In most chemical processes, variables are sampled at different rates which brings great challenges to traditional process monitoring methods that are built upon single sampling rate. In this paper, a multi-rate partial least squares algorithm is proposed. Compared to the traditional PLS method, the proposed algorithm takes use of the incomplete data samples through a modification of both of the covariance matrix of the input dataset and the covariance matrix between the input and output datasets. Iteration is used in the model training step to avoid the calculation of same parameters which requires complete training datasets. Then the fault detection and online prediction strategy is proposed based on this algorithm. A case study on TE process shows that the proposed method had an enhanced performance on both monitoring and online prediction, compared to the traditional PLS method.
