The analytical model - parametric torque data
IBM® Plant Performance Analytics creates the predictive model using parametric data obtained from a robot.
The predictive model is used to calculate the time to fault and probability of fault for occurring from historical data.
Model input
The torque data inputs for the predictive model are: the actual torque values from the different axes of the robot, (J1, J2, J3, J4, J5, J6), the ROBOT_ID, GROUP_NUMBER, GROUP_AXES, GROUP_TYPE, GROUP_ID, and TIME Stamp.
Model output
Anomalies that are detected in the parametric data, the time to fault and the Probability of the fault occurring.
Methodology
The predictive machine learning model consists of two stages – training and scoring. The inputs for the training model consists of the torque values of the robot with other identifiers for differentiating the operation mode of the robot. The training model uses unsupervised learning techniques to process the torque data and categorize the record as an anomaly or not. The labeled preprocessed data uses a supervised learning technique to predict the likelihood of fault or probability of fault. Finally a scoring model is used to score a new record by predicting the likelihood of fault and time to fault from a given time instance.