Failure Detection and Diagnosis (FDD) in Robotic Systems by ML techniquesFDD is a process monitoring and testing method for minimising downtimes, increase the safety of operations, and to reduce manufacturing costs.https://repo.valu3s.eu/method/failure-detection-and-diagnosis-fdd-in-robotic-systems-by-ml-techniqueshttps://repo.valu3s.eu/@@site-logo/logo_valu3s_green_transparent.png
Failure Detection and Diagnosis (FDD) in Robotic Systems by ML techniques
FDD is a process monitoring and testing method for minimising downtimes, increase the safety of operations, and to reduce manufacturing costs.
FDD is a process monitoring and testing method for minimising downtimes, increase the safety of operations, and to reduce manufacturing costs.
In the industrial robotics domain, faults have the potential to affect the efficiency of the underlying process, namely causing failures of internal physical components (e.g., robot, IPC, actuators), or even compromising the safety of humans interacting with the robot. When detecting a fault, usually a diagnosis process is induced in order to identify which internal components are involved. in order to detect anomalies in the to be process. This will be achieved through process mining and pattern recognition in data from the original assembly process used to develop and train ML model. The resulting model will be used for failure detection in manipulated data streams in the selected test scenarios.
Combination of FDD with advanced AI techniques such as machine learning techniques to provide useful and sophisticated diagnosis model
Deficit for using FDD approaches (particularly data-driven approaches) that are dedicated to detecting and diagnosing HRI-related faults. HRI is subject to uncertainty due to the fact that unexpected outcomes might lead to unknown faults and failed interactions.
Interaction-related faults between humans and robots, humans may have the tendency to compensate for a faulty behaviour of a robot during interaction.
A robotic system may provide a lot of data per se; thus, pure, data-driven approaches do not tend to exploit existing and available knowledge about the robotic system. As such, mining streamed data in an online manner for fault detection may be impractical for some reasons, and faults might not be detected immediately. Thus, when using fault injection (simulated or real), one cannot confidently account for all possible faults.
It is further a limitation when injecting a fault directly into a data stream since this strategy might not sufficiently represent the full impact of the fault on the whole system. Moreover, injecting a real fault may damage the robotic system, thus it has to be ensured that FDD experiments and tests are continuously supervised.
