Using the kernel function method, you can realize the self-organizing clustering of data in high-dimensional image space in the input space without knowing the specific form of the mapping f. Clustering identification of gear faults The test in this paper was carried out at a gear research institute. The test equipment is an electric closed gear fatigue tester. The test uses an acceleration sensor to obtain the acceleration signal of the gearbox vibration from the bearing housing, and the sampling gear is normal. There are 60 sets of experimental data in the three states of crack and broken teeth, and each group has 20 sets of data. For the validity of the test method, 8 time domain features commonly used in gear vibration signals are selected in cluster training, including maximum and minimum. Value, standard deviation, rms, mean, crest factor, margin factor, and pulse factor.
In order to compare the clustering performance between traditional SOM and KSOM, using the above training set data, the explosion-proof oil-cooled electric drums are continuously trained 20 times using traditional SOM and KSOM networks to compare their respective clustering performances. The SOM network uses the same network structure. KSOM uses radial basis functions, polynomial functions and functions as kernel functions to train, and the corresponding clustering performance is compared. The legends include circles, plus signs, triangles, and diamonds to represent traditional SOM. Based on the clustering performance of KSOM based on radial basis function, polynomial function and sigmoid function, cluster training can be carried out 20 times in succession. The topological resolution of KSOM training result is much smaller than that of traditional SOM, about 50 of the latter; The inter-class dispersion of KSOM training is larger than that of the conventional SOM, and it is also much more stable. When the KSOM method is used for learning classification, different categories can be better distinguished. In addition, when different kernel functions are selected, the topological resolution is The change in the dispersion between classes is small, which is similar to the training results when the kernel function is selected in the support vector machine.
Integrated Buried Sewage Treatment Equipment
Waste Water Treatment Equipment,Underground Waste Water Treatment,Underground Type Sewage System
Huatao Group , https://www.huataofilter.com