Real-time diagnosis and prognosis with sensors of uncertain quality

This work presents a real-time approach to the detection, isolation, and prediction of component failures in large-scale systems through the combination of two modules. The modules themselves are then used in conjunction with an inference engine, TEAMS-RT, which is part of Qualtech Systems integrated diagnostic toolset, to provide the end user with accurate diagnostic and prognostic information about the state of the system. The first module is a filter used to "clean" observed test results from multiple sensors from system noise. The sensors have false alarm and missed detection probabilities that are not known a-priori, and must be estimated - ideally along with the accuracies of these estimates - online, within the inference engine. Further, recognizing a practical concern in most real systems, a sparsely instantiated observation vector must not be problematic. Multiple hypothesis tracking (MHT) is at the heart of the filtering algorithm and beta prior distributions are applied to the sensor errors. The second module is a prognostic engine that uses an interacting multiple model (IMM) approach to track the "trajectory" of degrading sensors. Kalman filters estimate the movement in each dimension of the sensors. The current state and trajectory of each sensor is then used to predict the time to failure value, i.e., when the component corresponding to the sensor is no longer usable. The modules are integrated together and as part of the TEAMS-RT suite; logic is presented for the cases that they disagree.