A modified double-latent variable probabilistic model for monitoring of dynamic processes with multiple sampling rates

The monitoring of quality-correlated aspects in industrial production processes has become a crucial task in recent years. However, the challenges posed by multiple sampling rates and dynamic issues make it arduous to construct an efficient monitoring model. To address these issues, the present paper proposes a modified double-latent variable probabilistic (MDLVP) model that can deal with the measurement correlations across different sampling rates. Firstly, the MDLVP introduces two types of latent variables with minimum sample spacing for capturing quality-correlated and quality-uncorrelated information respectively. Secondly, a first-order Markov chain is utilized to describe the autocorrelation of the latent variables, thereby elucidating the dynamics of the multi-sampling rate process. The expectation–maximization (EM) algorithm is employed for the model training in an incomplete data collection. Finally, the model is adopted to develop a fault detection method, which is subsequently applied in two industrial cases. The experimental results demonstrate the superiority of the proposed model in handling dynamic in multi-sampling rate processes.