17-03-2014, 10:20 PM
Abstract
BP neural network is easily trapped into the local
minimum during the training process, which results
that it can’t get the optimal solution, even misjudging
in device fault diagnosis. Directing to the above
problems, a Hopfield-BP neural network fault
diagnosis method was proposed, which combined
Hopfield neural network, having the global optimal
neural network computing ability, with the BP neural
network, charactering the nonlinear classification
ability. It avoids the network to be trapped to a local
optimum. Implementing the new network into the fault
diagnosis of centrifugal fan has proven that fault
pattern recognition could be solved well, and the
accuracy of fault diagnosis is increased than that with
the method of BP neural network.
Key words: Hopfield neural network; BP neural
network; Fault Diagnosis; Hopfield-BP neural
network
1. Introduction
The essential of fault diagnosis is pattern
recognition, that is, the system state can be recognized
with measured eigenvector[1][2]. Neural network
technology is with the capabilities of high level selforganization,
self-learning, and nonlinear mapping,
which is good at solving the problem of pattern
recognition. It has been successfully applied to fault
diagnosis[3][4][5][6]. BP network is one of the most
mature and extensive application neural network, but it
is easily trapped into the local minimum in practical
application, which results in difficulty of acquiring
global optimum[7][8]. This paper proposes Hopfield-
BP neural network which combined with the
optimization of Hopfield neural network and the
advantage of BP network to solve the problem of nonlinear
classification. Validity and feasibility of the
method based on Hopfield-BP neural network is
verified by the example of centrifugal fan fault
diagnosis.
BP neural network is easily trapped into the local
minimum during the training process, which results
that it can’t get the optimal solution, even misjudging
in device fault diagnosis. Directing to the above
problems, a Hopfield-BP neural network fault
diagnosis method was proposed, which combined
Hopfield neural network, having the global optimal
neural network computing ability, with the BP neural
network, charactering the nonlinear classification
ability. It avoids the network to be trapped to a local
optimum. Implementing the new network into the fault
diagnosis of centrifugal fan has proven that fault
pattern recognition could be solved well, and the
accuracy of fault diagnosis is increased than that with
the method of BP neural network.
Key words: Hopfield neural network; BP neural
network; Fault Diagnosis; Hopfield-BP neural
network
1. Introduction
The essential of fault diagnosis is pattern
recognition, that is, the system state can be recognized
with measured eigenvector[1][2]. Neural network
technology is with the capabilities of high level selforganization,
self-learning, and nonlinear mapping,
which is good at solving the problem of pattern
recognition. It has been successfully applied to fault
diagnosis[3][4][5][6]. BP network is one of the most
mature and extensive application neural network, but it
is easily trapped into the local minimum in practical
application, which results in difficulty of acquiring
global optimum[7][8]. This paper proposes Hopfield-
BP neural network which combined with the
optimization of Hopfield neural network and the
advantage of BP network to solve the problem of nonlinear
classification. Validity and feasibility of the
method based on Hopfield-BP neural network is
verified by the example of centrifugal fan fault
diagnosis.