A neural network approach for the real-time detection of faults |
| |
Authors: | Yahya Chetouani |
| |
Institution: | (1) Département Génie chimique, Université de Rouen, Rue Lavoisier, 76821 Mont Saint Aignan Cedex, France |
| |
Abstract: | Fault detection is an essential part of the operation of any chemical plant. Early detection of faults is important in chemical
industry since a lot of damage and loss can result before a fault present in the system is detected. Even though fault detection
algorithms are designed and implemented for quickly detecting incidents, most these algorithms do not have an optimal property
in terms of detection delay with respect to false alarm rate. Based on the optimization property of cumulative sum (CUSUM),
a real-time system for detecting changes in dynamic systems is designed in this paper. This work is motivated by combining
two fault detection (FD) strategies; a simplified procedure of the incident detection problem is formulated by using both
the artificial neural networks (ANN) and the CUSUM statistical test (Page–Hinkley test). The design of a model-based residual
generator is intended to reveal any drift from the normal behavior of the process. In order to obtain a reliable model for
the normal process dynamics, the neural black-box modeling by means of a nonlinear auto-regressive with eXogenous input (NARX)
model has been chosen in this study. This paper also shows the choice and the performance of the neural network in the training
and test phases. After describing the system architecture and the proposed methodology of the fault detection, we present
a realistic application in order to show the technique’s potential. The purpose is to develop and test the fault detection
method on a real incident data, to detect the change presence, and pinpoint the moment it occurred. The experimental results
demonstrate the robustness of the FD method. |
| |
Keywords: | Safety Functioning risk Fault detection Reliability Neural network CUSUM |
本文献已被 SpringerLink 等数据库收录! |
|