Carlos Guardiola

Identifying static and dynamic prediction models for NOx emissions with evolving fuzzy systems

Abstract: Antipollution legislation in automotive internal combustion engines requires active control and prediction of pollutant formation and emissions. Predictive emission models are of great use in the system calibration phase, and also can be integrated for the engine control and on-board diagnosis tasks. In this paper, fuzzy modelling of the NOx emissions of a diesel engine is investigated, which overcomes some drawbacks of pure engine mapping or analytical physical-oriented models. For building up the fuzzy NOx prediction models, the FLEXFIS approach (short for FLEXible Fuzzy Inference Systems) is applied, which automatically extracts an appropriate number of rules and fuzzy sets by an evolving version of vector quantization (eVQ) and estimates the consequent parameters of Takagi-Sugeno fuzzy systems with the local learning approach in order to optimize the least squares functional. The predictive power of the fuzzy NOx prediction models is compared with that one achieved by physical-oriented models based on high-dimensional engine data recorded during steady-state and dynamic engine states.

An approach to model-based fault detection in industrial measurement systems with application to engine test benches

An approach to fault detection (FD) in industrial measurement systems is proposed in this paper which includes an identification strategy for early detection of the appearance of a fault. This approach is model based, i.e. nominal models are used which represent the fault-free state of the on-line measured process. This approach is also suitable for off-line FD. The framework that combines FD with isolation and correction (FDIC) is outlined in this paper. The proposed approach is characterized by automatic threshold determination, ability to analyse local properties of the models, and aggregation of different fault detection statements. The nominal models are built using data-driven and hybrid approaches, combining first principle models with on-line data-driven techniques. At the same time the models are transparent and interpretable. This novel approach is then verified on a number of real and simulated data sets of car engine test benches (both gasoline—Alfa Romeo JTS, and diesel—Caterpillar). It is demonstrated that the approach can work effectively in real industrial measurement systems with data of large dimensions in both on-line and off-line modes.

ON-LINE FAULT DETECTION WITH DATA-DRIVEN EVOLVING FUZZY MODELS

The main contribution of this paper is a novel fault detection strategy, which is able to cope with changing system states at on-line measurement systems fully automatically. For doing so, an improved fault detection logic is introduced which is based on data-driven evolving fuzzy models. These are sample-wise trained from on-line measurement data from scratch, i.e., the structure and rules of the models evolve over time in order to cope (1) with high-frequented measurement recordings and (2) on-line changing operating conditions. The evolving models represent (changing) dependencies between certain system variables and are used for calculating the deviation between expected model outputs and real-measured values on new incoming data samples (→ residuals). The residuals are compared with confidence regions surrounding the evolving fuzzy models, so-called local error bars and their behaviour is analysed over time by adaptive univariate statistical methods → anomalies in the residual signals indicate faults in the system. Due to local error bars, it is possible to react very flexibly on local regions within the system variables and hence to increase the fault detection performance significantly. Evaluation results based on high-dimensional measurement data from engine test benches are demonstrated at the end of the paper, where the novel fault detection approach is compared against static analytical (fault) models.

Exhaust pressure pulsation observation from turbocharger instantaneous speed measurement

In internal combustion engines, instantaneous exhaust pressure measurements are difficult to perform in a production environment. The high temperature of the exhaust manifold and its pulsating character make its application to exhaust gas recirculation control algorithms impossible. In this paper an alternative method for estimating the exhaust pressure pulsation is presented. A numerical model is built which enables the exhaust pressure pulses to be predicted from instantaneous turbocharger speed measurements. Although the model is data based, a theoretical description of the process is also provided. This combined approach makes it possible to export the model for different engine operating points. Also, compressor contribution in the turbocharger speed pulsation is discussed extensively. The compressor contribution is initially neglected, and effects of this simplified approach are analysed.

