Research On Fault Diagnosis Method Of Rare Earth Extraction Stirring System Based On Time-frequency Analysis And Machine Learning

Rare earth extraction production line consists of separation system,stirring system and so on,where the stirring system consists of feeding motor,stirring motor,belt,stirring shaft,etc.The role is to migrate the rare earth elements from the aqueous phase to the organic phase,which is the key link of the whole rare earth extraction production process.There are many factors that affect the mixing effect,such as motor failure(abnormal speed,voltage and current),belt slippage,stirring shaft failure(internal fracture,failure of external or internal shaft),which directly affect the product quality and even produce waste or defective products.Therefore,timely identification and location of faults is an important guarantee to maintain the normal operation of the extraction system,but the current fault determination mainly through the monitoring system,not only can only achieve post-facto alarm,but also there are missed and false alarms,which affect the quality of the product,it is necessary to study the appropriate method of fault diagnosis based on the system operating parameters,to achieve the prediction of the fault state,as far as possible in the early stage of the fault occurred in time to detect,locate,and propose This will enable the system to detect and locate faults as early as possible and propose effective strategies.In response to the above problems,a fault diagnosis method for the mixing system of rare earth extraction production lines based on a combination of time-frequency analysis and machine learning is proposed,as follows.(1)A multivariate Stacking integration learning method combined with mathematical statistics is proposed as an early fault warning method for the complex and variable working conditions of the existing rare earth extraction production line stirring system.Firstly,the state variables that are highly correlated with the motor operating temperature of the rare earth extraction stirring system are selected by using the improved grey correlation analysis and input to the Stacking fusion algorithm to fit the motor operating temperature,and the actual temperature is solved for the residual value with the fitted temperature.(2)For common typical faults such as stirring shaft breakage and impeller breakage in mixing systems,a fault diagnosis method based on time-frequency feature classification is proposed.Firstly,the time-frequency analysis method of non-linear signals is introduced,the principles of wavelet transform and short-time Fourier transform are explained,and the two-dimensional time-frequency maps after signal processing are input to three convolutional Lenet,Alexnet and Googlenet The accuracy of the validation set and the decreasing rate of the loss function are used as the evaluation criteria for the classification effect.After several comparative experimental analyses,it was shown that the method combining Alexnet and wavelet transform was better in terms of classification accuracy and convergence speed.(3)A small-sample fault diagnosis method based on a deep migration model is proposed for mixing systems with low frequency and insufficient data,but with serious fault consequences such as bearing and stator and rotor faults.Based on the migration learning theory,the learned knowledge is transferred to the target domain using model migration to stabilise the target domain recognition accuracy while reducing the training and parameter search time.The experiments show that the proposed method can effectively improve the accuracy of recognition under small samples.