This paper introduces a four-stage, human-centered process model for robust monitoring of process curves in industrial manufacturing, aiming to advance zero-defect production. Leveraging unsupervised machine learning, the model combines established industrial quality assurance tools, such as Failure Modes and Effects Analysis (FMEA), with domain expertise to detect process anomalies and optimize decision-making. Using a dataset of 72,460 annotated process curves from end-of-line tests, the model demonstrates enhanced detection of false positives and novel defect classes while minimizing information loss inherent in traditional evaluation methods. Key innovations include the use of hierarchical density-based clustering (HDBSCAN) and principal component analysis (PCA) for feature extraction and clustering in latent spaces. Results reveal improved process drift detection and actionable insights for quality improvement, underscoring the indispensable role of human experts in conjunction with AI systems. This approach provides a scalable, adaptable framework for real-time monitoring and continuous optimization in dynamic industrial environments. Future work focuses on extending the model for real-time inference to support adaptive manufacturing processes.