Andreas Eichhorn

An adaptive Kalman-filtering approach for the calibration of finite difference models of mass movements

Abstract Landslides are natural geomorphological phenomenas that can cause hazardous situations to men and infrastructure especially in densely populated areas. For the investigation of such events often numerical slope models are used, that describe the mechanical and geological properties of bedrock. However, the adaptation of the model to the monitored data is commonly done by ‘trial and error’ methods. In order to improve the adaptation process, adaptive Kalman-filtering techniques shall be used in terms of a realistic model calibration. Nevertheless, this method is very computationally intensive applied on full slope models with a larger grid size. Since the deformation process is normally restricted to limited parts of the investigation area (e.g. areas close to the surface), the Kalman-filter algorithm may be applied to selected parts of the grid with predicted major displacements. The first part of the paper is focussed on the monitoring design of the landslide ‘Steinlehnen’ (Tyrol, Austria). For this mass movement, a numerical model is currently under development and shall be calibrated by adaptive Kalman-filtering. In the second part, some investigation results for the adaptive Kalman-filtering approach are presented and discussed regarding a still simulated numerical test slope.

A Multi-Scale Monitoring Concept for Landslide Disaster Mitigation

There has already been much research on how the kinematic and geodynamic behaviour of landslides can be predicted. However, until now, there has been no decisive breakthrough for a monitoring and evaluation system combined with an alert system. Therefore an interdisciplinary international project OASYS (Integrated Optimization of Landslide Alert Systems) was commenced to progress research in this area. The members of the project, supported by the European Union, believe a multidisciplinary integration of different methods has potential for substantial progress in natural hazards management. This project proposes a new method consisting of three different steps: 1. Detection of boundary lines of potential landslides based on large scale information. 2. Detection of “taking-off-domains” and permanent local scale monitoring of these regions with high sensitivity geotechnical measurement methods. 3. Knowledge-based derivation of real time information regarding actual risks to support alert systems.

Concept of a Multi-Scale Monitoring and Evaluation System for Landslide Disaster Prediction

In 2006, OASYS, an EU funded project on a multi-scale monitoring concept for landslides as a basis for an alert system, was completed. 12 institutes from 6 countries tried to merge their multidisciplinary knowledge in the field of landslides and disaster management. The main goal of the research was to develop a cost saving concept for landslide disaster prediction in areas with a higher density of landslides. The present paper reports about the innovative steps and about some highlights of the research, emphasising mainly three tasks: & GIS integrated geological evaluations of remote-sensing data to delineate the high-risk areas in regions with a larger number of landslides & geometrical analysis of the monitoring data by fuzzy techniques as a basis for the design of the sensor network and & geomechanical modelling of the landslide by FD-methods as a basic information for an alarm system.

Applications of knowledge-based systems in technical surveying

Abstract In this paper the application of knowledge-based systems for the analysis and evaluation of geodetic measurements is described. The first part gives a nontechnical overview over existing software products like GeoFit and current research activities (e.g. TUNCONSTRUCT, KASIP) with special focuses on tunneling and natural disaster management. The second part of the paper describes a case study from technical surveying in detail: the knowledge-based evaluation of the quality of free stationing on construction sites (e.g. tunnel sites). One main goal of the knowledge-based approach is to automate the time consuming daily process of manually checking lots of datasets by human experts. The paper shows the development of the knowledge-based system ranging from knowledge acquisition over knowledge analysis and representation up to the software prototype. A practical example is presented which indicates the capability of the system even to reproduce the often intuitive ratings of human experts.

WiKaF - A Knowledge-based Kalman-Filter for Pedestrian Positioning

The precise, reliable and preferably ubiquitous positioning of mobile users is a substantial precondition for the provision of location and situation based information by Location Based Services. Within these applications, determining the location of pedestrians in ‘passive environment’ represents a special challenge. In this paper a new knowledge-based approach for the improvement of position quality is presented.

Map-independent positioning of land vehicles with causative modified motion equations

In this paper some results of the project ‘Positioning Component’ are presented, which was subject of a cooperation between the University of Stuttgart and DaimlerChrysler AG. The aim of the project was to develop a map-independent module for the autonomous positioning of vehicles within the framework of Location Based Services (LBS). The integration of the absolute position information (GPS) and the relative information (differential odometer and gyro) is realized with a central Kalman-filter. The system equations of this filter originally base on a kinematic model (constant velocity in one sampling interval). To reduce the filter’s inertia in the case of fast cornering it is necessary to modify these equations in a causative sense: the measured yawing (derived from the gyro) is used as a ‘geometrical’ correcting variable. The presented examples of filter results show different representative driving scenarios in inner urban areas with bad GPS quality and on highways. The mean errors of the estimated positions vary from approx. s P = 1,9 m (on highways) to 3,0 m (in urban areas with dense buildings and tunnels). The accuracy and availability requirements of the car manufacturer are achieved.