A vision for MPC performance maintenance

摘要: Model predictive control (MPC) is an advanced that has found widespread use in industries, particularly process industries like oil refining and petrochemicals. Although the basic premise behind MPC easy to comprehend, its inner workings are still generally viewed or regarded as too for actual plant operator understanding. This lack of understanding exposed when performance deteriorates sometime after commissioning, often case some commercially operated plants. Currently operators might have difficulty over reasoning about degradation formulating steps investigate cause. A tool described aims make more transparent operators. Commonly reported causes discussed ways which can recognise them they occur outlined. Issues addressed include: making set controlled variables be used a given manipulated simpler clearer; controller performing poorly identify source deterioration. An aim determine under what instances return previous levels request specialist support simply switch off. A goal avoid kind situation where gets worried deteriorating ends up taking knee jerk actions cause further problems MPC. At top maintenance hierarchy trends comparison group, reference graphical compared with counterpart. If any abnormality observed these trends, encouraged choose option from list preliminary diagnostic questions contained group below best describes abnormality. Each associated suspected causes. When chosen list, led symptoms investigation window, contains scripts detailing systematic examination each symptom, view either rejecting confirming suspicion. Assisted four background information windows: virtual without transfer function matrix window steady state gain, relative gain array (RGA) weight (RWA) window. The windows contain guidance refer time during symptom investigation. Development at design stage. key components been research implementing on three nonlinear models, continuous stirred tank reactor (CSTR), evaporator fluid catalytic cracking unit (FCCU). Case studies representing different scenarios simulated, followed by procedure diagnosing, isolating recovering such degradation, based assumed operator’s perspective expert’s technical reasoning. knowledge gained develop outline vision data-driven model help sensible judgements form direction diagnosis fault isolation, possibly, recovery procedure.