Large Language Model Agents as Prognostics and Health
Management Copilots
Sarah Lukens
1
, Lucas H. McCabe
1,2
, Joshua Gen
1
, Asma Ali
3
,
1
LMI, Tysons, VA, 22102, USA
sarah.lukens@lmi.org, lmccabe@lmi.org, joshua.gen@lmi.org
2
George Washington University, Washington, DC, 20052, USA
3
GE Vernova, Chicago, IL, 60661, USA
asma.ali1@ge.com
ABSTRACT
Amid concerns of an aging or diminishing industrial work-
force, the recent advancement of large language models
(LLMs) presents an opportunity to alleviate potential expe-
rience gaps. In this context, we present a practical Prog-
nostics and Health Management (PHM) workflow and self-
evaluation framework that leverages LLMs as specialized
in-the-loop agents to enhance operational efficiency without
subverting human subject matter expertise. Specifically, we
automate maintenance recommendations triggered by PHM
alerts for monitoring the health of physical assets, using
LLM agents to execute structured components of the standard
maintenance recommendation protocol, including data pro-
cessing, failure mode discovery, and evaluation. To illustrate
this framework, we provide a case study based on historical
data derived from PHM model alerts. We discuss require-
ments for the design and evaluation of such “PHM Copilots”
and formalize key considerations for integrating LLMs into
industrial domain applications. Refined deployment of our
proposed end-to-end integrated system may enable less ex-
perienced and professionals to back-fill existing personnel at
reduced costs.
1. INTRODUCTION
Industrial demographics have changed over time in several
domains, in part due to shifting occupational preferences,
shrinking generational cohorts, and lengthened professional
careers (Silverstein, 2008). Additionally, corporate finan-
cialization, technological change, and industrial outsourcing
Lukens Sarah et al. This is an open-access article distributed under the terms
of the Creative Commons Attribution 3.0 United States License, which per-
mits unrestricted use, distribution, and reproduction in any medium, provided
the original author and source are credited.
have left engineering organizations with numerous workforce
challenges that are not easily resolved by adapting hiring
practices alone (Muellerleile, 2009; Greenberg, 2010). As
a result, a so-called “experience gap” has caused concern in
operational fields (Rovaglio, Calder, & Richmond, 2012). In
particular, monitoring and maintenance of complex engineer-
ing systems typically requires the deployment of specialized
personnel with sophisticated domain expertise, and such staff
are in short supply. Although systemic approaches, such as
large-scale programs to increase vocational training access,
can be impactful, such strategies can be difficult for individ-
ual organizations to implement effectively. Instead, we con-
sider whether recent digital innovation - particularly that of
large language models (LLMs) - can help relieve these work-
force pressures by supplementing less experienced mainte-
nance and reliability professionals.
LLMs are (typically autoregressive) statistical models of to-
ken sequences, learned from large textual corpora (Chengwei
Wei and Yun-Cheng Wang and Bin Wang and C.-C. Jay
Kuo, 2024). In production, these models are often fine-tuned
for instruction-following (Ouyang et al., 2022), whereby
user-provided prompts induce a discrete distribution over
output sequences (Sordoni et al., 2024). These so-called
“instruction-tuned” models can serve as impressive con-
versational agents, but questions remain regarding effec-
tive application in industrial settings, including medicine
(Thirunavukarasu et al., 2023), design and manufacturing
(Makatura et al., 2023), and power engineering (Majumder
et al., 2024).
The so-called “copilot framework” - where artificial intelli-
gence (AI)-powered systems augment, rather than replace,
existing workflows - offers an opportunity to meaningfully
1
