Application of the Akaike Information Criterion to the interpretation of bender element tests

While the use of bender elements for geotechnical testing dates back to the sixties and seventies, the interpretation of this wave propagation test is still a debated topic. The most widely accepted method involves plotting a signal in the time domain and manually finding the arrival point. This requires identifying an arrival time corresponding to the velocity of a plane wave in a setting where the validity of plane wave theory is, at best, debatable. Inevitably, this manual interpretation process can become error-prone and subjective. While this paper does not pretend to solve the issue of applying plane wave theory outside its domain of validity, it mainly aims to make the interpretation process more objective and repeatable. To achieve this, a statistical tool named Akaike Information Criterion (AIC) is used to separate a signal into two stationary processes: before and after the arrival of a wave. The main challenge is to isolate a segment of signal where these two stationary processes can be identified. For this, two separate algorithms are built using different methods of identifying the optimal signal segment. Both algorithms are described and tested on a large set of signals recorded on Fontainebleau sand specimens.