Demonstration of faults/attacks detection with Data-driven Fault Detector

Short Description

This demonstrator shows how the use of the Data-driven Fault Detector, that is based on AI and system identification algorithms, is effective in detecting the faults/attacks that can occur on the system, also in the case of complex systems. The Data-driven Fault Detector tool is developed in Matlab. The tool implements a modelling procedure based on AI and system identification algorithms, and a detection algorithm that leverages such models. The goal of the tool is to create an automated system that helps in the detection of a fault or an attack that can occur on the system, that can be in general too complex to deal with, by analysing the system’s dynamics and comparing them with the modelled nominal expected behaviour. The Data-driven Fault Detector operates with the following sub-tasks: - Creation of a data-driven model based on the PCA algorithm using pre-processed data collected from the system - Setup a Kalman filter using the parameters derived from the training of the PCA-based model - Define thresholds to make the detector sensitive to changes that occur on the system’s dynamic when system deviates from nominal operations - Compare the Kalman filter state evolution with the threshold to detect an unexpected behaviour - If historical data of the possible failures can be collected a model of the faulty system can be created, and thus an identification of the failure can be processed other than the detection only The demonstrator implemented for UC6 consists of the scenario where the data measured by the IMU and the ones obtained by the 2 wheels of the robot are not coherent. In this respect, ESTE simulated such scenario on the robot and provided data for the algorithm testing.

Use cases

Responsible(s)/Contact(s)

TRL

Demonstrator Evaluation Criterias

Highlights/Rationale of the demonstrator

Transferability Comment Transferability defines how easily the technology can be applied in different systems or areas.

Transferability level 1-Minor, 2-Fair, 3-Good , 4-Very good , 5-Excellent (Minor: The purpose of the technology is connected to a specific system. It is hardly possible or even impossible to adopt the Technology in different systems. Fair: The technology is strongly coupled with another technology that is limited to be used elsewhere (e.g., due to licensing). Good: The technology is developed for a specific purpose but enables modifications to be applied elsewhere. Some specifics must be taken into consideration while adopting it. Very good: The technology is not purpose specific and can be applied elsewhere with small modifications. Excellent: The technology is general, implements standard interfaces, or has already been used in different areas with no or minor modifications.)

Improvement Level 1-Minor, 2-Fair, 3-Good , 4-Very good , 5-Excellent

Demonstrator Result/Measurement of V&V process Improvement List of V&V process improvement measurement based on VALU3S Evaluation Criterias per Use case Test Case and requirement

Demonstrator Result/Measurement of V&V SCP Improvement List of V&V of SCP improvement based on VALU3S Evaluation Criterias per Use case Test Case and requirement

Link to use cases

Qualitative Results

Qualitative Evaluation results obtained through QAM

Demonstrator pitch video Demonstrator pitch video hosted in VALU3S youtube channel