Toyota Motor Engineering & Manufacturing North America, Inc.

University of Connecticut

Donald Mcmenemy

Shailesh N Joshi

Krishna R Pattipati

Ali M Bazzi

Weiqiang Chen

John Kaminski

Ravi Rajamani

Muhamed K Farooq

Lingyi Zhang

A method, system, and non-transitory computer readable medium describing an autoencoder that creates a reduced feature space from healthy power electronics devices for training. Devices under test are then encoded and compared to the encoded features of the healthy devices to determine health of the other devices. Contextual information is used to build multiple models that compare power electronics devices from similarly operated vehicles with one another.

Autoencoder utilizing vehicle contextual information

Systems and methods of testing the health of vehicular power devices are disclosed herein. A method may include producing operating points as a function of cycling current (I) and voltage drain to source (V) when a subject device is conducting current. The method may further include determining a mean of moving distribution to adapt a center of the moving distribution contrasted with a plurality of known healthy devices. The method may also include indicating an imminent fault in the subject device based upon a discontinuity among operating points above a threshold.

Systems and methods for false-positive reduction in power electronic device evaluation

To predict a failure condition in a power module of a vehicle, it is determined whether a discontinuity in statistical data characterizing physical measurements of the power module meets a threshold criterion. Responsive to the discontinuity meeting the threshold criterion, a data offset in the physical measurements is computed at the discontinuity. A shift correction is applied to the physical measurements in accordance with the computed data offset responsive to a determination that the discontinuity is attributable to a restart of the power module. Other statistical data characterizing the shift-corrected physical measurements are computed and the statistical data and the other statistical data are provided to a machine learning processor that predicts the failure condition in the power module.

Mitigating data offsets for machine learning

Systems and methods for detecting a fault or model mismatch are disclosed. A system includes a processor, a memory, and one or more sensors. The sensors may detect data associated with an electronic device. The memory may store processor executable instructions to: compute Tand Q statistics, over a time period, and apply a model mismatch and fault detection logic based on the Tand Q statistics. The model mismatch and fault detection logic may: count consecutive instances where a Tstatistic exceeds a Tthreshold via a Tcounter, update a probability of fault based on the Tcounter, count consecutive instances where a Q statistic exceeds a Q threshold via a Q counter, update a probability of model mismatch based on the Q counter, and detect one of a fault or a model mismatch based on a probability of fault threshold and a probability of model mismatch threshold.

Growing community of inventors

Willington, CT, United States of America

Donald Mcmenemy