On data-driven profile monitoring using memory-type control charts: a sequential change-detection framework for chemical engineering processes

Abstract

Reliable interpretation of sequential measurements requires methods that can distinguish meaningful changes in process relationships from routine variability. This requirement is particularly important for heat exchangers, where progressive fouling may alter the overall heat-transfer coefficient, its response to operating conditions, and the associated measurement variability. This study develops a data-driven profile-monitoring framework based on a double cumulative sum (DCUSUM) procedure for detecting changes in linear process profiles. The proposed DCUSUM ₃ chart simultaneously monitors the profile intercept, slope, and error variance, which represent the response level, sensitivity to the explanatory variable, and residual variability, respectively. Coding the explanatory variable enables independent estimation and monitoring of these characteristics using routinely collected measurements without requiring a detailed physical process model. Performance is evaluated through Monte Carlo simulations under zero-state and steady-state conditions using average, dispersion, median, percentile-based run-length measures, and conditional expected delay. The results confirm that the chart maintains the prescribed in-control performance and that its detection delay and run-length variability decrease systematically as individual or joint shifts increase. Comparisons with Shewhart, EWMA, CUSUM, and DEWMA profile charts demonstrate competitive detection across a broad range of changes, with particularly strong steady-state performance. Application to controlled shell-and-tube heat-exchanger measurements illustrate earlier identification of fouling-related profile changes, thereby supporting reliable condition assessment and timely maintenance decisions. An interactive Shiny application supports chart calibration, performance evaluation, and practical implementation.