Online Monitoring of Exhaust Gas Emissions of a Boiler with Diesel/Biodiesel Fuel Blends

An online, low-cost, neural network based soft sensor was developed and implemented, with the aim of being compatible with the present industrial automation and control systems. The virtual sensor was designed to predict exhaust gas emission levels of an originally built diesel boiler with diesel/biodiesel mixture blends at different air ratios, using temperature, flow rates and pressure as inputs variables to the neural network model. With that purpose, experimental data were obtained for different diesel and biodiesel mixtures of fuel in the boiler, in order to validate the model. The online experimental data consisted of process variables, obtained using a SCADA system connected to fieldbus communication protocol instrumentation, and exhaust pollutant gas levels, obtained using a conventional gas analyzer. A detailed methodology was established for each phase of the study, including data collection and treatment, topology and neural training comparative studies, implementation of the neural network algorithm proposed in the SCADA system and the application of an online validation and maintenance procedure. Experimental online tests confirmed the compatibility between the exhaust gas emission levels inferred by the online soft sensor and those obtained with the analyzer. Scan acquisition intervals were six times smaller and maintenance proceedings were optimized, without demanding a large time interval. Thus, the automation solution can be used to provide pollutant monitoring, helping to achieve a more consistent operation, regarding production and environmental profits.