosis model established based on Fractional Fourier Transform(FRFT)and MWPE improves the model’s generalization performance.Experiments show that by adjusting the different order values of FRFT p,the proposed method can effectively extract vibration signal features.The average accuracy rate of the one-way valve fault diagnosis model is 98.975%.In summary,monitoring the operational state and diagnosing faults in high-pressure diaphragm pump one-way valves is crucial for enhancing the efficiency,safety,and costeffectiveness of slurry pipeline transportation systems.Furthermore,the twin fault diagnosis research with structural constraints and optimization advances fault diagnosis technology in the metallurgical industry.