The hybridization of powertrains is an important step toward efficient and clean mobility. In particular, the possibility of shifting the operation of the combustion engine to ranges with a higher efficiency level and representing purely electric driving modes is one of the main advantages of hybrid drives. This shifting of the load point can be further optimized on the basis of route data that includes the expected vehicle speed as well as the road gradient, and is considered to be the state of the art with regard to modern hybrid drives.
Combined with the development of predictive and automated driving functions, further potentials can be tapped. The key factor for an actual reduction of the energy requirement under real driving conditions is a precise forecast of the future development of a traffic situation. This forecast can be based on a multitude of potential sources, such as sensor data, high-resolution maps, and vehicle communication, whereby all the data is fused into a comprehensive environmental model.
Based on the information from this model, the longitudinal guidance of the vehicle and the powertrain control can be optimized. In cooperation with the Institute for Combustion Engines of RWTH University Aachen, Germany, FEV has developed a function structure that is capable of using a multitude of potential data sources. This creates a solution space for predictive speed profile optimization. This speed profile can then be used in order to optimize the operation of torque distribution between the hybrid components.
The function structure was integrated in a hybrid prototype vehicle constructed jointly with DENSO. A robust, real-time model predictive control algorithm is used in order to optimize the longitudinal guidance of the vehicle.
The HYBex3 concept vehicle
The HYBex3 (”HYBrid power exchange 3 modes“) vehicle was developed in order to determine the impact of a cost-effective DHT transmission concept on the driveability of the vehicle and test it under real conditions. It was developed jointly with DENSO AUTOMOTIVE Germany. The base vehicle is a MINI Cooper with a turbocharged 100 kW three-cylinder combustion engine. The serial transmission was replaced with the hybrid transmission to be examined, which was specially developed for the application case. The powertrain topology is equivalent to a mixed hybrid equipped with two electric engines (EE) in a P2/P3 layout. The P2 machine is located between the electrohydraulically powered clutch and the two-stage spur gear component. The synchronization elements are also actuated electrohydraulically. The P3 machine is positioned at the transmission output and therefore has a fixed transmission ratio to the wheel.
Various operating modes can be represented with this DHT transmission. For purely electric driving, the combustion engine is stopped and the clutch is opened. Electric engine P2 can therefore be operated in both transmission stages. In addition to a high starting torque in the first gear, this enables a maximum vehicle speed of 140 km/h in the second gear.
In hybrid operation, serial or parallel driving is possible. In parallel operation, one of the two gear sets is engaged. In serial operation mode, the transmission is shifted to neutral. The combustion engine is then exclusively connected to electric engine P2 while electric engine P3 operates the wheels. All gear changes are synchronized entirely electrically, so that the friction clutch can remain closed even in hybrid operation. The serial operation in the low speed range and the parallel operation at higher speeds enable a significant increase of the system efficiency level.
The operating strategy provides for the combustion engine being operated at a very low dynamic and the implementation of fast load changes by the electric path. The transmission ratios enable a significant reduction of the rotational speed of the combustion engine, without compromising the overall dynamic of the powertrain. The operating strategy was optimized with a Design of Experiments. For this purpose, the parameters of the stop-start strategy of the combustion engine were optimized simultaneously with the parameters of the battery charging strategy. For the final parameterization, a compromise between the layouts for different driving cycles was selected.
The distribution of the torques of the two electric engines, both in parallel operation and in fully electric driving, is determined by an online optimization patented by FEV. The search algorithm varies the torque distribution until the energetically optimal case is found. In doing so, both the battery limits and the power limits of the electric engines are taken into account for the current situation.
The function structure developed for predictive longitudinal dynamic control is designed in such a way that a multitude of data sources, optimization routines, and powertrain structures can be represented in said function structure.
The first step is an aggregation and fusion of the available data into an environmental model, followed by a prediction of the traffic situation. This enables an optimization of the speed profile. On the basis of that, an acceleration control of the vehicle is carried out. The planned speed profile can also be used in order to adjust the charging status strategy. If the desired charging performance is determined, the torque distribution between the powertrain components is carried out on the basis of said performance and the wheel torque requirement.
