Solve the Manufacturing Data Overload Dilemma

Finding process faults in sensor data is critical to improving operational metrics in manufacturing industries. Given the vast quantities of such data and the difficulties of aligning it with key response variables, the endeavor can become difficult, tedious and exhausting. Due to the pace of production schedules, frequent new product commercialization programs, and corporate and regulatory mandates finding the time and energy to maintain an analysis project is daunting and frequently is either not attempted or is soon dropped. That leaves a treasure chest of critical information and insights untapped, and this is a huge missed opportunity. According to Six Sigma philosophy, the inability to control variation in key process variables leads to defects, waste and poor customer satisfaction. Recent efforts to solve this dilemma reported in the literature propose various complex algorithms to sort through vast amounts of data to make such process faults visible. In contrast, this paper will illustrate an industrially proven method to apply basic statistics to efficiently and clearly distinguish normal from abnormal process behavior. Additionally, key statistical machine learning techniques are deployed to enhance process understanding through multiple attractive visualizations. The sample data set is selected from the well documented Tennessee Eastman challenge process, which has been utilized for over two decades in the literature to test both process control as well as fault detection and identification strategies. The enabling software presented is commercially available and provides a compelling opportunity for manufacturers to tap into this data treasure chest to optimize their processes and products by reducing defects, waste and energy usage and thereby improving customer satisfaction and profitability.