Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 74
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON THE APPLICATION OF ARTIFICIAL INTELLIGENCE TO CIVIL AND STRUCTURAL ENGINEERING
Edited by: B.H.V. Topping and B. Kumar
Paper 34

Non-destructive Testing of Ground Anchorages Using the GRANIT® Technique

A. Starkey, A. Ivanovic, R.D. Neilson and A.A. Rodger

Engineering Department, University of Aberdeen, United Kingdom

Full Bibliographic Reference for this paper
A. Starkey, A. Ivanovic, R.D. Neilson, A.A. Rodger, "Non-destructive Testing of Ground Anchorages Using the GRANIT® Technique", in B.H.V. Topping, B. Kumar, (Editors), "Proceedings of the Sixth International Conference on the Application of Artificial Intelligence to Civil and Structural Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 34, 2001. doi:10.4203/ccp.74.34
Keywords: artificial intelligence, condition monitoring, dynamic response, ground anchorage, non-destructive testing, vibration, neural network.

Summary
The GRANIT® system is a non-destructive integrity testing method for ground anchorages. It has won two major UK Awards - the Design Council Millennium Product Status in 1999 and the John Logie Baird Award in 1997. It makes use of novel artificial intelligence techniques in order to learn the complicated relationship that exists between an anchorage and its frequency response to an impulse. The GRANIT® system has a world-wide patent and is currently licensed exclusively to AMEC plc. It is widely recognised that non-destructive testing methods for ground anchorages need to be developed as a high priority with only between 1-5% of anchorages currently being monitored in service.

The GRANIT® technique works by applying a tensile axial impulse to the anchorage by means of a specially designed Impact Device. The resultant acceleration signal is measured by an accelerometer positioned on the Impact Device, and the signal recorded by a ruggedised computer system. These results are then processed remotely using MATLAB with Fourier analysis techniques and Wavelet techniques, in order to extract the relevant information that is contained within the response signature of the anchorage which relates to its load level. This information is then passed on to a neural network for training. A neural network is used at this point since the relationship between the load level of an anchorage and its response to an applied impulse may be a complicated non-linear relationship e.g. the response signature of the anchorage may change with a different anchorage head geometry, or a different diameter of anchorage. Therefore, the use of neural networks allows for the rapid and effective training and subsequent diagnosis of a number of different anchorage configurations.

A test site at AMEC's White Bear Yard, Adlington, UK was constructed by AMEC which included two bolt anchorages of similar construction which were of a specification that is commonly used in the field. The bolt anchorages were 12m in length, with a 3m rapid cement grouted fixed length, and a 9m free length. The protruding length for both anchorages was 0.7m, in order to ensure consistency with earlier tests undertaken on test anchorages in the laboratory at the University of Aberdeen. On this particular application of the GRANIT® system, training data was taken from a test bolt designated Bolt 1. This data was used to train a neural network. Further test data was taken from Bolt 1 for confirmation that the neural network had correctly learnt the relationship between the response of the anchorage and its load level. The results of the diagnosis of the network to this test data can be seen in Table 1, where the diagnosis of a number of test samples has been averaged and rounded to the nearest whole number. As can be seen in the results, the neural network has learnt the complicated non-linear relationship between load level and the dynamic response of the anchorage for Bolt 1 to a high degree of accuracy.


Diagnosis of Bolt 1   Diagnosis of Bolt 2
Actual (kN) Average (kN)   Actual (kN) Average (kN)
0 0   0 0
6 7          7 6
13 14          13 17
22 22          20 20
35 35          32 25

Table 34.1: Diagnosis of a neural network trained on data from Bolt 1 to test data from Bolts 1 and 2.


Further tests were taken from the adjacent bolt, Bolt 2. The same procedure was adopted throughout, with the Impact Device placed at the same distance from the anchorage head, and the same size of impulse being applied. The neural network that was trained on Bolt 1 was used to diagnose the results of data taken from Bolt 2. It should therefore be noted that the neural network has no "experience" of Bolt 2. However, since the two anchorages were installed in a similar fashion and should therefore possess similar characteristics, it was expected that the relationship between the dynamic response of Bolt 2 and its load level should be similar to the relationship that the neural network has learnt for Bolt 1. The results can be seen in Table 34.1. The results show that the neural network that has been trained on Bolt 1 is capable of diagnosing results taken from a different anchorage, of which it has no previous "knowledge".

These results demonstrate the potential of the GRANIT® system, whereby large numbers of similarly constructed anchorages can be tested by a single trained neural network. The training process (in this case) took place at the end of a single day of testing, with the results of the diagnosis available the next day for both anchorages. The rapid and effective deployment of the GRANIT® system as demonstrated in the test results at the Adlington site would enable large numbers of anchorages to be tested for loss in load quickly and competently. This new innovative non- destructive integrity testing technique for anchorages can be used to enable, for the first time, a cost-effective method of determining the condition of anchorages.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description