Comparison of data and model based sensor data pre-treatment in a hydro-cracker

AbstractDefect or non-available measurements due to damaged sensors or transmitters in chemical processes may lead to disastrous effects on quality and safety. In this work, a hydro-cracker is dealt with where faulty measurements may cause the cooling of the entire reactor and producing of low quality rejections. First, some state of the art data based methods are presented for outlier detection and data imputation. These methods assume some conditions (e.g. Gaussian distribution) which were not fulfilled in the presented plant. Therefore, two new model based methods were developed which allow to detect and to give an option to substitute outliers or non-available measurements in the quench of a hydro-cracker.