Barzin Mobasher Research Group

Journal 2024

Statistical Process Control of Fiber-Reinforced Concrete Precast Tunnel Segments

C. Pleesudjai, D. Patel, K. A. Williams Gaona, M. Bakhshi, V. Nasri, B. Mobasher

ACI Materials Journal, 121 (2), 127-139

View at publisher All publications

Abstract

Statistical process control (SPC) procedures are proposed to improve the production efficiency of precast concrete tunnel segments. Quality control test results of more than 1000 ASTM C1609 beam specimens were analyzed. These specimens were collected over 18 months from the fiber-reinforced concrete (FRC) used for the production of precast tunnel segments of a major wastewater tunnel project in the Northeast, USA. The Anderson-Darling (AD) test for the overall distribution indicated that the data is best described by a normal distribution. The initial residual strength parameter for the FRC mixture, fD600, is the most representative parameter of the post-crack region. The lower 95% confidence interval (CI) values for 28-day flexural strength parameters of f1, fD600 and fD300 exceeded the design strengths, and hence validate the strength acceptability criteria set at 3.7 MPa (540 psi).

A combination of Run Chart, Exponentially Weighted Moving Average (EWMA), and Cumulative Sum (CUSUM) control charts successfully identified the out-of-control mean values of flexural strengths. These methods identify the periods corresponding to incapable manufacturing processes that should be investigated to move the processes back into control. This approach successfully identified the capable or incapable processes. The study also included Bootstrap Method to analyze standard error in the test data and to determine the degree of reliability associated with different sample sizes when measuring residual strength parameters.

Key findings

  • The residual strength at a deflection of L/600, fD600, is within 1% of the mean flexural strength fDe,150 computed from the total toughness, so it is the most representative residual strength parameter of the post-crack region.
  • The Anderson-Darling test gave P-values of 0.176, 0.223, 0.289 and 0.509 for a normal distribution of f1, fD600, fD300 and fD150. The Weibull distribution gave P-values below 0.010 for the first three and 0.186 for fD150, and the lognormal distribution gave P-values below 0.005 for all four.
  • Between observations 300 and 450 the EWMA charts of fD600, fD300 and fD150 signal an out-of-control process against the lower control limit, and the same range holds the largest number of unqualified results.
  • The CUSUM and EWMA charts can show that the manufacturing process is unstable even when no unqualified sample is observed.
  • From the bootstrap analysis, 25 samples are required for the coefficient of variation to fall to 5%. It falls to 2.5% above 100 samples and stays below 1.5% above 200 samples.