A robust time delay auto-regressive exogenous fuzzy inference system for real-time estimation of catalyst temperature over engines coldstart operation: a multiobjective implementation scenario

Abstract

In the current investigation, the authors propose an intelligent real-time system to dynamically simulate the catalyst temperature in automotive engines over the coldstart period. In general, the behavior of an engine during the first few minutes of its operation, namely the coldstart period, is highly transient and nonlinear. However, it is important to the engineers of automotive industry to develop advanced simulation techniques capable of capturing the engine system's behavior during coldstart periods. The main motivation behind such a fact emanates in the pursuit of making successful strides towards decreasing the amount of unburned hydrocarbon emissions (HCs) of the engine. Catalytic convertors play a pivotal role in reducing the amount of this emitted pollutant. Therefore, here, the authors develop a knowledge based intelligent identification system, based on the integration of a low-level fuzzy identification machine (FIM) identifier and a hyper-level multiobjective optimizer, to monitor the catalyst temperature over the coldstart period for a given automotive engine system. In the current investigation, the authors propose the use of synchronous self-learning Pareto strategy for the multi-objective optimization of the architecture of FIM. To suit the intelligent scheme for online applications, the authors capture the required database using the concept of nonlinear auto-regressive exogenous systems. The developed intelligent approach can be used as a real-time soft sensor to observe the state of the catalytic convertor and send the required data to the coldstart controller. To ensure the efficacy of the resulting time delay FIM framework, it is used for capturing the catalyst temperature in three different experimental conditions. Through a comparative numerical study, it is demonstrated that the proposed soft sensor can be efficiently used for real-time estimations of the catalyst temperature. Besides, the results let us make sure that intelligent approaches have an aptitude to be effectively used as real-time observers instead of expensive commercial sensors. Generally, the results of the current investigation substantiates the authenticity of advanced computational approaches for coping with the considered identification problem, to help resolving one of the challenging issues in the field of automotive engineering, that is, the coldstart HC emission reduction.

Auto-regressive multiple-valued logic neurons with sequential Chua’s oscillator back-propagation learning for online prediction and synchronization of chaotic trajectories

“…In the rest of this section, the authors will explain how the concept of ARX system implementation can be used to prepare MVLNNs for real-time applications. ARX enables us to reform the basic structure of MVLNNs to a time-delay-based network which can be used for studying the underlying dynamics of nonlinear systems (Mozaffari and Azad, 2014b). In this context, we need to add a finite number of delays (lags) to the measured input/output signals.…”

“…where ε is the white noise signal measured at time t. Based on a theoretical analysis, it has been demonstrated that white noises can be modeled using a proper Gaussian function (Mozaffari and Azad, 2014b). Therefore, if experiments reveal that the realtime white noise error is significant, we can fit a Gaussian function to capture the error and further improve the performance of AMVLNN.…”

“…In addition to the aforementioned theoretical contribution, the authors intend to expand the realm of applications of MVLNNs, by applying it for online nonlinear system identification and real-time control tasks. To do so, the auto-regressive exogenous (ARX) system implementation, which is in-line with state-space representations required for control systems design, is considered for revising the structure of MVLNNs (Mozaffari and Azad, 2014b). Through an exhaustive numerical study, the authors demonstrate the high potentials of ARX-based MVLNN (AMVLNN) for online system identification tasks.…”