Abstract
Keywords
Exhaust Gas Sensors, Neural Networks, Emission Control, Sensor Array, Exhaust Emission Prediction.
Abstract
The monitoring and control of combustion-related emissions is a top priority in many industries. Availability of reliable sensors capable of detecting combustion gases along with predictive emission modeling tools would provide a better control of combustion leading to reduction of toxic emissions and subsequent energy savings. The automotive industry is an excellent example, where increased use of sensor and measurement technology has led to improvements in engine performance, higher energy efficiency and reduced pollutant emissions. The major commercial methods used to detect combustion gases fall short of practical applications for in-situ measurements, which require stability, sufficiently high operating temperatures, low cost and simplicity of use.
Moreover, there is a trend towards the use of more sophisticated design concepts in the development of new sensor systems, notably the so called smart sensor system, which combines sensing elements and signal processing, conversion and output units. This approach aims at simultaneous or sequential acquisition of more than one type of signal, mainly to resolve the problem of selectivity. This may include the use of more than two electrodes (arrays of electrodes) on a single sensor body, employment of sensor arrays made of many sensors or a multi-layer sensor with a filtering membrane, and applications of artificial intelligence for pattern recognition. In applications, the objective of such an approach may be difficult to achieve, as the sensor array functionality relies not only on the right combination or integration of different sensors but also on the stability of individual sensors. Any change or drift in one of the sensors requires a system-wise re-calibration or re-programming of the pattern recognition part.
Over the last several years, the National Science Foundation (NSF) Center for Industrial Sensors and Measurements (CISM) in collaboration with the Center for Automotive Research (CAR) has been developing several ceramic-based sensors, mostly for the monitoring of combustion gases. These include CO, CO2, O2, NOx and hydrocarbons.
This paper proposes a sensor array to predict the exhaust gas concentrations from a diesel engine. Sensors developed by researchers at CISM namely CO and NOx sensor were used for this application. Experiments were conducted at CAR on a diesel engine where the sensor responses along with operating conditions data were acquired. The acquired data was used to construct a two sensor array for CO and NOx gas concentration prediction. A neural network model was developed with the sensor responses as the input and CO & NOx concentrations as the output. It was demonstrated that the designed neural network when simulated after sufficient training produced reasonable predictions.