IJCA Vol 4 i1 2025 webmag - Flipbook - Page 20
20 The International Journal of Conformity Assessment
2023), while ICR can recognize
handwritten characters (Gunnoo,
2024). ICR enables handwriting
detection in digital documents
and converts it into editable text,
whereas OCR is limited to printed
content (Tang et al., 2024).
However, no PoC has yet examined
their use across multiple CoP
components under both ideal and
real-world production conditions.
This gap presents further
research potential for scientific
investigation.
The ICR model Generative
Pretrained Transformer 4 (GPT4v) represents a significant
breakthrough in AI by enabling
visual data in large language
models (LLMs) (Kaushik, 2024;
Microsoft Corporation, 2024).
Building upon GPT-3, it combines
advanced language with visual
content processing (Kaushik, 2024).
2.3 Research Questions and
Objectives
GPT-4v combines OCR capabilities
with AI and can therefore be
characterized as an ICR system.
It allows for accurate detection
of text in images—including
handwritten text—and converts it
into electronic format (Kaushik,
2024; Microsoft Corporation, 2024;
Olesia, 2023).
While GPT-4v does not perform
direct image processing on its
own, it can interpret and enhance
OCR results from computer
vision systems. Its multimodal
capabilities enable it to process
both text and images, generate
image descriptions, answer
questions about visual content,
and create images from text
prompts (Kaushik, 2024; Microsoft
Corporation, 2024).
Based on the identified research
needs (Sections 2.1 and 2.2), the
following research questions arise:
Component classification and PoC
for detection technologies:
1. Which detection technology is
suitable for each component of
the Conformity of Production
(CoP) process?
2. How can the CoP components
be effectively classified to align
with the appropriate detection
technology in PoCs?
Digitalization concept:
3. What type of detection
technology can be applied to
each specific component within
the CoP framework?
4 Component-ID Classification
The current CID process
encompasses over 300 different
components (BMW Group, 2021).
To support the PoCs presented
in Section 5, it is essential to first
determine suitable component
categories within this range.
Therefore, a classification system
will be developed to identify
representative components prior to
initiating the PoCs.
Section 4.1 defines the relevant
characteristics and classifies
the components accordingly.
The results of this classification
are presented in Section 4.2 for
application in subsequent PoCs.
4.1 Definition of Properties and
Classification of Components
To define appropriate component
categories, it is necessary to
establish relevant properties. CoP
experts were consulted to identify
these properties.
A total of 35 domain experts from
the BMW Group, representing
internal plants, participated in the
3 Methodical Structure of the evaluation. The materials of the
300 different CoP components
Contribution
were categorized into glass,
elastomers, thermoplastics,
Building upon previous findings
thermoset and metals. Defined
in the current state of the art, the
properties include materials
following methodical structure is
used, accessibility in the vehicle
used to answer the questions in
this paper, as illustrated in Figure 1. or complexity of integration,
To date, only a few scientific
PoCs have investigated GPT4v’s potential in identifying and
extracting information from
components or comparing
different materials (Shahriar et al.,
2024; Wu et al., 2023).
Conclusion for Relevant Research
Objective
Current research has demonstrated
PoCs for both transmitter-receiver
and optoelectronic systems.
Figure 1: Methodical structure and description of the approach
