Reliability improvement in CNC manufacturing through lean six sigma DMAIC: an empirical study on mean time between failures

Abstract Unplanned equipment failures at CNC manufacturing facilities create production delays that are difficult to schedule around, and published Lean Six Sigma DMAIC studies of Mean Time Between Failures (MTBF) rarely report variance reduction as an outcome in its own right, despite highly variable failure intervals being at least as disruptive to production planning as low average reliability. This study applies the DMAIC framework to MTBF at a Government Tool Room and Training Centre (GTTC) in Bengaluru, India, a public-sector precision manufacturing facility, using six months of maintenance work-order records (January–June 2025, 502 unplanned breakdown events). Because system-based failure tracking began at the start of this window with no prior failure history available, the first recorded failure for each of 52 machines could not be paired with a genuine previous failure and was excluded as a left-censored observation, yielding a final analysis dataset of 450 events with a median baseline MTBF of 402.34 h (mean 429.92 h; CV, calculated from the mean and standard deviation, = 65.2%, a level of dispersion that limits the practical use of any single figure for production scheduling). Non-parametric Mann-Whitney testing (required by confirmed non-normality, Anderson-Darling A-squared = 3.81, p < 0.005) identified three factors statistically associated with lower MTBF: incomplete daily operator maintenance (median gap 104.5 h, p < 0.001), preventive maintenance not completed on schedule (gap 99.3 h, p < 0.001), and delayed component replacement (gap 113.4 h, p < 0.001; observed in 66.9% of events). Interventions addressing each factor were piloted on CNC milling machines and extended to four further machine families. Median MTBF rose to 558.70 h (+ 38.9%; Mann-Whitney, p < 0.001), with CV falling from 65.2% to 6.3%. The improvement persisted over a subsequent four-month Control-phase observation ( n = 267): every individual month remained significantly better than baseline ( p < 1.3 × 10⁻⁷ in the weakest month), and the statistically appropriate one-sided regression test found no evidence of regression relative to the immediate post-improvement result ( p = 0.985), with the Control-phase median (564.03 h) in fact marginally exceeding it. This study’s principal contribution is methodological as much as empirical: it demonstrates that MTBF variance reduction, not only the mean or median shift, can be treated as a primary DMAIC outcome, and it reports a transparent left-censoring exclusion rule that other facilities beginning failure-tracking from a fixed start date can apply directly. Findings are presented from a single facility and are not claimed to be representative of a broader setting. As a before–after study with no concurrent control group, the results show an association with the interventions, not proof of causation, and four months of Control-phase data do not establish long-term stability.