Published September 20, 2026 | Version v1.0.0

ForgeMind: an agentic predictive-maintenance pipeline, with the evaluation artifacts for "When Does a Large Language Model Earn Its Place in an Agentic Maintenance Pipeline?"

  • 1. Indian Institute of Management Calcutta

Description

ForgeMind is an open agentic pipeline for scenario testing on a digital twin of a five-machine electronics assembly line. An operator's free-text fault report is converted into a structured severity interface (sensor, severity band, magnitude), injected into a 50-step, 18-channel sensor window, passed to a convolutional-recurrent remaining-useful-life model, and turned into a machine status and a dispatch instruction.

This deposit archives the pipeline together with the artifacts behind the accompanying paper: the prompt banks, the recorded outputs of every language-model and classifier arm, the replayed predictions, the result tables, the pre-registered analysis plan, and the verification script that recomputes every number reported in the paper from the raw files.

Third-party data are not redistributed. The PRONOSTIA bearing recordings and the FMUCD work-order corpus must be obtained from their original sources; only the sampled record identifiers and the outputs produced from them are included here.

Files

vybhav72954/agent-factory-v1.0.0.zip

Files (18.4 MB)

Name Size
md5:86e0d060089f638739a934a498622ba0
18.4 MB Preview Download

Additional details

Related works