Page 64 - IJAMD-2-2

International Journal of AI for
Materials and Design
A unified industrial AI foundation framework

These include the lack of systematic domain-specific basis of domain knowledge, which contains accumulated
knowledge integration, a tendency to focus on theoretical experience, research insights, and best practices from
conceptualization without clear and objective guidelines, AI-driven approaches. With the rise of IoT technology,
and an overemphasis on data and models without sufficient things, including machines, sensors, and other devices,
attention to systematic thinking that connects knowledge, generate vast amounts of data, which is an essential
data, and models. Motivated by these gaps, the following prerequisite for conducting industrial AI analytics,
two sections revisit the role of industrial AI and present modeling, and decision-making in modern industry.
a unified industrial AI foundation framework aimed at Moreover, systems such as CPS, IIoT systems, and other
addressing these critical shortcomings. intelligent industrial systems use AI models to automate
operations, enable intelligent decision-making, and
3. Rethinking the role of industrial AI improve real-time control.
The role of industrial AI can be best understood through While the two-layer perspective provides a conceptual
a two-layer perspective, as shown in Figure 1, where the understanding of how industrial AI bridges the physical
physical layer consists of humans, things, and systems, and digital domains, real-world industrial settings present
while the digital layer consists of knowledge, data, and several challenges. In practice, developers and engineers
models. Based on recent emerging technologies such face persistent difficulties in transforming heterogeneous
as IIoT, digital twin, 18,19 CPS, 13,14 industrial big data data from IIoT devices into reliable, AI-ready formats due
5,6
analytics, 11,12 deep learning, and LLMs, industrial AI to inconsistent data quality, lack of standard preprocessing
7,8
1,2
acts as a dynamic bridge between the physical and digital pipelines, and fragmented metadata management. CPS,
5
layers, continuously refining and applying AI-driven although widely discussed, often remains difficult to
insights to improve and optimize the real-world industrial operationalize, as maintaining consistency between real-
systems. More specifically, human expertise forms the world operations and virtual models requires continuous
data synchronization and robust model updating.
13
Despite its promise, digital twin development is challenged
by the complexity of integrating multi-source data
streams, real-time analytics, and simulation modeling
requirements. Furthermore, the increasing availability
18
of LLMs introduces new opportunities but also pose
practical challenges in domain knowledge understanding,
extraction, and interpretability for industrial use cases.
7
Based on our observations from industry collaborations
and project experiences, these complexities frequently
result in ad-hoc solutions, isolated development efforts, and
inefficiencies in scaling AI-driven applications. Therefore,
these observations and challenges motivate the need for
a structured foundation framework that systematically
connects knowledge, data, and models, as presented in the
following section.
4. Industrial AI foundation framework
In this section, we propose a unified industrial AI
foundation framework designed to systematically guide the
development and deployment of industrial AI solutions,
as illustrated in Figure 2. The framework consists of three
core modules: (i) knowledge module; (ii) data module;
and (iii) model module. Within each module, we identify
four key components, resulting in 12 important aspects
that collectively provide structured direction for industrial
AI developers and practitioners. These modules enhance
and extend the industrial AI methodology platform by
Figure 1. The role of industrial artificial intelligence in bridging the providing a structured, modular foundation for existing
physical and digital layers. Image created by the authors. and emerging industrial AI methodologies. Figure 2

Volume 2 Issue 2 (2025) 58 doi: 10.36922/IJAMD025080006