SELF-OPTIMIZING MANUFACTURING SYSTEMS BASED ON ARTIFICIAL INTELLIGENCE AND REAL-TIME PRODUCTION ANALYTICS

Abstract

Statistical process control extended with rule-based adaptive adjustment remains the dominant response to production drift in most discrete-manufacturing lines, yet its single-parameter, single-objective logic leaves substantial performance on the table wherever defect rate, cycle time, and energy consumption respond jointly and nonlinearly to the same set of process inputs. This study proposes Closed-Loop Autonomous Process Optimization (CLAPO), a framework in which a continuously updated Gaussian-process surrogate model of the production process is queried by a multi-objective Bayesian optimization routine operating within safety-bounded parameter limits, with inline quality inference from vision and force sensors feeding the surrogate in real time and an autonomous decision layer committing parameter adjustments without requiring operator initiation, subject to override. The framework was evaluated on a representative injection-molding line through comparative modeling of three regimes: manual operator-set parameters, existing SPC-triggered rule-based adaptive control, and the proposed CLAPO system. The proposed system reduced defect rate from 4.8% to 1.2%, raised first-pass yield to 98.4%, lowered energy consumption per unit by 23%, and reduced operator interventions per shift from 22 to 3, while increasing overall equipment effectiveness from 64.2% to 84.6% and generating a five-year net present value nearly three times that of the standard rule-based approach.