A Self-Attention and GRU based Method for Operational State Assessment of Hydrocyclone

Hydrocyclone is extensively known as an important equipment for separating mineral mixtures with different particle sizes in the mineral processing. However, the undesirable blockage faults seriously threaten the safety and stability of the hydrocyclones. Therefore, it is valuable to evaluate the operational state of the hydrocyclone and detect the blockage fault timely. To solve the above problems, a Self- Attention and GRU based operational state assessment algorithm of hydrocyclone is proposed in this paper. Firstly, the acquired multivariate temporal sequences are split into the many discrete window data by sliding window technique. Then, K-mean method is used to cluster the obtained window data, and the state labels can be assigned to each window data in chronological order. Lastly, the self-attention mechanism is used to obtain the dependencies between the window data variables, while the corresponding temporal features are captured by Gated Recurrent Unit (GRU) network. Based on the obtained variable dependencies and temporal features, an operational state assessment model of the hydrocyclone is built by using a feedforward neural network. Verification experiments using the operational data of the hydrocyclone show that the proposed algorithm is effective and feasible.