Study On Nondestructive Detection Of Anchorage Quality Of Rock Bolts Based On RBF Neural Network

The detection of anchorage quality of rock bolt is very important due to its supporting function in the construction of slope, deep excavation and tunnel, etc. Identifying defects and defects types for anchorage are significant in detection of anchorage quality of rock bolt. Combing with theoretical analysis、model rock bolt testing and practical project case, a study on nondestructive detection in anchorage quality of rock bolt has been carried out in this paper. On the basis of wavelet analysis, particle swarm optimization(PSO) algorithm, a radial basis function(RBF) neural network for anchorage are put forward to classify defect types intelligently, overcoming the shortcomings of personal experience. The main achievements of study works are as follows:(1)A wavelet threshold denoising method was improved to filter the measured signal of rock bolt. The simulation results show that, compared with soft threshold methods, hard threshold methods and other threshold methods in some literature, the improved threshold denoising method detects parameters of rock bolt more accurately and can provide more precise data for identifying the defect type of anchorage.(2)Wavelet transformation is performed on the acceleration signal of anchorage system, extracting energy of each band as feature vector for identification of defect types. The sample of RBF neural network is composed of feature vector extracted and defect type, with feature vectors being input of RBF neural network and defect type being output.(3)Improve the RBF neural network model. First, the number of hidden layer nodes is determined by subtraction clustering algorithm; the initial center and width of the hidden layer are determined by K-means and the weight between hidden layer and output layer are obtained by the least squares method. Second, the fixed inertia weight PSO algorithm, Linear Decreasing Inertia PSO algorithm and an improved inertia weight and location formula PSO algorithm are used to optimize the the hidden layer centers and widths of RBF neural network respectively, and three different PSO-RBF neural network model are built. Finally, test samples are utilized to verify the performance of the PSO-RBF neural network model. The result shows that the improved PSO algorithm leads to better accuracy and higher speed, compared with the fixed inertia weight PSO algorithm and Linear Decreasing Inertia PSO algorithm.(4)According to the anchorage with different defect types in laboratory, improved threshold denoising method, wavelet energy spectrum analysis and improved PSO-RBF neural network are used to identify the anchorage. In addition, anchorages of practical project are identified. The results show that the improved PSO-RBF can identify the defect types of anchorage accurately.