Study Of The Vibrating Screen Damage Detection Based On Neural Network Model

With the improvement of the mechanical manufacturing technology and the needs of modern industrial development, the structure of machinery and equipment becomes more and more complex, the high efficiency of production dependent on strong production capacity provided by machinery and equipment, any failure of equipment can produce a huge loss. Therefore, fault diagnosis to mechanical equipment in production is very meaningful. Fault diagnosis could analyze and diagnose the fault state, fault position of the mechanical equipment and could get the development of fault on the basis of monitoring equipment operating state.Time domain and frequency domain analytical methods can not accurately determine the fault status and fault development trend of the equipment. This paper use system identification to establish the model of the vibrating screen, through the analysis of the characteristics of the identification model, study of fault diagnosis and crack development trends on the basis of system identification.In order to efficiently extract feature information from the collected signals, firstly, preprocess the measured vibration signal and normalize the data, and then using wavelet denoising method to denoise, using the least squares method to eliminate the signal trend part, finally remove the DC component of the signal.Study of various system identification methods when the vibrating screen scale model lower beam cracks. Respectively use linear models, nonlinear models, neural network models to model the vibrating screen. Then analyze degree of fitting between model and the actual system, combine with a large number of tests. The research result shows that the neural network model has higher accuracy used to fit the shaker system. Further to compare degree of fitting of model, to test residual of different neural network model, choose neural network NNARX identification model. At last, determine the various parameters of the model, including the number of storey, the number of hidden layer of neural network, activation function. And analyze the effect of these parameters on the model.Under different conditions that vibrating screen scale model lower beam cracks or not, respectively identify the neural network model of the system, and by analysis of amplitude spectrum of vibration signal,the model virtual response spectrum, model weights, obtain the properties of the analysis models. These properties can be the basis of judging whether the vibrating screen cracks.Finally, apply the method of analyzing the properties of model to the study of actual vibrating screen crack development trend. Identify the model of actual vibrating screen that cracks, obtain vibrating screen vibration signal by the way of increase in the number of days, study the change of the identification model weight in different times. By the statistical analysis of these changes, with the increase in the number of days, the weight of the model was gradually decreased and trend towards concentration. The experiments show that using the analysis of the identification model weights to study the development trend of the vibrating screen cracks is feasible, and also meaningful.