Analysis of the data-driven fault detection process

Use Case Id

Description of evaluation

The data collected from the system are used to derive a linear or nonlinear model of the system to describe its dynamics. The dynamical behavior of such system is used together with a Kalman filter to check if a change in the evolution of the dynamics occurred. In particular, a PCA model based on ARX modeling, extended with Poly-Exponential modeling if necessary, is built based on the collected data, and the ARX parameters describing the system are used as the state of a Kalman filter. Then, the analysis of the state evolution of the Kalman filter corresponds to the analysis of the dynamical behavior of the system. If a fault/attack is occuring on the system, it will produce a change in the dynamical behavior, and thus in the Kalman filter evolution. Thus, by monitoring the state of the filter, and comparing it with tresholds defined ad-hoc based on the system under study, it is possible to detect whether a fault/attack is occurring.

Test case Test case where the V&V method has been used to improve a evaluation criteria. The test case where the V&V process improvement has been achieved

Requirement The validated requirement where a VALU3S V&V method has been used.

Evaluation Scenario The evaluation scenario where the V&V evaluation result has been obtained

V&V method

V&V Tool

Evaluation Criteria for Safety, Cybersecurity, and Privacy (SCP)

VALU3S SCP Criteria

Quantitative results Description and specification of the evaluation criteria used to measure the improvement (D5.6)

3 different scenarios analyzed.

100% of the scenario have been recognized.

Long Description

We started from the operations of the AgriRobot from UC6. Two maneuvers in teleopereted mode have been analyzed, and operating data have been collected. Data are related to the acquisition from three sensors: percentage speed (from -100% to 100%) of left and right wheels and, from IMU, angle of the robot with respect to z-axis (orientation of the robot). The objective is to detect wrong behaviours of the robot with respect to its normal (nominal) operations. Two nominal operation have been generated on the robot. Then a faulty trajectory of the robot has been simulated. The fault detector has been able to both detect such an anomaly, and recognize the correct operation (i.e., no faults occurred) of the nominal trajectories.

Contents

There are currently no items in this folder.