US 12,585,980 B2 · Annotation of time series data and validation for generating machine learning models

Abstract

Conventionally, applying analytics on dataset is the scarcity of labelled data. With increase of data there is cost fact effecting nature of servicing required for data (e.g., cost in terms of resource and time and effort is high for data annotation). Though data is analysed, it may be prone to error. Present disclosure provides systems/methods for reducing volume of data to be annotated for time series data thereby reducing time and effort of resources, thus resulting in effective utilization of system's resources (e.g., memory, processor, etc.). More specifically, the method of the present disclosure adaptively modifies the volume of the data to be annotated based on the performance of the unsupervised learning method applied in the system. Moreover, in the absence of an annotation mechanism for clusters of time series data, meta data associated with the time series data is utilized for annotation and validation of dataset.

A processor implemented method for annotating time series data and validating thereof for generating machine learning models, the processor implemented method comprising:

obtaining, at an edge device, a first set of time series data captured by one or more sensors attached to one or more equipment deployed in an industry, wherein the first set of time series data is captured for a first time interval, wherein the one or more equipment include a Friction Stir Welding (FSW) machine or an IoT device, wherein the one or more sensors attached to the one or more equipment include a force sensor, a power sensor or a torque sensor and the first set of time series data from the one or more sensors include values of force, torque, power, revolutions per minute (rpm), velocity, and ultimate tensile strength (UTS), and wherein the one or more sensors are either an integral part of the FSW machine or the IoT device or externally connected to the FSW machine or the IoT device;
applying, at the edge device, a clustering technique on the first set of time series data to obtain a first set of clusters of the first set of time series data, wherein each of a sensor value in the first set of clusters of the first set of time series data is associated with a cluster label;
transmitting, by the edge device, at least a subset of the first set of clusters of the first set of time series data to a cloud based on a corresponding cardinality assigned to each of the clusters from the first set of clusters in a percentage value, wherein the cluster of time series data with one of a low cardinality, a medium cardinality or a high cardinality is transmitted to the cloud for annotation or label correction, and wherein the clusters from the first set of clusters have a cardinality either above or below a pre-defined or empirically determined threshold based on the arrival rate, arrive time, and historical pattern of the time series data, wherein the transmission of the subset of the first set of clusters of the first set of time series data from the edge device to the cloud by reducing bandwidth usage;
obtaining, at the edge device from the cloud, a first validation of at the subset of the first set clusters of the first set of time series data, wherein the step of obtaining the first validation of at least the subset of the first set of clusters of the first set time series data comprising obtaining the first validation of the cluster label associated with each of the sensor value in the first set of clusters of the first set time series data, wherein the first validation of the cluster label associated with each of the sensor value in the first set of clusters of the first set time series data is indicative of an error rate, wherein the first validation of the subset of the first set clusters of the first set of time series data comprises a corrected label for each of the cluster, and wherein a machine learning (ML) technique is applied on the validated subset of the first set of clusters of the first set of time series data to generate a model file;
obtaining, at the edge device, a second set of time series data captured by the one or more sensors attached to the one or more equipment deployed in the industry, wherein the second set of time series data is captured for a second time interval;
applying the clustering technique on the second set of time series data to obtain a second set of clusters of the second set of time series data, wherein each of the sensor value in the second set of clusters of the second set of time series data is associated with a cluster label;
obtaining a second validation of at least a subset of the second set of clusters of the second set of time series data, wherein the step of obtaining the second validation of at least the subset of the one or more clusters of the second set of time series data comprising obtaining a second validation of the cluster label associated with each of the sensor value in the one or more clusters of the second set of time series data, and wherein selection and validation of the subset of the second set of clusters of the second set of time series data is based on the error rate associated with the first validation of the subset of the first set of clusters of the first set of time series data, wherein the error rate refers to an expected level of accuracy that is set in the edge device, wherein the error rate is represented as an integer or a value or a percentage, wherein an amount of the second set of time series data to be transmitted to the cloud by the edge device is decided based on a number of the corrected label for each of the first set of clusters of the first set of time series data, wherein the first validation and the second validation is performed by at least one of the cloud, a domain expert and based on metadata indicative of an operating status of the one or more sensors, and wherein the metadata comprises log records of each value of the one or more sensors and associated characteristics;
learning, by the edge device, a pattern of the label correction and applying the learnings on the second set of time series data to reduce the error rate, wherein the edge device learns the label correction or the first validation even when the label correction or the first validation is performed by the cloud, wherein the learnings of the first validation enables the edge device to automatically decide whether the second time series data from second set of clusters of the second time series data has to be transmitted from the edge device to the cloud, wherein the edge device performs the second validation by applying the learnings from previous iteration without having to transmit the second set of clusters of the second time series data to the cloud for annotation, and wherein the automatic decision enables the edge device to optimize utilization of resources including a memory, a processor of the edge device for further processing;
training, by the edge device, a model using at least one ML technique based on the first validation and the second validation to automatically decide whether the first set of time series data and the second set of time series data has to be transmitted from the edge device to the cloud, wherein the at least one ML technique comprises (i) un-supervised ML technique, (ii) a semi-supervised ML technique, and (iii) a supervised ML technique;
classifying the second set of clusters of the second set of time series data using the model file;
and
applying an ensemble of the supervised ML technique and the un-supervised ML technique in the edge device for new incoming time series data and iteratively performing the steps of applying the clustering technique on the first set of time series data to obtain the first set of clusters till obtaining the second validation, results in generating new set of results enabling automatic annotation of the new incoming time series data by the edge device.

9 more elements

Annotation of time series data and validation for generating machine learning models

Life of this patent

Annotation of time series data and validation for generating machine learning models

Life of this patent

Abstract

Claims9