This article introduces opportunities that are seen at the intersection of electrification, connectivity, and smart controls in the automobile industry. Computational Intelligence provides the vehicle the ability to reason, adapt, and learn based on historical usage data, the present operating conditions, and the predicted future states. Modern automobiles continue to grow in complexity and sophistication. Electrified powertrains now provide vastly improved fuel efficiency by utilizing high-voltage systems to overcome some of the shortcomings of traditional combustion engines. Smart controls have enabled a wealth of new vehicle features ranging from automatic climate control to vehicle dynamic control. Vehicle connectivity, having already empowered the driver through infotainment and telematics, now promises new computing resources and information that can be leveraged directly for improved vehicle performance. At the intersection of these three vehicle mega trends lies a field that is rich for development. In the future, drivers will benefit in everything from enhanced drivability to more durable vehicles.
The C-MAX Hybrid headlines Ford’s transformed lineup, one-third of which will feature a model with 40 mpg or more in 2012.
The automobile of the future is electric. It is connected. It is smart. Clearly, we see that the automotive industry is in the midst of a major migration toward electrified vehicle technology. The number of electrified vehicle nameplates offered in the U.S. market has grown from two in 2001 to more than 30 today. The reason for this explosive growth is the substantial benefit that electrification offers in increasing overall vehicle efficiency and thereby greatly improving fuel economy. At the same time, we as a society are moving toward a world where we are always “connected.” This extends into the automobile where technologies such as Cellular, Bluetooth, WiFi, Vehicle-to-Vehicle and Vehicle-to-Infrastructure (V2V/V2I), and Power Line Communication (PLC) are already being integrated into new vehicles. This vehicle connectivity presents many new opportunities ranging from using off-board memory and computation capability to accessing broad ranging and real-time online information sources. In order to build these automobiles of the future, we need an intelligent control system that optimizes the vehicle’s operation for each customer on every trip over the entire life of the vehicle. Computational Intelligence provides the vehicle the ability to reason, adapt, and learn based on historical usage data, the present operating conditions, and the predicted future states. This article will introduce some of the opportunities that we see at this intersection of electrification, connectivity and smart controls and give some examples of the new functionality that may be coming to your vehicle soon.
Automobiles have Grown in Complexity and Sophistication since Their Invention at the end of the 19th Century
In recent decades, this trend has accelerated as advanced electronics and embedded controls have become ubiquitous in modern vehicles. These technologies facilitate rapid introduction of new features that improve everything from safety and efficiency to comfort and convenience. Indeed, the vast majority of new features added to cars are now software enabled.
Today’s automotive consumers continue to grow more interested in vehicle efficiency. Rising fuel prices and concerns about global warming are driving customers to seek more fuel-efficient vehicles in their purchase decisions. Automakers have responded by introducing technologies designed to improve efficiency. Vehicle electrification has been one of the most effective technologies, providing fuel economy improvements of up to 50%. As automakers find ways to reduce the cost of electrified components and systems, the technology will continue to propagate throughout their product portfolios.
Vehicle connectivity is another rapidly emerging trend in the industry. Today, vehicles already connect to the outside world through embedded cellular systems in the vehicle (e.g. GM’s OnStar system) or through “brought-in” systems using a driver’s phone connected to the vehicle through Bluetooth technology (e.g. Ford’s Sync system). Until now, this connectivity has primarily served to empower the driver by bringing the “always connected” mindset into the car. Future vehicles will also leverage connectivity to empower the vehicle to achieve improved vehicle system performance.
Clearly tomorrow’s vehicles will be smart, connected, and electrified.
Electrified powertrains are now a familiar technology in the automotive industry. The first production hybrid vehicle of the modern vehicle era was launched in the United States over ten years ago and most automakers now have at least one hybrid vehicle in their portfolio. Moreover, electrification offerings have grown to include plug-in hybrids and battery-only electric vehicles as well. Each of these configurations offers customers unique benefits in energy efficiency but they also come with unique challenges to overcome to reach their full potential. To understand these challenges, it is useful to consider each individually.
