A data-driven methodology for monitoring Total Quality Management (TQM) systems in the Industry 4.0 era

Monitoring Total Quality Management (TQM) implementations has long been challenging due to subjectivity, lack of resources, and siloed data. A notable gap remains in the evaluation of the human factor within TQM systems. In the Industry 4.0 era, new technologies have become available to assess the quality of operator-generated data. We present a knowledge discovery in databases (KDD) method for continuous monitoring of TQM systems by evaluating the accuracy of quality control (QC) data collected by line operators. It consists of four steps: calculation of operators’ accuracy using the laboratory measurements as a reference, comparison of operators’ accuracy against the number of complaints, development of Model A to find features causing the poor operators’ accuracy, and implementation of Model B to determine the reliability of operator measurements for which coincident laboratory data is unavailable. We validated the method using data from a TQM-based multinational plant that (over three years), covering from raw material processing to packaging, with anonymized variables. Some QC variables were well established, while others were useless for detecting defects. The Model A identified the operators who may need training and others incapable of distinguishing intermediate QC categories, suggesting a reduction in the number of the categories. The Model B provided more data for the QC managers to evaluate operators’ accuracy and product quality in near-real-time by increasing 50 times the total number of reliable operator measurements. Our proposal represents a step forward in providing better tools for monitoring TQM systems and evaluating human factors in the industry. • A KDD method is validated and successfully implemented in an industrial plant. • Our method showed effectiveness in evaluating operatoŕs accuracy and hidden causes. • We found QC variables unfit for finding defects or incorrectly measured by operators. • A data-driven model identifies required changes in quality control (QC) categories. • Reliable operatorś measurements were increased by 50 times using LASSO regression.