Pengembangan dan Evaluasi Model Generative AI Berbasis GPT Untuk Menghasilkan Rekomendasi Intervensi Ergonomis
Rheina Khaisa Rhehani Putri, Ir. Ardiyanto, S.T., M.Sc., Ph.D., AEP., IPM.; Ir. Achmad Pratama Rifai, S.T., M.Eng., Ph.D.
2026 | Tesis | S2 Teknik Industri
Gangguan otot rangka (Gotrak) merupakan salah satu masalah
kesehatan kerja yang umum terjadi pada sektor Usaha Mikro, kecil, dan Menengah
(UMKM) akibat aktivitas kerja manual yang melibatkan postur tidak ergonomis, gerakan
berulang, dan durasi kerja yang panjang. Kondisi ini dapat menurunkan
produktivitas serta kualitas hidup pekerja. Meskipun intervensi ergonomis
terbukti efektif dalam mengurangi risiko Gotrak, penerapannya masih terkendala
oleh keterbatasan akses terhadap ahli ergonomi, biaya yang tinggi, dan
keterbatasan sumber daya. Oleh karena itu, penelitian ini bertujuan untuk
mengembangkan solusi berbasis Generative Artificial Intelligence (GAI) menggunakan
Large Language Model (LLM) berbasis Generative Pre-trained
Transformer (GPT) untuk menghasilkan rekomendasi intervensi ergonomis
secara otomatis dan kontekstual.
Sistem ini dirancang untuk menghasilkan rekomendasi intervensi ergonomis secara otomatis dan kontekstual berdasarkan deskripsi teks dan gambar mengenai situasi kerja. Pengembangan dilakukan menggunakan GPT Builder dengan knowledge base ergonomis yang kemudian diuji menggunakan studi kasus UMKM. Evaluasi dilakukan oleh tujuh evaluator terhadap rekomendasi yang dihasilkan oleh model custom GPT, Basic GPT, dan Pakar Ergonomika berdasarkan tujuh aspek, yaitu accuracy, completeness, inclusion of rationale, provision of citation, human-likeness, readability, dan satisfaction. Data kuantitatif dianalisis menggunakan uji Kruskal-wallis dan uji Dunn’s post hoc comparisons, sedangkan data kualitatif dianalisis menggunakan analisis tematik dengan inductive coding. Selain itu, tiga studi kasus tambahan digunakan untuk mengevaluasi akurasi antar model GPT dengan menggunakan metode BERTScore, serta validitas referensi dianalisis secara manual.
Musculoskeletal disorders (MSDs) are among the most
common occupational health problems in the Micro, Small, and Medium Enterprise
(MSME) sector, resulting from manual work activities involving non-ergonomic
postures, repetitive movements, and prolonged working durations. These
conditions can reduce workers’ productivity and quality of life. Although
ergonomic interventions have been proven effective in reducing the risk of
MSDs, their implementation remains constrained by limited access to ergonomics
experts, high costs, and resource limitations. Therefore, this study aims to
develop a Generative Artificial Intelligence (GAI)-based solution using a Large
Language Model (LLM) based on Generative Pre-trained Transformer (GPT) to
generate ergonomic intervention recommendations automatically and contextually.
The system was designed to generate ergonomic
intervention recommendations automatically and contextually based on textual
and visual descriptions of work situations. The development process was
conducted using GPT Builder with an ergonomics knowledge base and subsequently
evaluated using MSME-related case studies. The evaluation involved seven
evaluators who assessed recommendations generated by the custom GPT, Basic GPT,
and Ergonomics Expert models based on seven aspects: accuracy, completeness,
inclusion of rationale, provision of citation, human-likeness, readability, and
satisfaction. Quantitative data were analyzed using the Kruskal–Wallis test and
Dunn’s post hoc comparisons, while qualitative data were analyzed using
thematic analysis with inductive coding. In addition, three supplementary case
studies were used to evaluate accuracy between GPT models using the BERTScore
method, and reference validity was analyzed manually.
The results showed significant differences in the
inclusion of rationale and readability variables, while the other five
variables showed no significant differences among the models. Dunn’s post hoc
comparisons indicated that the custom GPT differed significantly from the
Ergonomics Expert in the inclusion of rationale variable and differed
significantly from the Ergonomics Expert in the readability variable together
with the Basic GPT. The BERTScore results showed that the custom GPT was
slightly more accurate than the Basic GPT (F1 = 0.6004), with superior
precision, whereas the Basic GPT demonstrated superior recall. Reference
validation showed that the custom GPT had the highest proportion of correct
references (91.30%) compared with the Ergonomics Expert (85.00%) and the Basic
GPT (57.69%), and it generated no hallucinated references. Qualitative analysis
identified five main themes: quality of ergonomic recommendations, quality of
explanations and justifications, quality of language and presentation, quality
of scientific references, and suitability for the MSME context. These findings
indicate that generative AI has the potential to serve as an effective support
tool for developing ergonomic recommendations in MSMEs, although further
evaluation using more comprehensive approaches is still required.
Kata Kunci : Generative AI, GPT, Rekomendasi Intervensi, Ergonomis