Tjahjaningsih, Yustina Suhandini (2026) Integrasi Pengendalian Kualitas Berbasis AI dengan Pendekatan Human-Centric menuju Zero Defect Manufacturing. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 7010231003_Doctoral.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
7010231003_Doctoral.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy |
Abstract
Zero Defect Manufacturing (ZDM) merupakan paradigma peningkatan kualitas yang dikembangkan untuk mengatasi keterbatasan metode tradisional seperti Six Sigma, Lean, dan Total Quality Management (TQM) dalam menghadapi kompleksitas sistem manufaktur di era Industri 4.0. Penelitian ini bertujuan mengembangkan dan memvalidasi model ZDM berbasis Artificial Intelligence (AI) dengan pendekatan human-centric yang mengintegrasikan empat strategi utama ZDM, yaitu deteksi (detection), prediksi (prediction), perbaikan (repair), dan pencegahan (prevention), ke dalam satu kerangka implementatif untuk pengendalian kualitas di industri manufaktur. Sebagai bentuk implementasi dan validasi lapangan, model diwujudkan dalam bentuk prototipe sistem berbasis mobile (Surface Defect Detection/SDD) untuk mendukung proses inspeksi dan rekomendasi tindakan mutu secara terintegrasi. Pada strategi deteksi, dikembangkan model Convolutional Neural Network (CNN) dengan ResNet50V2 sebagai backbone utama untuk klasifikasi surface defect melalui tahapan preprocessing, augmentasi data, pelatihan, evaluasi menggunakan confusion matrix, serta konversi ke format TensorFlow Lite (TFLite) agar dapat dioperasikan pada perangkat mobile. MobileNetV2 digunakan secara terbatas pada studi kasus corrugated asbestos sheet sebagai uji kelayakan awal inspeksi berbasis CNN. Pada dataset gabungan lintas produk, hasil evaluasi menunjukkan bahwa model CNN berbasis ResNet50V2 mencapai akurasi pengujian 97,66% dengan F1-score 0,9766 serta average confidence 0,9866, yang mengindikasikan kemampuan generalisasi dan konsistensi prediksi yang tinggi untuk mendukung inspeksi kualitas di lingkungan industri. Strategi prediksi direalisasikan melalui penyusunan Defect Tracking Matrix (DTM) yang memetakan hubungan atribut teknis–jenis cacat–tahapan proses, disusun dengan prosedur DTM dan diintegrasikan ke basis data untuk mendukung pelacakan pola defect dan analisis prediktif. Strategi perbaikan dan pencegahan diformulasikan menggunakan Failure Mode and Effect Analysis (FMEA) untuk mengidentifikasi mode kegagalan, penyebab, serta rekomendasi tindakan berbasis risiko. Seluruh luaran empat strategi diintegrasikan dalam satu sistem agar rekomendasi yang dihasilkan konsisten dengan konteks proses dan risiko. Validasi sistem dilakukan melalui pengujian fungsional (blackbox testing), serta User Acceptance Test (UAT). Hasil UAT menunjukkan tingkat penerimaan 86% (kategori “Sangat Setuju”), menegaskan bahwa implementasi model dinilai mudah digunakan dan relevan bagi kebutuhan inspeksi kualitas. Pendekatan human-centric diterapkan dengan menempatkan operator dan quality control sebagai validator dan pengambil keputusan akhir, sementara AI berperan sebagai alat bantu cerdas yang memperkuat konsistensi inspeksi dan rekomendasi mutu. Implementasi pada industri particle board menunjukkan bahwa model yang dikembangkan mendukung pengendalian kualitas yang lebih sistematis, adaptif, dan kolaboratif dalam mendukung pencapaian Zero Defect Manufacturing.
========================================================================================================================
Zero Defect Manufacturing (ZDM) is a quality improvement paradigm developed to address the limitations of traditional methods such as Six Sigma, Lean, and Total Quality Management (TQM) in dealing with the complexity of manufacturing systems in the Industry 4.0 era. This study aims to develop and validate an Artificial Intelligence (AI)–based ZDM model with a human-centric approach that integrates four main ZDM strategies—detection, prediction, repair, and prevention—into a single implementable framework for quality control in manufacturing industries. As a form of implementation and field validation, the model is realized as a mobile-based system prototype (Surface Defect Detection/SDD) to support integrated inspection processes and quality action recommendations. In the detection strategy, a Convolutional Neural Network (CNN) model is developed using ResNet50V2 as the main backbone for surface defect classification through preprocessing, data augmentation, training, evaluation using a confusion matrix, and conversion to the TensorFlow Lite (TFLite) format for deployment on mobile devices. MobileNetV2 is used in a limited manner for a corrugated asbestos sheet case study as an initial feasibility test of CNN-based inspection. On a combined cross-product dataset, the evaluation results show that the ResNet50V2-based CNN model achieves a testing accuracy of 97.66% with an F1-score of 0.9766 and an average confidence of 0.9866, indicating strong generalization capability and prediction consistency to support quality inspection in industrial environments. The prediction strategy is realized through the development of a Defect Tracking Matrix (DTM) that maps the relationships among technical attributes, defect types, and process stages. The DTM is constructed following a defined DTM procedure and integrated into a database to support defect pattern tracking and predictive analysis. The repair and prevention strategies are formulated using Failure Mode and Effect Analysis (FMEA) to identify failure modes, causes, and risk-based recommended actions. All outputs from the four strategies are integrated into a single system to ensure that the generated recommendations are consistent with process contexts and risk considerations. System validation is conducted through functional testing (black-box testing) and a User Acceptance Test (UAT). The UAT results show an acceptance level of 86% (classified as “Strongly Agree”), confirming that the implementation is considered easy to use and relevant to quality inspection needs. The human-centric approach is applied by positioning operators and quality control personnel as validators and final decision-makers, while AI functions as an intelligent support tool that enhances inspection consistency and quality recommendations. Implementation in the particle board industry demonstrates that the developed model supports more systematic, adaptive, and collaborative quality control in achieving Zero Defect Manufacturing
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | Artificial Intelligence (AI), Human-Centric Approach; Quality Control System, Zero Defect Manufacturing (ZDM). Artificial Intelligence (AI), Human-Centric Approach, Quality Control System, Zero Defect Manufacturing (ZDM). |
| Subjects: | T Technology > TS Manufactures > TS156 Quality Control. QFD. Taguchi methods (Quality control) |
| Divisions: | Faculty of Industrial Technology > Industrial Engineering > 26001-(S3) PhD Thesis |
| Depositing User: | Yustina Suhandini Tjahjaningsih |
| Date Deposited: | 02 Feb 2026 07:29 |
| Last Modified: | 02 Feb 2026 07:29 |
| URI: | http://repository.its.ac.id/id/eprint/131766 |
Actions (login required)
 |
View Item |