Innovative models and algorithms for the optimization of layout and control strategy of complex diesel HEVs

This study is focused on the design, optimization and analysis of non plug-in parallel and complex vehicles and on the evaluation of their potential to reduce fuel consumption and NOx emissions, in comparison with a reference vehicle. The simulated vehicles are equipped with compression ignition engines; two different engines were considered for the layout optimization process, and the related data were provided by GMPT-E (General Motors PowerTrain-Europe). A tool has been developed and employed to identify the optimal layout of each vehicle on the basis of the minimization of the overall powertrain costs during the whole vehicle life. These costs include the initial investment due to the production of the components as well as the operating costs related to fuel consumption and to battery depletion. The control strategy has been defined as the algorithm that selects the transmission gear and that manages the power to be provided by the engine and the electric machines of the power-train. In this framework, the transmission gear and the power management are the two control variables. Identification of the optimal control strategy is necessary in order to fully exploit the potential of the hybrid architecture to reduce fuel consumption and pollutant emissions. It is therefore carried out by the so-called optimizer in terms of a specific objective function. This function aims at maximizing the fuel economy with some constraints to the pollutant emissions and to the battery energy and life consumption, according to the application. To this end, two global optimizers, one of a deterministic nature and another of a stochastic type, have been developed, applied and compared. These methods are fundamental for the definition of the vehicle optimal control strategy. They are indeed referred to as benchmark optimizers. A zero-dimensional kinematic model of the vehicle has been developed in the Matlab environment in order to evaluate the evolution of the system variables, as a function of the vehicle velocity and of the control variables. A new mathematical technique has been developed and applied to the vehicle simulation model in order to decrease the computational time of the optimizers. First, the vehicle model equations were written in order to allow a coarse time grid to be used, then, the control variables were discretized, and the values of the system variables were evaluated and stored in a matrix, for all the possible combinations of control variables and for each time node, before the optimization process. However, since the benchmark optimizers are not suitable for on-board applications, one static optimizer and two different rule-based optimizers, which are referred to as real-time optimizers, have been also developed, compared to the benchmark tools and implemented in the vehicle control unit, in order to perform an on-board optimization. Usually this kind of optimization is based on heuristic techniques that may lack of performance in a broad range of applications. In this thesis, machine-learning techniques have been introduced to train the real-time tools. The training procedure that is applied to the rule-based optimizers consists of two parts: the input variable clustering and the rule definition. The vehicle velocity and power, as well as the battery state of charge, have been selected as the input variables. A clustering algorithm has been coded to discretize the input domain of the rule itself into a mesh, i.e., each combination of input variables is associated to an unique cluster. The rule connects every cluster to one of the discrete values of each control variable. Two different approaches have been followed in this study to develop the rule-based optimizers. A clustering algorithm has been developed for the first tool to generate the mesh that is associated to the rule, while genetic algorithms are applied to generate the action to take for each cluster of the mesh itself. In fact, the tool is referred to as Cluster Extracted Rule Optimized (CERO). Genetic algorithms have been instead applied for the second tool to generate the optimal mesh that is associated to the rule, while the most frequent action, in the set of actions that have been suggested by the benchmark optimizer in a cluster of the mesh, is correlated to that cluster. The tool is referred to as Cluster Optimized Rule Extracted (CORE). The vehicle control unit is required to receive the data about the instantaneous vehicle velocity, power and battery state of charge during the trip. These data are processed to identify the cluster they belong to, which is used to index the rule to extract the discrete values of the control variables. The control unit is therefore able to actuate the power-train components to drive the vehicle. The performance of the hybrid vehicles has been evaluated over several driving missions for different oriented optimizations and a detailed energetic analysis has been carried out in order to clearly identify the key operating modes that contribute most to the fuel consumption and NOx mission savings of the different hybrid architectures.