The study has developed a technique of sensing unstable behavior of slopes by conducting x-MR univariate statistical analyses of individual sensors and T² multivariate statistical analyses which makes it possible to consider correlations between various sensors for Pore pressure and water content acquired while flume tests on granitic weathered soil and gneissic weathered soil were being conducted.
When failure types observed during flume tests are classified according to Varnes' criteria (1996), granitic weathered soil belongs to active retrogressive single, active advancing single, active enlarging successive, and gneissic weathered soil belongs to active retrogressive single and active advancing successive.
As an analyzing method of early detecting unstable behavior of slopes, univariate and multivariate statistical analyses were conducted. To detect individually unstable behaviors of monitoring data acquired position by position of sensors (upper, middle, lower position) and kind (Pore pressure, water content) by kind, univariate statistical analyses were conducted. To analyze unstable behavior of all the sensors acquired at the time of the flume test, multivariate statistical analyses were conducted which makes it possible to simultaneously analyze numerous data sets in correlation.
To utilize, as the criteria for a forecasting and warning system, the results of the analysis of x-MR control charts, a type of the univariate analytic approach, analyses were conducted after the control limit line was divided into three steps (1σ, 2σ, 3σ). The analysis indicated that time of unstable behavior varied depending on the position and kind of the sensor, and reflected traits of the sensor position relative to behavior of the ground.
The data acquired from the measuring sensor installed on the same slope were correlated with each other. Accordingly, to consider correlations between the measured data, Hotelling's T² statistic, one of multivariate statistical analyses, was conducted, and a confidence interval was set at 95.0% and 90.0%. As a result of the analysis, collapse of a slope was detected hundreds ~ thousands of seconds ahead of the time of collapse of the slope.
Univariate and multivariate statistical analyses shows that the time prior to failure of T² value with a confidence level at 90.0% was similar to, or faster than, the results acquired by the x-MR control chart analysis.
The univariate statistical analysis comes in handy when studying individual data traits concerning various factors measured to failure prediction of slopes. However, the time predicted of the collapse of slopes varied on the same slope depending on the position and kind of the measuring sensor used, and sometimes errors occurred in sensing unstable behaviors. Since the multivariate statistical analysis, which comes up with the only control limit line for the time of collapse of slopes by simultaneously taking into numerous data sets in correlation, can supplement weaknesses of the univariate statistical analysis. Therefore, both univariate and multivariate statistical analysis need to be applied simultaneously.