Papers by Keyword: Probability Neural Network

Abstract: It is important to monitor and diagnosis for the health of ball bearing. We use the wavelet package to decompose the vibration of running bearing. The energies in frequency domain are input into a probability neural network. The output of the neural network is employed to justify the health of bearing and detect locations of fault. The experimental shows a good result of this method.

Abstract: Significant interrelationships between skirt shapes and properties of fabrics have been found in skirt design. In this paper, the determination method, classified fabrics based on Euclidean distance and optimum cluster can be implemented by multivariate ANOVA was proposed, after testing tensile, shear and bending properties of 30 different kinds of silken fabrics, selected as subjects, with KES. Fisher discrimination module was developed by using Visual Basic 6.0 language based on Fisher discrimination functions through SPSS 10.0. Probability neural network (PNN), established based on mechanical properties of train samples, was employed to study the classification of new sample. And the classification results were studied comparatively. The results showed that the proposed method, based on Euclidean distance and multivariate ANOVA, is feasible, and those silken fabrics can be classified into three clusters. The results also indicated that Fisher discrimination module and PNN is feasible to distinguish cluster of new sample. Discrimination of silken fabrics is easy to operate because of Fisher module, and has strong robust property in noises of test samples for the reason of PNN.

Abstract: In order to accurately estimate tool life for milling operation, a novel tool condition monitoring system was proposed to improve classifying precision in different cutting condition. Lots of features were extracted from cutting forces signal, vibration signal and acoustic emission signal by different signal processing method, only a few features selected by principal component analysis (PCA) according to contribution rate, and constructed as input vector. The relation between tool condition and features was built by radial basis probability neural network which control parameter of kernel function and hidden central vector were optimized by improved genetic algorithm. The experimental results show that the method proposed in the paper achieves higher recognition rate, good generalization ability and better available practicality.

Abstract: In this research, we used nondestructive test based on ultrasonic test as inspection method, and made up inspection robot in order to control of ultrasonic probe on the SWP surface, and programmed to signal processing code and pattern classifying code by user made programming code. For evaluation of flaw signal is reflected on welding flaw, user-made program codes are composed of signal processing and probability neural network (PNN) and backpropagation neural network (BPNN). And then, we actually confirmed to the theoretical advantage of each neural network method compared probability neural network with backpropagation neural network for classification and recognition rate. For the application of classifier to SWP inspection system, BPNN classifier is adequate in the first stage. And then, the application of PNN classifier is adequate as the next stage. Because of PNN application need enough sample data that is due to probabilistic density function.