US 12,367,464 B2
Predictive device maintenance
Adam Lee Griffin, Dubuque, IA (US); Shikhar Kwatra, San Jose, CA (US); Matthew Alzamora, Poughkeepsie, NY (US); Patricia Wynne Mchann, Hernando, MS (US); Christopher Denis Hardt, Hyde Park, NY (US); and David Beltran, Beacon, NY (US)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Sep. 1, 2020, as Appl. No. 17/009,290.
Prior Publication US 2022/0067669 A1, Mar. 3, 2022
Int. Cl. G06Q 10/20 (2023.01); G06F 16/93 (2019.01); G06F 16/951 (2019.01); G06F 40/205 (2020.01); G06N 5/02 (2023.01); G06Q 30/0282 (2023.01); G16Y 10/75 (2020.01); G16Y 40/20 (2020.01)
CPC G06Q 10/20 (2013.01) [G06F 16/93 (2019.01); G06F 16/951 (2019.01); G06F 40/205 (2020.01); G06N 5/02 (2013.01); G06Q 30/0282 (2013.01); G16Y 10/75 (2020.01); G16Y 40/20 (2020.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
utilizing a trusted circle of Internet of Things (IoT) devices to support a circle of trust containing IoT devices by exchanging information within the circle of trust, wherein the IoT devices within the circle of trust function as peers in a blockchain environment such that they are able to acquire software images from the trusted circle of IoT devices, share software images to and from other IoT devices within the circle of trust, and protect software images from being divulged to an unauthorized party using a blockchain protocol;
periodically polling each of the Internet of things (IoT) devices that is registered to be utilized in the trusted circle of IoT devices to determine the operational status;
in response to determining that the IoT device is working correctly, collecting input data about the working device;
generating a risk score as a weighted measure of the collected input data and a frequency of failure of the IoT device, wherein the risk score indicates an estimate of future loss of function of the IoT device, wherein the risk score is generated by the following:
performing a Latent Dirichlet Allocation (LDA) analysis for topic modeling and extracting features from a document;
using a Bidirectional Encoder Representations from Transformers (BERT) model to extract a start and stop end span from a paragraph of the document;
using the BERT model to generate a sentiment from an output of the LDA analysis by computing measures or characterizations of input text using the extracted start and stop end span, wherein one or more measures or characterizations have a singular or multi-response which is associated with hidden topics and/or categories forming a data set; and
splitting the data set into a training set and a test set, wherein the training set is used train a model to calculate an initial risk score for a failure mode of the IoT device and/or components thereof, wherein the test set is used to update the initial risk score;
determining an optimal time to trigger a predictive notification pertaining to the IoT device using the risk score; and
at the determined optimal time, issuing the predictive notification about the working devices.