The precise forecast of the current traffic situation requires the aggregation of all available data. This includes, for instance, RADAR sensors, LIDAR sensors, or optical cameras that traffic participants can identify with the help of image recognition techniques. Usually, these sensors indicate the type (passenger car, truck, pedestrian, etc.), the relative positions and, potentially, the relative speed of the detected objects.Further information can be obtained from the on-board navigation systems, which indicate speed limits, road gradients and curvatures as well as, potentially, intersection data for the most probable path of the vehicle via an “electronic horizon”. If the navigation system is connected to the internet, data on average speeds along the planned route and traffic jams can be provided.
Additional data can be obtained through the future connection of vehicles using 5G or ETSI ITS G5. This Vehicle-to-everything (V2X) communication should include, among other things, the positions, direction, and speeds of other vehicles, as well as the layout of intersections and the status of traffic light systems. The vehicle communication can therefore provide data that goes beyond the horizon detectable via on-board sensors.
Since the same object can therefore be detected multiple times by various data sources, the data aggregation must also include a functionality for data fusion. This is especially advantageous for hardware setups with different types of sensors, e.g. a RADAR sensor and camera sensor. The RADAR sensor can precisely define the distance to and the relative position of a vehicle driving ahead, but cannot determine the lateral position of the vehicle in relation to the road markings. In contrast, the camera sensor can only provide estimates regarding the relative speed and the distance, but can precisely determine whether the detected object is in the same lane as the vehicle under consideration. After the fusion of several data sources, an aggregated object list is created, which only contains valid and relevant data for all detected objects, and generates a corresponding environmental model.
Before an optimization of the vehicle trajectory can be carried out, there must be a forecast of the development of the current situation. This forecast is based on the relevant objects that the environmental model provides. The first step is the determination of the speed limit along the prediction horizon. Based on that and the current condition of detected vehicles driving ahead, the speed and position trajectory of these vehicles is forecast.
On the basis of this, a solution space is spread out in which the downstream optimization algorithm can operate. The function structure developed by FEV and the Institute for Combustion Engines enables the implementation of different algorithms to this end. Depending on the requirement, simple, rule-based approaches, as well as model predictive control or discrete dynamic programming methods can be represented.
Application in the vehicle
To test the function structure, a real time-compatible model predictive control (MPC) was implemented in the rapid prototyping control unit of the HYBex3 concept vehicle and various test scenarios were carried out. In a first demonstration, the functionality and real-time compatibility of these scenarios for a predictive adjustment of the HYBex3 concept vehicle was proven. With an efficient implementation of the MPC using the qpOASES tool, an optimization of the speed curve for a horizon of 10 s can be carried out within less than 100 µs.
In the future, the modular design of the function structure can be used to expand the forecast horizon of the vehicle – for instance, with traffic lights ahead – or to represent predictive, automated driving functions such as Predictive Cruise Control (PCC).
Legislative requirements for emissions and fuel consumption reduction are driving OEMs to develop innovative powertrain and vehicle technologies. In addition to continued development of new technologies with conventional internal combustion engines (ICE), there is an increasing trend toward electrification. These trends make it essential to develop relevant means of assessing the NVH performance of electric drive tunits (EDUs). These components do not generate the amplitudes of noise and vibration observed from internal combustion engines (ICE). As such, the methods used for NVH assessment and target development of IC engines are not sufficient for electric machines: While the objectives of ICE-based NVH development are reduction and refinement of source excitations, EDU-based NVH development focuses on the elimination of potential objectionable noise behavior in the context of ever-changing or missing masking noise content. For example, there is a reduced background noise for masking tonal noise in the absence of a running internal combustion engine.
The expectation for interior noise content from ICE-based vehicles (i.e., “powertrain presence”) depends highly on the vehicle class and target demographic; while luxury cars target low interior noise content, performance vehicles demand some level of powertrain noise feedback (with an emphasis on development of the desired “brand character”). Conversely, the tonal noise typically associated with EDUs is universally considered annoying; hence the goal is to minimize perception of this content in the vehicle. This becomes challenging, given that the reduced overall noise content available to hide (mask) this tonal noise content is lower on electric vehicles than ICE-based vehicles. Figure 1 below illustrates the difference in typical noise levels observed in ICE-powered vs. electric vehicles (EV) in the form of FEV scatterbands. Clearly, significant reduction in overall noise levels on EV are evident, especially at low-to-mid vehicle speeds.
