Non-invasive diagnostic systems and methods for using the same
AssigneeMiami University
InventorsMark Scott, Matt Boubin
Life of this patent
Estimating the term…
Abstract
A method of measuring electromagnetic interference (EMI) to noninvasively identify component degradation or failure in power electronics circuitry. The method involves characterizing the degradation or failure characteristics of the component and modeling those characteristics to enable a machine learning algorithm to identify EMI frequency distribution characteristics that correspond to the degradation or failure. The EMI frequency distribution is measured and the data provided to the machine learning algorithm whereupon the algorithm identifies degradation or failures indicated by the measured data.
Claims20
A method for determining whether a circuit fault exists in one or more components of a power electronic device comprising:
- modeling a component of the power electronic device;
- modeling one or more a plurality of failure modes for the component;
- identifying an electromagnetic interference (EMI) characteristic of the component, wherein the EMI characteristic is indicative of each of the plurality of failure modes;
- training a machine learning algorithm to detect each of the plurality of modeled failure modes based on the EMI characteristic;
- monitoring the EMI of the component;
- analyzing the monitored EMI via the machine learning algorithm to detect each of the plurality of modeled failure modes;
- and
- providing feedback that no faults exist if none of the plurality of modeled failure modes are detected.
A method of detecting whether a circuit is fault-free faults comprising:
- identifying a component;
- modeling a plurality of failure modes of the component;
- measuring the electromagnetic frequency spectrum produced by the circuit when the component is experiencing each of said failure modes;
- training a machine learning algorithm to detect the modeled failure modes using the measured electromagnetic frequency spectrum;
- recording the electromagnetic frequency spectrum of the circuit;
- analyzing the recorded electromagnetic frequency spectrum using the machine learning algorithm to detect each of the modeled failure modes of the component;
- and
- determining that no faults exist if none of the failure modes are detected.
A non-destructive testing system for power electronics:
- an electromagnetic interference (EMI) filter having a frequency sensing element and a diagnostic system, wherein the diagnostic system includes:
- a processing circuit operably connected to a memory having programmable instructions stored thereon, wherein the programmable instructions, when executed by the processing circuit, causes the diagnostic system to:
- model a component of the power electronic device;
- model a plurality of failure modes for the component;
- identify, for each of the plurality of failure modes, an (EMI) characteristic of the component, wherein the EMI characteristic is indicative of the respective modeled failure mode;
- train a detector via a machine learning algorithm to detect each of the plurality of modeled failure modes based on its indicative EMI characteristic;
- monitoring the EMI of the component;
- analyzing the monitored EMI to detect each of the plurality of modeled failures mode;
- and
- indicate to a user that the component is without fault if none of the plurality of modeled failure modes are detected.
Dependencies are inferred from the claim text.