Optimizing Air-Handling-Unit Assembly Through Lean Six Sigma DMAIC Methodology: A Case Study in High-Mix Low-Volume Manufacturing

This study investigates the application of Lean Six Sigma (LSS) through the Define-Measure-Analyze-Improve-Control (DMAIC) framework to optimize assembly operations at a Portuguese air-handling-unit (AHU) manufacturer, a paradigmatic high-mix low-volume (HMLV) production environment characterized by substantial product configurational variability. A key methodological contribution is the development of a novel two-level Complexity Index, comprising the AHU Unit Complexity Index (ICUTA) and its order-level aggregate (ICOP), that enables normalized performance assessment through the metric h/ICOP (assembly hours per unit of order complexity). Calibrated against 46 baseline production orders (123 AHUs), the ICOP exhibited a Pearson correlation of r = 0.95 (p < 0.001) with actual assembly times; the coefficient of determination R2 = 0.90 indicates strong in-sample fit within the calibration dataset. The intervention targets, cycle time reduction ≥ 25%, variability ≥ 50%, rework ≥ 60%, and delivery ≥ 75%, were anchored in internal benchmarking using first-quartile baseline performance (1.93 h/ICOP). Baseline diagnostics revealed a mean assembly efficiency of 3.05 h/ICOP (coefficient of variation, CV = 53.1%), a rework rate of 41.3%, and an on-time delivery rate of 41.0%. Root cause analysis using Ishikawa diagrams identified critical deficiencies in methods, human resources, materials, and measurement systems. Targeted interventions, 5S workplace organization, material kitting, enhanced communication protocols, and visual management, yielded: an 11.15% reduction in h/ICOP (from 3.05 to 2.71), a 44.26% decrease in variability (CV from 53.1% to 29.5%), a 61.26% reduction in rework (from 41.3% to 16.0%), and a 23 percentage-point increase in on-time delivery (from 41.0% to 64.0%; +56.10% relative). A counterintuitive “simplicity trap” was identified, low-complexity AHUs exhibited poorer normalized performance than high-complexity variants, hypothesized to arise from reduced operator vigilance and lower process standardization for apparently simple tasks; this interpretation is theoretically informed and requires future empirical validation. Statistical process control charts confirm sustained process stability within the monitoring period. These findings support LSS-DMAIC as an effective framework for HMLV-process improvement, with the ICOP offering a potentially transferable normalization methodology pending external validation.