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, sensors, 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. Applying FDD for industrial robotics is a relatively new approach, thus, there exists a wide range of different types of industrial robots, and on the other hand there exist different FDD approaches such as data-driven, model-based, and knowledge-based approaches. FDD approaches can be distinguished into data-driven, model-based, and knowledge-based approaches as highlighted in the classification of existing FDD approaches [cf. figure]. 

Detecting faults emerging from external sensors of highly dynamic HRI work environments is mandatory as faults may lead to incorrect decisions within the robotic control unit and eventually to unexpected behaviours of the collaborative robot. Analytical or stochastic a priori models are particularly used in respect to internal sensors of a robotic system when the system operates in a well-known work environment. For robotic systems operating in unknown environments, predicting the values of external sensors is hardly possible. For these situations data-driven approaches are the better choice by applying sensor fusion techniques for external sensor fault detection. This means that multiple sensors sense different aspects of the environment (e.g. orientation and location), while their readings can be fused to form a consensus. Sensor-fusion-based fault detection approaches for robotic systems include different algorithms such as: Kalman filters, Dempster-Shafer, correlation and distribution-based, and Bayesian networks.

The figure below illustrates a robotic system which is composed out of hardware and software modules. Hardware components such as internal sensors, power, controller, and actuators might be subject to different hardware faults, which are typically diagnosed by model-based or knowledge-based diagnosis approaches. However, for the internal and external (exteroceptive) sensors (bottom left), data-driven approaches are more appropriate for FDD; in particular, sensor fusion for sensor fault detection, and data analytics techniques such as machine learning and particle filtering for utilizing sensor-readings for diagnosis.