Thesis for the Degree of Master of Science


Year 2008
Vesa-Matti Tikkala

Fault Diagnosis of a Board Machine Based on Causal Digraph Method

Fault diagnosis systems, which detect and locate the faults and the causes of abnormal conditions in process systems, are demanded by the industry. The requirements concerning quality, efficiency, environment and safety are increasing constantly. Meanwhile, the industry is imposed by the tightening competition and the increasing complexity of the processes. By providing supportive information for the operative personnel, the fault diagnosis systems help to keep the processes running efficiently.

Research in the field of fault diagnosis has been very active for over three decades. There have been developed numerous methods to deal with the faults in process systems. The causal digraph method has attracted considerable attention due to its cause-effect model structure and the powerful inference mechanism for fault isolation. However, the causal modelling of the processes is not straightforward.

The causal digraph methods for process fault diagnosis are reviewed in the literature part of this thesis. Also, the construction of the digraph models and their applications in the field of pulp and paper industry are presented. In the experimental part, the simulation-aided design of the causal digraph models of processes is studied. A methodology for the model construction is proposed and tested with a case study concerning the stock preparation process of the three-layer board machine. A causal digraph model is built by applying the methodology and tested with three consistency sensor faults simulated with dynamic process simulator.

The causal digraph model is built successfully and assessed with simulations using industrial data. By applying the proposed methodology it is possible to obtain a representative causal model of the process for fault diagnosis purposes. In the fault diagnosis test, the model was able to detect each consistency sensor fault and to successfully isolate two out of three.

This info last modified 26 Sep 2017 by Jerri Kämpe-Hellenius