Hybrid Electric Vehicles
Hybrid electric vehicles (HEVs) combine a conventional powertrain with electric motor(s), power electronics, and a high- voltage battery for the purpose of increasing overall powertrain efficiency. Figure 1 shows a powersplit HEV configuration that includes two electric motors coupled to the engine through a planetary gearset and to the wheels through the driveline. The motors are also connected electrically (not shown) to a high- voltage battery. This is only one of many types of HEV configurations.
The efficiency improvements of HEVs are achieved through engine downsizing, engine start/stop, regenerative braking, and the ability to dynamically trade off fuel usage with electrical energy usage in the vehicle using controls. The last item, known as energy management optimization (Sciarretta & Guzzella, 2007), is typically done by trying to minimize the vehicle’s fuel usage over a drive cycle while maintaining the battery’s state of charge.
Although there are many types of HEV configurations, each of them has at least 2 degrees of freedom for selecting the powertrain operating points, u(t), that can be used to achieve this optimization. To date it has not been possible to fully achieve this optimization in a production vehicle, however, because it relies on a priori knowledge of the drive cycle, which typically hasn’t been available.
Battery Electric Vehicles
Battery electric vehicles (BEVs) are the simplest form of electrified vehicles from a complexity perspective. In BEVs, the powertrain and fuel tank of a conventional vehicle are replaced by an electric motor and a high-voltage battery (see Figure 2). The powertrain torque control in a BEV is relatively straight forward. Since these vehicles have only one power source, it is simply a matter of converting the driver’s demand into a motor torque request under the constraints of the driveline components. Instead, the challenge for BEVs is to deliver the maximum possible driving range. Because these vehicles only have the energy of the onboard battery available for propulsion, their driving range is typically limited to 100 miles or less before the battery needs to be recharged from the electrical grid. Cost, weight, and packaging considerations prevent the use of larger batteries. Moreover, the time required to charge the battery typically ranges from three to eight hours depending on the power capabilities of the vehicle and the charging system. The combination of these factors has motivated automakers to look for ways to squeeze out every additional mile of range while also minimizing “range anxiety” of drivers.
Automakers have limited opportunity to use controls to give BEV drivers more range and less anxiety. Unlike HEVs, BEVs have no available internal degrees of freedom to use to improve efficiency through powertrain control. Instead, the only opportunities lie with being able to influence how drivers operate their vehicles, for instance by choosing energy efficient routes or by reducing driving aggressiveness. In order to reduce driver anxiety, automakers need to provide the best possible estimate of the “distance until empty (DTE)” for the vehicle, which in turn requires a prediction of the expected vehicle energy use along the planned route. Realization of all of these functions requires learned or forecasted knowledge about the driver and the vehicle’s operating environment.
Plug-in HEVs (PHEVs) provide the benefits of both HEVs and BEVs at the cost of a larger battery than would typically be in an HEV. Because PHEVs, like HEVs, can be run on fuel only without depleting the battery, they offer driving ranges similar to HEVs and can be easily refueled at a gas station. In addition, because of the larger battery, like a BEV, they also offer some electric-only driving range.
PHEVs have one additional degree of freedom beyond an HEV. Because customers expect to charge their PHEVs from the grid after each trip, the vehicle needs to operate in a way that will fully deplete the battery during the trip. If the trip is longer than the electric-only driving range of the PHEV, selection of the depletion profile of the battery charge provides another opportunity for optimizing energy usage. Like in an HEV, optimizing this energy usage depends on a priori knowledge of the drive cycle.
Athough some predictive controls can be performed based on locally collected data in the car, a connected vehicle will have several advantages in terms of access to data and computational resources. There are many different ways to “connect” a vehicle, using different infrastructures, technologies and transmission mediums. In this paper we will focus on the common mobile Internet connectivity, typically the same sort as is used for handheld smartphones and laptops.
Mobile Internet is a collection of techniques used to connect mobile consumer devices such as smartphones to the Internet. It uses existing cellphone networks to send and receive IP data packets effectively making the device a portion of the Internet. However, as it uses wireless transmission to the cellphone infrastructure, the quality of the connection varies depending on distance to cellphone towers and the surrounding environment. Some of the key characteristics in mobile Internet are:
Speed or throughput is the measure of how much data can be transmitted at once and is the most common attribute. The maximum theoretical throughput varies depending on technology with typical rates of approximately 30-100 kbps for 2nd generation mobile systems (GSM/GPRS, CDMA2000/IX), 0.4-7 mbps for 3rd generation systems (UMTS, EV-DO or “3G”) and 100-300 mbps for 4th generation systems (LTE, WiMAX, or “4G”). The maximum theoretical throughput is seldom achieved as this assumes a fixed location in close proximity to the tower, and the speeds available for a moving car are therefore much lower. Figure 3 compares the speed of various mobile technologies.
