Amid the immense complexity of modern Industry 4.0 manufacturing, traditional Statistical Process Control (SPC) approaches, which are reliant on manually defined thresholds and expertise-driven control charts, are proving increasingly insufficient. The vast amount of data, continual evolution of products and processes and the necessity for timely, meaningful alerts have exposed key limitations in existing methods, including scalability, adaptability and transparency. This paper introduces a novel automated, AI-driven methodology called Self-learning AI-based Q-Rules (SLAQ) to overcome these challenges. The SLAQ methodology combines optimization techniques with reinforcement learning to dynamically generate and refine quality rules (Q-Rules) for anomaly detection in production data. Initial Q-Rules are derived from historical data via optimization algorithms. These rules are subsequently refined through reinforcement learning using human feedback to enhance detection performance and adaptability to address changing process conditions.