Quality-Related Dynamic Process Monitoring: Part II

In the current dynamic process monitoring studies, several methods deriving from the PLS model, such as DTPLS and DiPLS, perform oblique decomposition, which will result in a false alarm to KPI. In fact, a false alarm is a key indicator in industrial process to keep the system stable. For example, in the steel rolling process, the fault of the hydraulic sensor in the fifth stand bending roll system has little impact on the thickness (KPI variable), which means it is quality-irrelevant. Once the above fault causes an alarm in the finishing exit thickness, it will be considered as a false alarm. As a result, the steel rolling process will be interrupted, and the burden on the finances and the manpower will be increased. Induced by the oblique decomposition of the aforementioned methods, useless information unrelated to KPI exists in the quality-related space (QRS). In addition, since the augmented matrix stores historical data, the quality-unrelated information will accumulate over time in QRS. When a quality-unrelated fault occurs, a large number of false alarms will be alarmed in QRS. To solve this problem, a feasible approach consists in mapping along the orthogonal projection operator. Based on this idea by performing orthogonal decomposition on the process data, the DM-PLS model is proposed, which reduces the false alarm for quality-unrelated fault. Nevertheless, a clear dynamic inner model to establish the relation between the process data and latent variables is neither provided nor is provided a clear dynamic outer model to establish the relation between the input latent variables and output latent variables.