Sensor unit with on-device learning and anomaly detection
AssigneeSTMicroelectronics International N.V.
InventorsFederico RIZZARDINI, Lorenzo BRACCO
Life of this patent
Estimating the term…
Abstract
A sensor unit is coupled to a machine and configured to detect anomalous behavior of the machine. The sensor unit includes a low power microcontroller that learns to recognize a plurality of operations of the machine. The sensor unit generates mean vector and inverse of a Cholesky decomposition matrix for each operation. During a detection mode the sensor unit computes a Mahalanobis distance for each feature vector, mean vector and first matrix. The sensor unit detects anomalous behavior or classifies the operation of the machine based on the Mahalanobis distances.
Claims20
A method, comprising:
- receiving, at a sensor unit coupled to a machine, a request to train the sensor unit to recognize a first operation of the machine;
- generating first sensor data with a sensor of the sensor unit while the machine performs the first operation;
- generating a mean vector based on the first sensor data;
- generating a first matrix corresponding to a covariance matrix based on the first sensor data;
- and
- generating a second matrix corresponding to an inverse of a Cholesky decomposition of the first matrix.
A method, comprising:
- generating, with a sensor unit coupled to a machine, a respective first matrix for each of a plurality of operations of the machine;
- generating sensor data with the sensor unit;
- computing a respective Mahalanobis distance value for each feature vector and the corresponding mean vector and first matrix;
- and
- determining that the machine is operating in an anomalous state if all of the Mahalanobis distance values are greater than a threshold value.
A sensor unit, comprising:
- a plurality of sensors;
- and
- a sensor processing unit coupled the plurality of sensors and configured to operate in a learning mode and in a detection mode, wherein:
- in the learning mode the sensor processing unit is configured to generate a mean vector and a first matrix based on first sensor data received from plurality of sensors while a machine performs a first operation;
- and
- in the detection mode the sensor processing unit is configured to receive second sensor data from the plurality of sensors and compute a squared Mahalanobis distance based on the feature vector, the mean vector and the first matrix.
Dependencies are inferred from the claim text.