Abstract In response to the need for dynamic adaptability in adhesive manufacturing, this paper introduces a machine learning model and a new error-based concept drift detection mechanism based on Page-Hinkley Test (PHT) which is a statistical method used for change point detection in time-series data for enhancing accuracy and robustness in adhesive manufacturing process. Upon its implementation, the model monitors the prediction error in the manufacturing process such that it dynamically adjusts its responses in the presence of significant deviation. The integrated machine learning and PHT-based approach successfully detected multiple drift points, starting at t = 1,007 and going up to t = 1,342, which covers both abrupt and gradual shifts. Contrasting with the model, the study was able to work with only one pre-defined drift point, such as t = 1,000, after each detection of drift, the cumulative error resets, allowing the model to relearn from the present data distribution. This repeated recalibration yielded a final root mean square error (RMSE) of 2.0550, showing a great error reduction, thereby improving the product quality. This increased flexibility is very useful for adhesives production within the oil and gas industry, where variation in material quality, environmental conditions, or equipment performance might otherwise affect product quality. Compared with the traditional static model, the integrated machine learning and PHT-based system greatly improved the accuracy and response time of predictions, demonstrating its value in real-time quality control and defect rate minimization for manufacturing.
Nwachukwu et al. (Thu,) studied this question.