PCA Guided K-Means and Fuzzy C-Means for Fault Diagnosis of Robotic Systems

Robotics, machine learning, and other data-intensive technologies are becoming increasingly prevalent in smart manufacturing environments. Industries reliant on these technologies include automotive, aerospace, and other critical industries that demand accuracy and reliability. Therefore, there is a growing demand for robotics technologies that can perform a greater number of tasks in increasingly complex configurations with other assets and resources. Robotic systems will naturally degrade with time, and with this degradation, there will be an increasing number of faults in the robot, a fault being a discrepancy between the desired and actual value associated with a robot's operation. To ensure accuracy and reliability in increasingly complex and demanding manufacturing environments, it is necessary to develop methods for analyzing robotic health data, identifying faults, and determining root causes of faults before production-compromising failures occur. This project investigated a robot arm dataset that included both expected and actual values for the joint positions of a robot arm. The dataset was analyzed using Principal Component Analysis guided k-means clustering and fuzzy c-means clustering. The operational region most prone to joint position errors and degradation was discovered without prior engineering knowledge or error data. This paper presents a technique that can be built upon or combined with other models to enable fault detection within highly complex systems using generalized, unsupervised methods..