This paper presents a tool designed for analysing fault propagation and fault impact on large-scale process performances. The analysis is based on structural description of the process. The main physical variables are associated to each subsystem and a relational model linking these variables for all the different functioning modes of the system is determined. In large-scale systems every component must provide a certain function in order to make the overall system working satisfactorily. When a fault or a badly tuned parameter affects a control loop, the required function cannot be fulfilled which may cause a failure. Therefore, some Loop Performance Indexes (LPI) indicating if the control loops operate properly are necessary to evaluate the impact of the failure on the overall process performances represented by a high level index Key Performance Index (KPI). Structural models provide an interesting approach for the analysis of a system and also studying the impact of a fault because they only need a limited knowledge about the behaviour of the system. Generic component models can be used to describe the system architecture. At the first level different statistical tests are applied to the KPI. When a set of LPI or KPI deviate from their nominal or desired values, the elements which are source of an eventual malfunctioning can be searched in the structural graph by searching the nodes predecessors. The selected LPI are tested in their turn by mean of statistical tests. A node is declared to be “faulty” if the value of the corresponding LPI is out of the acceptable (pre defined) limits. The procedure is iterated until the last level of the model is reached. This procedure researches the possible cause of KPI value significant deviation. The procedure was applied on a board machine. In this process, the main KPI is the value of moisture of the board at the end of the production chain. The corresponding structural model which relates the moisture (top node) to the control loops (nodes) has been developed. In order to validate the large-scale capabilities of such approach, the model has been integrated within the PREDICT's SOA(Services Oriented Archiecture) software platform: KASEM (Kowledge and Advanced Services for E-Monitoring). The platform enables to apply the “on-line” statistical test to the KPI and LPIs of the board machine and supports the iterative procedure Indeed, the iterative procedure based on the structural graph was integrated as one of the KASEM diagnostic tools with a dynamic and animated graph and used during the KASEM workflow to solve the problem.
[1]
N. Wiener,et al.
The prediction theory of multivariate stochastic processes
,
1957
.
[2]
Richard D. Braatz,et al.
Fault Detection and Diagnosis in Industrial Systems
,
2001
.
[3]
Carsten O. Daub,et al.
The mutual information: Detecting and evaluating dependencies between variables
,
2002,
ECCB.
[4]
D Xavier,et al.
On the Design of a Failure Detection Service for Large-Scale Distributed Systems
,
2003
.
[5]
M. V. Iordache,et al.
Diagnosis and Fault-Tolerant Control
,
2007,
IEEE Transactions on Automatic Control.
[6]
Christian Commault,et al.
Sensor Location for Diagnosis in Linear Systems: A Structural Analysis
,
2007,
IEEE Transactions on Automatic Control.