Latency is the time it takes for a single data package to reach the destination. This varies considerably in mobile networks with averages ranging from 10ms up to several hundred milliseconds (and sometimes seconds) under bad conditions.
Packet Loss is measured in percentage and indicates how many packets never reach their destination. In good coverage areas, this should be minimal, but as a car moves in and out of coverage, several percent of loss during a trip is possible.
Another important property of mobile internet is that, just as with home networks, the mobile device is seldom given a public internet address. This does not affect messages and requests originating from the car, but it prevents cloud-based servers from directly contacting the vehicle without the vehicle inviting the connection first.
By allowing communication between the car and the “cloud” several advantages can be realized:
Outsourcing of computation. We can perform computations or run algorithms that would require much more powerful computers (faster or more memory) than could be installed in a car.
Access to persistent personal storage. Data for predictive algorithms could be based on a driver’s usage of more than one car, letting the predictive algorithm work in an optimal way even for new or temporary vehicles (e.g. rental cars).
Real-time and dynamic data access. An optimization algorithm could make use of factors not available in the vehicle—the driver and/or car might behave differently based on weather, traffic or personal schedule which can be available from the cloud. Cloud connectivity can also give access to a large set of static or dynamic data sources useful for the control algorithms, such as high-resolution maps, or topography information.
Crowd-sourcing. By allowing access to data collected from all drivers, personalized algorithms that are depending on historic driver data could still be made to produce beneficial results even when there isn’t enough personal data available, either because the situation is new or the driver has not been at a specific destination before.
As mentioned in the section above, the unique characteristics of mobile communication introduce a unique set of challenges. First, since coverage is never guaranteed, a vehicle can never rely wholly on connectivity and there will always need to be a default functionality implemented onboard. Secondly, even if connectivity exists, it might be too slow to be useful. Unfortunately this is not always possible to measure and the vehicle must always be prepared to switch over to alternative local functionality.
Smart Control System Design
People ultimately use their vehicle to get from one location to another. On any given trip there are three key ways that the overall driving experience can be affected. The first is the choice of route, e.g. shortest distance, shortest time, greenest. The second is the choice of vehicle operation. In this, the driver can choose to drive aggressively, smoothly, or anywhere in between. The third is the choice of propulsion system operation. The vehicle control systems choice of the battery power, engine torque, and other operational parameters has a significant impact on trip efficiency. The first two are choices that can be influenced by the control system, but ultimately the driver makes the decisions. The third is determined by the vehicle control system.
Connectivity offers unprecedented levels of information that can be used by the vehicle control system to help the driver get from point A to point B in the way that the driver is looking for. Smart control system design can help drivers pick an optimal route, coach drivers in how to best operate their vehicle, and automatically pick the optimal operating point of the powertrain to maximize the system efficiency.
The level of attribute performance (e.g., fuel consumption, emissions, drivability) on a trip from point A to point B is related to the amount of information richness. As shown in Figure 4, as the control system learns more about the upcoming situation, the opportunity becomes bigger for attribute improvement through smart control system design. At one end of the spectrum, drivers can provide information about their operational preferences. At the other end of the spectrum, the control system has perfect preview information. Across this spectrum of information richness, there is an increasing opportunity for control system optimization and improved vehicle attributes.
Driver Initiated Control System Configuration
Drivers can inform their vehicle about their driving preferences and the control system can adapt its logic to increase the vehicle performance. With a limited amount of energy stored in the high voltage battery and long recharge times, driving a BEV comes with additional concerns about range anxiety. Drivers can inform their vehicle about their driving habits by preconfiguring the start times for their next trip, as well as their climate control preferences. The vehicle then automatically charges the battery during the most cost effective time of day or night, while ensuring that the charging event completes by the start of the next trip. Additionally, the cabin temperature and battery temperature can be preconditioned while the vehicle is plugged into the grid. In this way, the cabin and battery temperatures are controlled to the desired setting in advance of driving using grid energy rather than the precious energy stored in the battery. This leaves more energy available for propulsion, thereby extending the driving range and reducing range anxiety. Figure 5 shows an example of a BEV owner configuring their charging and temperature preferences.