To predict the perceptibility of tonal noise content in-vehicle, masking band analysis can be used. As shown in the figure below, the order content can be compared to surrounding 3rd octave levels to determine how much noise is available in adjacent frequencies to mask the tonal noise. If the order level (of whine noise) is high relative to the corresponding 3rd octave band noise levels, this is an indication that there is insufficient background noise to mask the order, resulting in a perceptible, and hence, objectionable whine noise. Also shown below is a masking surface which illustrates the masking content for various orders over the operating range of an example vehicle. At higher vehicle speeds, wind noise is more prominent; this results in more masking content and an associated reduction in perception of whine noise.
NVH issue root-cause analysis & mitigation
Increased trends in electrification and associated technologies have posed new challenges in NVH development. In addition to minimizing tonal noise content in the vehicle’s interior, there are multiple potential NVH issues related to transient instabilities (e.g., gear rattle or other driveline issues). FEV utilizes a structured approach, with extensive experience in 8D analysis and Design-of-Experiments (DoE) to address such problems. As part of this root-cause analysis, FEV utilizes a combination of industry-standard methods (e.g., Ishikawa diagrams), as well as FEV developed tools and processes. FEV’s Vehicle Interior Noise Simulation (VINS) is an example of a unique methodology that can be effectively utilized in the support of root cause analysis with complex noise issues. The VINS process is a unique time-domain transfer path analysis which provides insights into noise sources and transfer paths which contribute to sound quality issues under steady-state or transient conditions. Any noise issues identified at the vehicle’s interior can be broken down to identify contributions of various structureborne and airborne noise paths. The critical noise paths can be further decomposed to identify any potential opportunities for improvement (mount isolation, attachment point stiffness, vibroacoustic sensitivity, acoustic attenuation, etc.). Because the results generated are in the time-domain, advanced analysis methods or subjective evaluations (listening studies) can be used for assessment of the overall simulated noise or individual path contributions. Figure 2 schematically shows the integration of the VINS methodology in a structured 8D root-cause analysis process.
Component-level EDU NVH assessments
FEV has established standard testing procedures for quantifying radiated noise, sound power, and vibration at the component-level to facilitate assessment of source-level inputs to support electric vehicle NVH development. Analogous to ICE-based powertrain NVH testing, overall EDU radiated noise levels are typically assessed based on average radiated noise, measured at a distance of 1 m from the EDU (e.g., using SAE J074 standard). Additionally, it is common practice for electric machines and EDUs to augment these assessments with measurement of sound power, utilizing a hemispherical or parallelepiped array (e.g., IS0 3744 or 3745 standards). Structureborne excitations can be assessed by measurement of vibration at the EDU mounting locations (i.e., interface points between the EDU and vehicle).
Comparison of average overall sound pressure levels between ICE-based powertrains and EDUs in the figure below illustrates that noise levels radiated from EDUs are significantly lower than those observed from ICE powertrains. As such, assessment of individual orders excited by the electric machines and/or gear meshing frequencies is more relevant (than overall noise levels) for quantification of EDU NVH performance. An example of order content relative to overall radiated noise levels is illustrated below. This comparison provides information regarding the contribution of orders to overall noise levels. Additional investigation of the frequency content of the component noise levels can provide insights into perceptibility of this noise in a test cell environment. However, component level data analysis alone does little to predict the perceptibility of these orders by the customer in-vehicle. For this, a vehicle-centric data analysis approach is required, as described below.
Vehicle-centric EDU NVH target development
Derived from the VINS methodology, FEV has developed an additional process for interior noise prediction, called dBVINS. Unlike VINS (which utilizes vehicle-specific noise transfer functions), the dBVINS process predicts interior noise based on a combination of source data (noise and vibration, as measured in the test cell) and standardized vehicle noise transfer functions. These “standardized” noise transfer functions are based on median vehicle noise sensitivity performance, derived from the extensive database of vehicles assessed by FEV. By standardizing the transfer functions, the interior noise relevant NVH performance of a given component (e.g., EDU) can be judged based on component-level tests from a NVH test bench. This allows for direct comparison of the expected interior noise performance of different EDUs or design variants of a development EDU. Specific to EDU development, this process allows for prediction of relevant order content at the vehicle interior. Comparison of this order content to the masking noise levels discussed above provides insights into the potential perceptibility of tonal noise issues by the customer. Appropriate design changes using a combination of CAE (e.g., MBS/FEA) and test-based approaches (e.g., calibration changes, NVH countermeasure development) can be employed to improve the component-level NVH performance of the EDU, utilizing such a vehicle-centric approach.