Turbo-Generator Vibration Fault Diagnosis Based on PSO-BP Neural Networks

To overcome the flaws of traditional BP learning algorithm of its low convergence speed and easy falling into local extremum during turbo-generator vibration faults diagnosis, a novel algorithm called PSO-BP is proposed for artificial neural network (ANN) learning based on particle swarm optimization (PSO) in this paper. The algorithm covers the two phases. Firstly, PSO algorithm is applied to optimize the weight values of neural networks based on training samples. BP algorithm is then applied for farther optimization based on verifying samples till the best weight values are achieved. In addition, to simplify the structure of the networks, rough set theory (RST) is applied to implement the attributes reduction from the diagnostic decision table quantified by Kohonen networks. Thus the learning speed and the misjudgment rate of the diagnostic networks are improved, dramatically. Finally, the trials in turbo-generator vibration faults diagnosis indicate that the proposed method possesses better forecasting accuracy and lower misjudgment rate, and is an ideal pattern classifier. Key-Words: Turbo-Generator; Particle Swarm Optimization (PSO); BP neural networks; Rough set; Kohonen networks; Evaluation