Measuring Cause and Effect between Process Variables

Disturbances in chemical processes spreading through plants affect a number of process variables. The operator or the monitoring scheme in place often detects the disturbance at a variable critical to the process. This variable might not be close to the actual cause and isolation of the root cause becomes therefore a task for the control engineer. The aim of this study is to find ways of identifying cause and effect and thus the direction of propagation of a disturbance using only historical data of the process variables. A recently introduced method, predictability improvement (PredI), is based on embedded vectors and finding their nearest neighbors. The modified PredI algorithm works by exploiting, firstly, time delays which often occur between two measuring points and, secondly, attenuation of the signature of the disturbance related to the distance of the root cause. A new case study of an industrial reaction process is presented. The modified PredI method is applied to a disturbance observed by the operators at the bottom of the reactor and successfully shows that the disturbance originates from a root cause further upstream.