Vision-Based Motion Process Monitoring Using Bayesian Fault Diagnosis

In this paper, the sparse optical flow (SOF), in form of three-way array, is used as the feature data of principal component analysis (PCA). The characteristics of the retained loading vectors can be demonstrated through denotative matrix. Then multisubspace division algorithm is employed based on the similarity of the loading vectors. Because current status is influenced by finite elapsed time behavior, the probabilistic statistic and the control limit is defined to monitor the motion process in each divided subspace. Finally, the contribution plot of the motion related variables is used for labeling the specific area of the frame that arouse the fault warning. The vision-based fault warning can help the operator directly, which obviously outperforms the traditional warning curve method. Case study of robotic-arm based marking system (RABMS) is taken to evaluate the performance of the proposed methods. The results demonstrate the capability of the proposed methods.