A bias correction method for fast fuel-to-air ratio estimation in diesel engines

λ probes in turbocharged diesel engines are usually located downstream of the turbine, exhibiting a good dynamic response but a significant delay because of the exhaust line transport and the hardware itself. With the introduction of after-treatment systems, new sensors that can measure the exhaust concentrations are required for optimal control and diagnosis. Zirconia-based potentiometric sensors permit the measurement of nitrogen oxides and oxygen with the same hardware. However, their dynamic response is slower and more filtered than that of traditional λ probes and, in addition, the sensor location downstream of the after-treatment systems increases this problem. The paper uses a Kalman filter for online dynamic estimation of the relative fuel-to-air ratio λ−1 in a turbocharged diesel engine. The combination of a fast drifted fuel-to-air ratio model with a slow but accurate zirconia sensor permits the model bias to be corrected. This bias is modelled with a look-up table depending on the engine operating point and is integrated online on the basis of the Kalman filter output. The calculation burden is alleviated by using the converged gain of the steady-state Kalman filter, precalculated offline. Finally, robustness conditions for stopping the bias updating are included in order to account for the sensor and model uncertainties. The proposed algorithm and sensor layout are successfully proved in a turbocharged diesel engine. Experimental and simulation results are included to support validation of the algorithm.

Assessment of a sequentially turbocharged diesel engine on real-life driving cycles

The article presents the development of the control manager of a parallel sequential turbocharger system. This control manager must decide the transition between the different operation modes of the boosting system. The control manager must protect the system from surge and over-speed risks.The methodology followed in the development process was based on the use of concurrence survey data from a similar engine, the simulation with a 1D code and finally the assessment on engine test bench. Real-life driving cycles were used during the development process, which were acquired in three different driving scenarios (city, road and mountain road).

A learning algorithm concept for updating look-up tables for automotive applications

Abstract Look-up tables are commonly used in the automotive field for handling operating point variations. However, constant maps cannot cope with systems variations and ageing. Methods, such as Kalman filter or Extended Kalman filter for non-linear cases, can be used for table adaptation providing an optimal solution to the problem. But these methods are computationally intensive, making difficult to implement them on commercial engine control units. The current paper proposes a learning method for online updating of look-up tables or maps. This algorithm uses precalculated membership functions based on a standard Kalman filter observer for weighting the adaptation. The main contribution of the method is the derivation of a steady-state Kalman filter observer that lowers the calculation burden and simplifies the implementation against the standard Kalman filter implementation that requires higher computational cost. As far as table is updated online while engine runs, this allows correcting drift errors and the unit-to-unit dispersion. The method is illustrated for mapping engine variables such as λ − 1 and N O x in a Diesel engine by using an adaptive look-up table, and its characteristics make it suitable for implementing in commercial engine electronic control units for online purposes.

Applying evolving fuzzy models with adaptive local error bars to on-line fault detection

The main contribution of this paper is a novel fault detection strategy, which is able to cope with changing system states at on-line measurement systems fully automatically. For doing so, an improved fault detection logic is introduced which is based on data-driven evolving fuzzy models. These are sample-wise trained from online measurement data, i.e. the structure and rules of the models evolve over time in order to cope 1.) with high-frequented measurement recordings and 2.) online changing operating conditions. The evolving fuzzy models represent (changing) non-linear dependencies between certain system variables and are used for calculating the deviation between expected model outputs and real measured values on new incoming data samples (rarr residuals). The residuals are compared with local confidence regions surrounding the evolving fuzzy models, so-called local error bars, incrementally calculated synchronously to the models. The behavior of the residuals is analyzed over time by an adaptive univariate statistical approach. Evaluation results based on high-dimensional measurement data from engine test benches are demonstrated at the end of the paper, where the novel fault detection approach is compared against static analytical (fault) models.

DFT-based controller for fuel injection unevenness correction in turbocharged diesel engines

Although combustion failure diagnosis techniques have been widely developed over the last few years, real-time correction of fuel injection failures, such as drift, are still deficient. In this paper, a controller for the correction of fuel injection failures is presented; the aim of the algorithm is to ensure that the same quantity of fuel is injected in each one of the cylinders. The controller is based on a linear model that relates the low-frequency region of the Fourier transform of a dynamic engine signal and fuel injection unevenness. Model inversion is used as an injection failure observer, and discrepancies in the injected fuel mass in each cylinder can be estimated, even in the case of multiple and simultaneous failures. This observer is used for closing the loop and performing the control action via an integral controller. Theoretical bases are given for the controller, and the stability and settling error are related to the error of the linear engine model assumed. This technique can be used for different engine signals, like crankshaft speed, exhaust manifold pulsation, and turbocharger instantaneous speed. Experimental results obtained on a diesel turbocharged engine, where the turbocharger instantaneous speed was used as input information of the controller, are presented proving the performance of the fuel quantity control algorithm

On Board NOx Prediction in Diesel Engines: A Physical Approach

For pollutant emissions predictive physical modeling in diesel engines, three key points have to be taken into account: