Dta-qc: an AI-driven framework for adaptive quality control and intelligent test optimization in 5 G manufacturing

Abstract In modern 5 G radio manufacturing, traditional quality control methods based on black fixed thresholds are increasingly inadequate, often failing to capture nuanced fault patterns and requiring substantial manual intervention. This study presents DTA-QC, an AI-driven framework for adaptive thresholding and intelligent test optimization in 5 G production environments. The proposed system introduces three core innovations: (1) dynamic thresholding using LSTM autoencoders and regression models to detect anomalies under evolving production conditions, (2) supervised fault classification via convolutional neural networks trained on time-windowed sensor data, and (3) a four-level severity classification system (Normal, Warning, Worse, Stop) to support real-time decision-making in manufacturing environments. DTA-QC is implemented and validated on Ericsson AB’s 5 G radio production line, achieving high anomaly detection accuracy (ROC-AUC: 0.89–0.94) and significantly reducing manual review efforts, without requiring specialized hardware. To assess generalizability, DTA-QC is further evaluated on a public benchmark dataset. A comparative analysis of three architectural variants revealed trade-offs in complexity, latency, and deployment feasibility. These results underscore the value of embedding AI-driven analytics in industrial test workflows, contributing to the broader goals of intelligent manufacturing and adaptive, data-driven quality assurance.