Supported by: VEC
In conventional automotive vehicle applications, only the measurement of low frequency temperature variations is usually required, and standard robust sensors such as thermocouples, resistance temperature detectors (RTD) and thermistors suffice. However, recent advances in engine design have resulted in the need for robust temperature sensors that have fast response characteristics. An important example where such sensors are now required for control and diagnostics would include on-board diagnosis (OBD) of catalyst malfunction.
In many sensors, the smaller the sensing elements, the faster will be the response but at the expense of durability and ease of manufacture. Therefore, most sensors involve a compromise between performance and the conflicting requirements for ruggedness and low cost. Experimental work on exhaust systems at QUB showed that during transient operation, conventional thermocouple sensors gave errors of up to 200°C. A reduction in wire diameter dramatically improved the accuracy, but in the harsh environment of an exhaust system, a lower limit to the diameter is quickly reached, below which sensor failure occurs.
This new multidisciplinary project within the Virtual Engineering Centre (VEC) uses dual sensors with different characteristics, and then mathematically reconstructs the actual gas temperature from the two outputs. The approach originates from work into combusting turbulent flows conducted at the NASA-Lewis Research Centre and at the Combustion Laboratory of Nagoya Institute of Technology (NIT), based on frequency domain and time domain methods respectively. While effective for some scenarios, both suffer from shortcomings such as sensitivity to noise in the measurements and in the case of the NASA method, the sensor characteristics must be known a priori.
In earlier QUB dual sensor research, a more robust time domain method and one based on an extended Kalman filter were developed and successfully used to reconstruct temperatures in the exhaust of an engine during a transient. However, both methods require that the sensor characteristics are known a priori which limits their applicability. Recently, researchers at Oxford University have adapted the QUB Kalman filter approach to allow in-situ estimation of the sensor characteristics. Their results have highlighted the errors due to offsets in the measurements and rapid variations in exhaust gas velocity; the Kalman filter requires a finite time to adapt to the system dynamics and consequently there is uncertainty regarding the estimates for the sensor parameters in highly pulsating flow.
We are researching a completely novel discrete-time linear identification framework, which allows in-situ evaluation of the sensor characteristics. This eliminates the major shortcoming of earlier QUB and other approaches which require that the dual sensor characteristics are known a priori. Extensive simulation studies have shown that the new methods reduce the sensitivity to noise on the inputs.
Much of our recent theoretical work is concerned with the evaluation of alternative identification schemes. Regular Least Squares (LS) has proved unsatisfactory as it produces biased parameter estimates while more powerful techniques such as Generalised Total Least Squares, which accommodate coloured input and output noises, have been shown to provide bias-free estimates.
Ongoing research involves three main areas of work:
Simulation: The work on sensor characterisation and temperature reconstruction was first performed in simulation, allowing both the bandwidth of the temperature signal and statistical nature of the noises to be selected in a controlled manner.
Air Flow Rig: The next stage was to move to dual sensor measurements from a laboratory test rig. Initially the velocity of the air flow was held constant and the air temperature was varied in a sinusoidally manner.
In-cycle Gas Temperature Measurements: The aim here was to use the low bandwidth sensors to measure the temperature signal at the exhaust port of a real engine over one engine cycle. Initial work has shown that due to the rapid variations in exhaust gas velocity, including flow reversals where the time constants for the thermocouples become very large, all existing methods for sensor characterisation fail; the sensor parameters are dependent on gas velocity, and may have a higher bandwidth than that of the gas temperature.
Current theoretical work is concentrating on adaptive, sliding window and blind equalisation solutions to some of these fundamental issues, in conjunction with a substantial experimental effort.