HIDDEN MARKOV MODELING FOR INTELLIGENT CONTROL OF INDUSTRIAL PROCESSES

THE PRINCIPAL OBJECTIVE OF PHASE I IS DEVELOPMENT OF METHODS FOR OBTAINING PROCESS MODELS FROM IN-SITU AND POSTMORTEM MEASUREMENTS THAT CAN BE USED FOR EITHER QUALITY CONTROL OR FEEDBACK CONTROL. A NOVEL FEATURE OF THIS APPROACH IS TO IDENTIFY THE PROCESS BY USING THE HIDDEN MARKOV MODEL (HMM), WHICH HAS BEEN USED EXTENSIVELY IN SPEECH RECOGNITION. HMMS PROVIDE A NATURAL MEANS TO USE DATA FOR IDENTIFYING A COMPLEX, UNCERTAIN, NONLINEAR SYSTEM SUCH AS A PROCESSING PLANT. IN PHASE I, THE HMM METHODOLOGY IS BEING EXTENDED FOR CONTROL SYSTEM DESIGN, TRAINED WITH SIMULATED AND ACTUAL DATA, AND TESTED. A SECOND RELATED OBJECTIVE, MORE SUITED TO PHASE II, IS DEVELOPMENT OF COMPUTER-AIDED CONTROL ENGINEERING (CACE) DESIGN TOOLS THAT APPEAL TO THE NEEDS OF PROCESS CONTROL ENGINEERS. DESPITE MANY THEORETICAL AND COMPUTATIONAL ADVANCES IN THE CONTROL SCIENCES, THERE REMAINS A SUBSTANTIAL GAP IN ITS APPLICATION BY ENGINEERS IN THE PROCESS INDUSTRIES, WHERE SUCH A DESIGN APPROACH COULD SIGNIFICANTLY ENHANCE PRODUCTIVITY. WITH THE APPROPRIATE CACE TOOLS, THESE ENGINEERS CAN PERFORM SOPHISTICATED DESIGNS WITHOUT DEMANDING SUBSTANTIAL INCREASES IN SKILL LEVEL. SUCH TOOLS ALSO EXTEND THE LIFE CYCLE OF CONTROL DESIGNS, DECREASING THE NEED FOR MAINTENANCE OF CONTROL LOOPS. THE OVERALL BENEFIT TO THE PROCESS INDUSTRIES IS IMPROVED PERFORMANCE FROM AUTOMATION PROJECTS, WHICH ENABLES MORE PROFITABLE, CLEANER, AND SAFER OPERATION OF PROCESS PLANTS.