Residual life prediction method of roll-packing units based on multi-component vibration model and CIndRNN neural network

The degradation information, such as vibration and wear, of the equipment components in the roll-packing units and their production and operation data have a significant impact on predicting the residual life and maintenance cycle of the whole machine. Such factors also seriously affect the formulation of equipment maintenance cost quotas. This paper presents a multi-component vibration model that generates the cumulative vibration displacement of vulnerable components as the key feature dataset based on damage accumulation theory. The model fuses this dataset with two other feature datasets, namely the specified model output and startup time. A Convolutional Independent Recurrent Neural Network (CIndRNN) model is established based on multivariate sequence input and local feature learning to accurately predict the residual life of the roll-packing units. Using the ZB45 SB#3 machine as an example, the results show that this method can reflect the correlation between the diverse influencing factors and the prediction object. Furthermore, the residual life prediction value of the unit is positively correlated with the yield, and the prediction error is only 3.4%. This method provides an algorithmic basis for accurately predicting the residual life of the roll-packing units, and reasonably formulating the equipment maintenance cost quota standard.