There are many pieces of information available to a vehicle that can be used to construct preview information. Using a planned (or predicted) route, map data, and real time traffic information, a smart control system can construct a prediction about the upcoming situation that the vehicle will experience. Electrified vehicles are particularly well suited to take advantage of this preview information due to their onboard energy storage.
For example, in a PHEV, a key goal of the control system is to utilize all of the available energy in the battery prior to the next charging event. It is well understood that depleting the battery evenly across the entire trip is more efficient than depleting the battery at the maximum rate in the beginning of the trip (Tulpule, Marano, & Rizzoni, 2009). In this example, a single piece of preview information (driving distance until the next charging event) can be used to configure the control system and deliver improved fuel economy. Predicting this information can be done based on past driving history (Naghshtabrizi, Kristinsson, Yu, & McGee, 2011) or through the navigation system.
Moving beyond only driving distance information, the control system can use preview information about the upcoming road segments, e.g. city or highway, to employ pattern-based state-of-charge (SOC) pre-planning algorithms and further improve fuel economy. Pattern construction, classification, and real time distance based SOC control are all key technologies required to leverage this information. The distance-only information has been shown to improve fuel economy of a PHEV by more than 6%, while including the pattern based preplanning technology improves fuel economy by nearly 8% (Yu, Kuang, & McGee, 2011). Figure 6 compares the fuel consumption for the Base SOC control, the Linear Discharge control, and the Pattern Based SOC control for one particular drive cycle. The second part of Figure 6 shows the accumulated fuel consumption across the driving distance. For this particular example, the default strategy initially uses less fuel due to the higher battery usage. However by the end of the drive cycle the default strategy consumes more fuel.
The Quest for a Fully Optimized Control System
As more and more information is accessed, a clearer picture of the upcoming driving profile can be constructed. Depending on the time (or distance) horizon, different information sources can be fused into a coherent prediction. For example, if we know nothing about the trip, the control system can guess based on past usage. For instance, if it is 7 a.m. on a Tuesday morning and the car is started in the owner’s garage, then it is highly likely, based on driving history data, that the person is driving to work. Once the system knows the destination, either from the navigation system or through a prediction, the system can determine the route. With a route, the system can consult various geographical data sources to determine the elevation, speed limit, road curvature, and various other road attributes for each segment of the trip. Given the map data, the system can overlay traffic signal timing and real time—and predicted-traffic data and obtain a likely traveling speed and elevation profile for that particular driver. Alternatively, clustering algorithms can be used to summarize the operational profiles into a distinctive set of patterns (Yu, Tseng, & McGee, 2012).
In this example, we can see that increasing information richness leads to a more accurate prediction across a longer horizon. At the extreme, if a control system has perfect preview information, dynamic programming (DP) based solutions can be used to determine provably optimal control laws. While DP is a useful tool for benchmarking the best possible solution, it is not practical because there are no perfect predictions. However, control system researchers are constantly finding new ways to make use of every new piece of information.
In Figure 7 we can see that the preview horizon extends from a few meters to an entire lifetime of driving. Across this spectrum, different information sources are used to inform the vehicle control system and different approaches are used to improve electrified vehicles. An exemplary idea is given for each category to illustrate potential applications.
Modern automobiles continue to grow in complexity and sophistication. Electrified powertrains now provide vastly improved fuel efficiency by utilizing high voltage systems to overcome some of the shortcomings of traditional combustion engines. Smart controls have enabled a wealth of new vehicle features ranging from automatic climate control to vehicle dynamic control. Vehicle connectivity, having already empowered the driver through infotainment and telematics, now promises new computing resources and information that can be leveraged directly for improved vehicle performance. At the intersection of these three vehicle mega trends lies a field that is rich for development. In the future, drivers will benefit in everything from enhanced drivability to more durable vehicles.