REAL TIME PROCESS CONTROL USING NEURAL NETWORKS

THIS PROJECT WILL DEMONSTRATE THE FEASIBILITY OF USING NEURAL NETWORKS TO CONTROL MAGNETIC BEARINGS TO SUPPORT A LEVITATED, FRICTIONLESS ROTATING SHAFT. NEURAL NETWORK CONTROLLERS WILL BE DESIGNED TO PROVIDE THE DYNAMIC CONTROL NECESSARY TO ALLOW A SHAFT SUPPORTED BY A MAGNETIC BEARINGS TO SPIN UP AND SPIN DOWN THROUGH ITS RESONANCE FREQUENCIES. NEURAL NETS WILL PROVIDE DYNAMIC SHAFT DAMPING FOR TRANSIENT RESONANT MOMENTS THAT RENDER CURRENT CONTROL TECHNIQUES IMPRACTICAL. THIS WORK WILL DEMONSTRATE IN A LARGER SENSE THE ABILITY OF NEURAL NETS TO IMPLEMENT OPERATIVE REPLACEMENTS FOR DIGITAL FILTER CONTROLLERS WITHOUT THE REQUIREMENT FOR EXTENDED DEVELOPMENT PROCESSES. THE ALGORITHMS DEVELOPED WILL EMULATE THE PERFORMANCE OF CONVENTIONALLY USED DIGITAL SIGNAL PROCESSING ALGORITHMS, SUCH AS THE INFINITE IMPULSE RESPONSE (IIR) FILTER. STATIC SHAFT LEVITATION AND SIMPLE DYNAMIC CONTROL WILL BE DEMONSTRATED DURING PHASE I TO SHOW FEASIBILITY. IN PHASE II A GENERAL PURPOSE ACTIVE CONTROL SYSTEM WILL BE CONFIGURED AND TESTED, WHICH PROVIDES ADAPTIVE DYNAMIC CONTROL AND EXHIBITS ADAPTIVE LEARNING AND RESPONSE TO LONG TERM CHANGES IN THE STATIC AND DYNAMIC CHARACTERISTICS OF THE SYSTEM. POTENTIAL APPLICATIONS INCLUDE CONTROL OF ANTI-MATTER FOR ROCKET FUEL, CONTROL OF MAGNETS FOR FUSION REACTORS, AND SEPARATION OF IMPURITIES FROM SEA WATER IN DIFFERENTIAL-VORTEX DESALINATION PLANTS. IT WILL ALLOW EXTREMELY PRECISE POSITIONING OF CUTTING TOOLS VIA NEURAL NETWORK CONTROL OF NON-LINEAR PIEZO MICROPOSITIONERS. THIS TECHNOLOGY WILL BOOST RELIABILITY THROUGH REMOVAL OF IMPERFECTIONS WHICH LEAD TO FAILURE AND WILL ELIMINATE EXCESS WEIGHT WHICH IS DESIGNED INTO PRODUCTS TO PROVIDE MARGINS OF SAFETY FOR UNKNOWN PART-OUT-OF-TOLERANCE CONDITIONS.