Online fault detection of industrial processes by applying pseudo-random sequences

The systems in process industries are often large and complex. A typical system is characterised by hundreds or thousands of input and output variables and strong interactions between the sub processes. These factors often present difficulties for system monitoring. This paper presents the use of multiple-input multiple-output (MIMO) identification techniques in analysing the complex industrial systems online through frequency responses. Compared to more traditional identification methods, where a system is analysed through single-input experiments in the time domain, the proposed techniques provide several advantages such as shorter experiment time. This paper shows a practical approach and implementation to obtain the frequency responses. Experimental measurements are shown from a physical process which emulates the traditional headbox of a paper machine. The measurements are performed under normal operation conditions as well as under a fault condition. In addition, a simple computational method to recognise the fault condition based on the measured data is shown.