The use of adaptive (neural) networks for fault diagnosis and process control is explored. Adaptive networks can be used as fault recognition systems, adaptive non-linear process models, and as controllers. Connection strengths representing correlation between inputs (alarms and sensor measurements) and outputs (faults) are made to learn by the network using the Back Propagation Algorithm (BPA). Results are presented for diagnosing faults in a Heat Exchanger CSTR system. The network employed in the present study has eight input nodes corresponding to the state variables and eight output nodes corresponding to the eight malfunctions (faults). A hidden layer of an optimum number of eleven nodes is chosen based first on heuristics and later on various trial and error combinations. A learning rate of 0.7 is used. The training used the lower and upper limits as keys for training as well as the limits for normalization. The final fault diagnosis (after network training) involved rounding of the values from the network. The network is able to identify all the faults correctly.
[1]
Lyle H. Ungar,et al.
Adaptive networks for fault diagnosis and process control
,
1990
.
[2]
Timo Sorsa,et al.
Neural networks in process fault diagnosis
,
1991,
IEEE Trans. Syst. Man Cybern..
[3]
Timothy Masters,et al.
Advanced algorithms for neural networks: a C++ sourcebook
,
1995
.
[4]
Venkat Venkatasubramanian,et al.
A neural network methodology for process fault diagnosis
,
1989
.
[5]
Masahiro Abe,et al.
Incipient fault diagnosis of chemical processes via artificial neural networks
,
1989
.
[6]
Ramaswamy Vaidyanathan,et al.
Process fault detection and diagnosis using neural networks
,
1990
.
[7]
Liya Hou,et al.
Diagnosis of multiple simultaneous fault via hierarchical artificial neural networks
,
1994
.