US 12,461,952 B1
Systems and method for constructing troubleshooting trees
Lasitha Vidyaratne, Fremont, CA (US); Xian Yeow Lee, Santa Clara, CA (US); Aman Kumar, San Jose, CA (US); Tsubasa Watanabe, Spring Lake, MI (US); Dipanjan Ghosh, Santa Clara, CA (US); Ahmed Farahat, Santa Clara, CA (US); and Chetan Gupta, Sunnyvale, CA (US)
Assigned to Hitachi, Ltd., Tokyo (JP)
Filed by Hitachi, Ltd., Tokyo (JP)
Filed on Jun. 14, 2024, as Appl. No. 18/744,398.
Int. Cl. G06F 16/00 (2019.01); G06F 16/31 (2019.01); G06F 16/338 (2019.01); G06F 16/383 (2019.01); G06F 40/205 (2020.01)
CPC G06F 16/322 (2019.01) [G06F 16/338 (2019.01); G06F 16/383 (2019.01); G06F 40/205 (2020.01)] 20 Claims
OG exemplary drawing
 
1. A method for constructing an equipment troubleshooting tree, the method comprising:
at a preprocessing pipeline, receiving an unstructured text source comprising sections;
using a parsing tool to segment the unstructured text source to extract a listing of sections, each section being associated with at least one of a component of an equipment, a unit, or a process;
in response to determining that the parsing tool has not fully extracted the listing, performing steps comprising:
providing the unstructured text source to a pre-trained language model; and
instructing the pre-trained language model to identify each section, by using the listing of sections as a context;
for each section, performing steps comprising:
obtaining metadata comprising a section-specific text and a corresponding section title and page range associated with the unstructured text source; and
applying a non-static process, which enhances a coherency of a similarity matching, to preprocess the section to obtain segmented data;
using an indexing process that increases a retrieval efficiency in a troubleshooting tree constructing pipeline by embedding the segmented data and the metadata into index nodes;
using the index nodes in the troubleshooting tree constructing pipeline, which performs steps comprising, for each component:
using the pre-trained language model to generate a catalog comprising a fault, a cause, or repair recommendation; and
structuring at least some of the fault, a cause, or repair recommendation as nodes of the troubleshooting tree;
in response to at least one of the equipment, the unit, or the process being queried by a user query, generating the troubleshooting tree;
using the metadata in a grounding process that evaluates an accuracy of the troubleshooting tree, the grounding process comprising cross-referencing at least some of the troubleshooting tree with the unstructured text source; and
outputting the troubleshooting tree.