Automatic Analysis of AUV Log Data for Expert System Generation
|Supervisor:||Horst Hellbrück , Torsten Teubler|
Expert systems are a well known technique in Artificial Intelligence (AI) as they provide means to establish the experiential knowledge of human experts in a computer system. Expert systems are widely used in medicine mainly for diagnosis tasks. In robotics expert systems are also used to control autonomous robots.
In this work an expert system for online diagnosis and failure detection for an Autonomous Underwater Vehicle (AUV) has to be developed using the CLIPS expert system tool. The usual approach to establish the knowledge in the expert system is to ask human experts. In the approach which has to be developed in this thesis the knowledge is taken from log files recorded during supervised operation mode of the AUV. Therefore, the log data is considered to represent the normal operation mode. This log data has to be examined for correlations. An example is a correlation between the depth control the pressure sensor. If the AUV dives the pressure increases in normal operation. If this is not the case either the pressure sensor or the diving mechanism does not work. These correlations have to be implemented in the expert system by first analyzing the log data and finding correlations. In a second step a code generator has to be written which generates expert system code which identifies deviations for the normal operation mode based on the correlations found in the first step. There is already some existing work (e.g. Failure Detection in an Autonomous Underwater Vehicle) in the field but unlike our approach correlations are found manually and the expert system is automatically generated.
- Implement a technique to find correlations in the AUV log data automatically
- Visualize the found correlations and check for soundness manually
- Implementing a code generator for an expert system for identifying deviations form the normal operation mode
- Evaluation of the generated expert system with log data and modified test data
- Good programming skills in C/C++, Java, Python, etc.
- Willingness to incorporate in AI, expert systems, and signal processing
- Willingness to incorporate in other software tools like CLIPS