System to determine load cell failure
AssigneeAmazon Technologies, Inc.
InventorPaul Eugene Munger
Life of this patent
Estimating the term…
Abstract
Shelves or other fixtures may be used to support items at a facility. Load cells at the fixtures may be used to acquire weight data indicative of changes to the fixture as items are added or removed from the fixture. The load cells may occasionally fail to provide accurate information. Described are techniques for comparing the data from a first sensor with data from other sensors to determine if the first sensor has failed. A linear regression may be performed using the data from the other sensors, and the results compared with the output from the first sensor. If the difference exceeds a threshold amount, an error in the first sensor may be determined. Once determined, other actions may be taken, such as generating a repair ticket to fix the first sensor, disregarding weight from that sensor, and so forth.
Claims20
A system comprising:
- a shelf having four corners;
- a plurality of load cells comprising:
- a first load cell configured to measure weight associated with the shelf proximate to a first corner and generate first channel data;
- a second load cell configured to measure weight associated with the shelf proximate to a second corner and generate second channel data;
- a third load cell configured to measure weight associated with the shelf proximate to a third corner and generate third channel data;
- and
- a fourth load cell configured to measure weight associated with the shelf proximate to a fourth corner and generate fourth channel data;
- and
- a computing device comprising:
- a memory storing first computer-executable instructions;
- and
- a hardware processor to execute the first computer-executable instructions to:
- acquire a first time series comprising channel data from the plurality of load cells over a period of time;
- apply a Gaussian filter and an absolute value function to elements of the first time series to generate a filtered time series;
- generate, using the filtered time series, a second time series that omits data based on the first channel data, wherein elements of the second time series include the channel data obtained from before and after a specified time;
- perform a linear regression on the filtered time series to determine regression data;
- determine an error value based on a difference between the regression data and the first channel data obtained at the specified time;
- determine the error value exceeds a first threshold value;
- generate data indicative of an error of the first load cell;
- and
- perform a mitigation action in response to detecting the error of the first load cell, wherein the mitigation action comprises one or more of:
- disregarding weight data from the first load cell,
- deactivating the first load cell, or
- generating a work order to replace the first load cell.
A method comprising:
- detecting an event involving a fixture;
- responsive to the detecting the event involving the fixture, accessing weight data indicative of a weight of at least a portion of the fixture, wherein the weight data is obtained from a plurality of load cells of the fixture;
- applying a filter to the weight data to generate filtered data;
- generating, from the filtered data, first data that omits the weight data from a first one of the plurality of load cells;
- generating regression data using the first data;
- determining a best fit value using the regression data;
- determining error data based on the best fit value and the weight data from the first one of the plurality of load cells;
- determining a failure of the first one of the plurality of load cells based on the error data;
- and
- responsive to the determining the failure of the first one of the plurality of load cells, designating the weight data obtained from the first one of the plurality of load cells as erroneous.
A system comprising:
- a computing device comprising:
- a memory storing first computer-executable instructions;
- and
- a hardware processor to execute the first computer-executable instructions to:
- access weight data obtained by a plurality of load cells;
- apply a filter to the weight data to generate filtered data;
- generate regression data using the filtered data associated with a first subset of load cells;
- determine, using the filtered data associated with a second subset of load cells, a best fit value using the regression data;
- determine error data based on the best fit value;
- determine a failure of one of the plurality of load cells based on the error data;
- and
- responsive to the determination of the failure of the one of the plurality of load cells, disregard the weight data obtained by the one of the plurality of load cells.
Dependencies are inferred from the claim text.