Research On Nondestructive Detection In Anchorage Quality Of Rock Bolts Based On BP Neural Network

Rock bolts now used widely in the world, such as tunneling, mining, high slope and deep foundation excavation etc. However, in our country nondestructive testing of rock bolt system remained in using the hydraulic jacks for destructive pull-out test phase. This method wasted time and work, it belonged to damage detection. Therefore, searching for a quickly and accurately method of nondestructive testing of rock bolt system was becoming the top priority.Stress wave reflection method was used widely at home and abroad, but it was short of corresponding theoretical system support, nondestructive testing mainly relied on inspection person's empirical judgments, and it could not be accurately judged because of the interference of external factors. Therefore this paper used BP neural network for evaluating the quality of rock bolt system and achieved the following results through learning and training a large number of measured waveform:(1) Through analyzing the basic principle of non-destructive testing of rock bolt system based on the method of low-strain stress wave reflection, I found the limitation of this method. Identified the limitation such as difficult to identify the bottom reflection and difficult to judge the length of bolt, I proposed the method of using the reflection wave-shaped curve, divided these curves into 16 levels in these time domain direction, then i got a set of reflection amplitude array. I took the array as the factors of judging the quality of rock bolt system.(2) In the same training conditions (input vectors and the number of neurons of the network model, etc), compared the error curves with traingdm, traingda, trainlm and trainscg algorithms after training 200 steps, I found that the network had very fast convergence speed using the conjugate gradient algorithm, and it was suitable for the complex BP neural network.(3) In this paper, I took the amplitude values of 90-group reflection wave-shaped curves as the learning samples and the amplitude values of 20-group reflection wave-shaped curves as the test samples, chose the conjugate gradient algorithm for the optimization of BP neural network, founded 17-13-4 BP neural network model. Through studying and training the samples of 90-group, the training error of network reached 0.00995888 after 220 steps, achieved the goal 0.01 of training requirements. I took the samples of 20-group input the BP neural network for positive operation, got the quality codes of rock bolt system. Compared with the actual quality codes of rock bolt system, I found that this BP neural network had a good recognition accuracy and definite value.