Dharma, Fajar Pitarsi (2026) Pengembangan Model Quality Monitoring Di Multistage Manufaktur Tekstil Menggunakan Pre-trained Machine Learning Model. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 7010222003-Doctoral.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
7010222003-Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy |
Abstract
Industri tekstil Indonesia menghadapi tantangan kritis dalam pemantauan kualitas manufaktur tekstil multistage, dimana inspeksi manual tradisional menghasilkan accuracy rendah (50-70%), inkonsistensi tinggi, dan keterlambatan deteksi cacat yang berdampak pada peningkatan biaya produksi. Meskipun menjadi salah satu dari lima sektor prioritas Making Indonesia 4.0, penelitian integrasi machine learning untuk quality monitoring multistage masih sangat terbatas dengan hanya 7 dari 3.882 publikasi yang relevan terhadap konteks industri tekstil pertenunan Indonesia. Penelitian ini mengembangkan kerangka kerja pemantauan kualitas dengan strategi ganda yang mengintegrasikan model convolutional neural network untuk klasifikasi jenis cacat dan model deteksi objek untuk lokalisasi spasial cacat pada kain. Metodologi mencakup analisis bibliometrik komprehensif, evaluasi komparatif terhadap berbagai arsitektur deteksi pada dataset publik, pengembangan model klasifikasi dengan mekanisme attention hierarkis pada tahap greige hingga finishing, dan optimasi hyperparameter sistematis untuk meningkatkan kinerja deteksi. Hasil penelitian menunjukkan bahwa model klasifikasi yang dikembangkan mencapai accuracy 94% dalam mengidentifikasi jenis cacat, melampaui model baseline sebesar 22 poin persentase. Model deteksi teroptimasi mencapai peningkatan mean average precision sebesar 17% dan precision 22% dibandingkan konfigurasi dasar, dengan kinerja superior terhadap state-of-the-art methods yang tersedia. Validasi statistik dengan multiple random seeds menunjukkan reprodusibilitas dan stabilitas model yang tinggi. Kontribusi penelitian mencakup tiga novelty utama: (1) mekanisme attention hierarkis untuk integrasi feature multiscale yang adaptif terhadap variation across manufacturing stages (greige, dyeing, printing, finishing), memungkinkan pre-trained model untuk generalize tanpa stage-specific retraining (2) paradigma optimasi yang memprioritaskan konfigurasi model sebelum ekspansi kapasitas arsitektur, dan (3) kerangka kerja pemilihan model tiga dimensi yang mengintegrasikan metrik teknis dengan kendala operasional industri. Hasil penelitian telah divalidasi melalui lima publikasi peer-reviewed dan memberikan roadmap praktis untuk transformasi digital quality monitoring industri tekstil Indonesia selaras dengan Making Indonesia 4.0.
=============================================================================================================================
The Indonesian textile industry faces critical challenges in multistage manufacturing quality monitoring, where traditional manual inspection yields low accuracy (50-70%), high inconsistency, and delayed defect detection that significantly increases production costs. Despite being one of five priority sectors in Making Indonesia 4.0, research on machine learning integration for multistage quality monitoring remains severely limited, with only 7 of 3,882 publications relevant to the Indonesian textile weaving industry context. This research develops a dual-strategy quality monitoring framework integrating convolutional neural network models for defect type classification and object detection models for spatial defect localization on fabric. The methodology encompasses comprehensive bibliometric analysis, comparative evaluation of multiple detection architectures on public datasets, development of classification models with hierarchical attention mechanisms across greige-to-finishing stages, and systematic hyperparameter optimization to enhance detection performance. Results demonstrate that the developed classification model achieves 94% accuracy in identifying defect types, surpassing the baseline model by 22 percentage points. The optimized detection model achieves 17% improvement in mean average precision and 22% improvement in precision compared to baseline configuration, with superior performance against available state-of-the-art methods. Statistical validation using multiple random seeds confirms high model reproducibility and stability. Research contributions include three key novelties: (1) Hierarchical attention mechanism for multiscale feature integration that is adaptive to variation across manufacturing stages (greige, dyeing, printing, finishing), allowing pre-trained models to generalize without stage-specific retraining, (2) optimization paradigm prioritizing model configuration before architectural capacity expansion, and (3) three-dimensional model selection framework integrating technical metrics with industrial operational constraints. Research findings have been validated through five peer-reviewed publications and provide a practical roadmap for digital transformation of quality monitoring in Indonesian textile industry aligned with Making Indonesia 4.0.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | smart manufacturing, quality monitoring, multistage manufacturing, machine learning, industri tekstil, Industry 4.0, fabric defect detection, YOLO, CNN, hyperparameter optimization smart manufacturing, quality monitoring, multistage manufacturing, machine learning, industri tekstil, Industry 4.0, fabric defect detection, YOLO, CNN, hyperparameter optimization |
| Subjects: | Q Science Q Science > Q Science (General) Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TS Manufactures T Technology > TS Manufactures > TS156 Quality Control. QFD. Taguchi methods (Quality control) |
| Divisions: | Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26001-(S3) PhD Thesis |
| Depositing User: | Fajar Pitarsi Dharma |
| Date Deposited: | 02 Feb 2026 06:52 |
| Last Modified: | 02 Feb 2026 06:52 |
| URI: | http://repository.its.ac.id/id/eprint/131718 |
Actions (login required)
 |
View Item |