Research On Intelligent Fault Diagnosis Of Slurry Pump Based On Joint Denoising And Deep Learning Methods

Slurry pump is the key media conveying equipment in coal processing plant.Attention should be paid to the health status of slurry pumps to ensure the normal operation of coal processing plant.Solid-liquid media is usually transported by slurry pumps.Many solid particles are present in the media.What’s more,slurry pumps are generally operated under heavy loads.Under the coupling effect of two factors,failures occur frequently in the pumps,resulting in serious economic losses.A slurry pump state identification method should be proposed to ensure the safe and stable production of coal processing plant.Based on the method,faults can be detected timely and guidance can be provided for equipment engineer to develop maintenance plan.Rolling bearing and impeller are parts of the slurry pump bearing-rotor system.Once faults occur,fault characteristics will be reflected in the vibration signal collected from the bearing box.Therefore,faults can be diagnosed via detecting the vibration signals.Considering the presence of strong noise in the coal processing plant,fault diagnosis research of slurry pumps should be carried out from two aspects.On the one hand,vibration signal denoising method and bearing-rotor compound fault diagnosis method were studied based on noisy dataset collected in the laboratory.On the other hand,based on above methods,fault diagnosis method of 300ZJ-I-A70 slurry pump was proposed to meet the equipment diagnosis requirement of Hongliu coal processing plant.The main research results are as follows:Firstly,six vibration signal conversion methods were used to convert onedimensional vibration signals into two-dimensional images.Based on the noisy twodimensional datasets,Res Net and Swin Transformer were used to conduct rolling bearing fault diagnosis research.Meanwhile,a comparison with the 1D convolutional neural network was conducted.The effect of noise on deep learning fault diagnosis methods was clarified by the experiments.What’s more,fault classification accuracies were determined based on different deep learning methods and noisy datasets with different SNR.According to the experimental results,the fault diagnosis method based on Short-time Fourier Transform and Res Net was determined as the optimal benchmark model with 88% average classification accuracy.Secondly,for influence of noise on the deep learning diagnosis method,a joint vibration signal denoising method based on SVD-CEEMDAN-WPT was proposed.Based on the denoising method,gaussian noise and residual burr noise can be effectively suppress.What’s more,the Res Net network was improved.An intelligent fault diagnosis method was proposed based on the proposed joint denoising method and Conv Ne Xt.Compared with the benchmark model,the classification accuracy was improved by 3.08% on average.For the presence of random noise,varying working conditions and compound faults,three generalization experiments were carried out to verify the generalization performance of the proposed method under the aforementioned three conditions.The experiments showed that the accuracies of the proposed method reached 96.31%,69.52% and 70.45% respectively on the three datasets.The accuracies were improved separately by 5.9%,9.52% and 8.68% with the joint denoising method.The outstanding diagnosis performance was proved under simulated complex coal processing plant environment.Subsequently,the industrial application was carried out on 300ZJ-I-A70 slurry pump of Hongliu coal processing plant.Vibration signals were collected from the slurry pump with one healthy state,two single fault states and three compound fault states respectively.Considering low quality of data collected in the industrial site,an abnormal data screening method was proposed based on LOF and working condition analysis.Then,screened signals were denoised by the joint denoising method.The quality of slurry pump vibration data set was improved based on the above data processing.On the one hand,the fault characteristics were clarified with envelope analysis.On the other hand,the intelligent fault diagnosis of slurry pump was carried out on the basis of Conv Ne Xt and 99.94% diagnosis accuracy was achieved.Based on the above research,the condition monitoring and fault diagnosis system of slurry pump was developed.The system was applied in 300ZJ-I-A70 slurry pump of Hongliu coal processing plant and realized the fault diagnosis of rolling bearing and impeller.Useful guidance was provided for the equipment maintenance work in coal processing plant.