Gain vehicle insights with virtual sensors
A short history of how our startup was born
COMPREDICT is a deep-tech startup based in Darmstadt/Germany. We provide the automotive and mobility industries with software algorithms and analytics that enable large-scale exploitation of in-vehicle data, using machine learning (ML) and data-driven approaches.
The two co-founders, Dr.-Ing. Rafael Fietzek and Dr.-Ing. Stéphane Foulard, founded COMPREDICT in 2016 after their PhD studies at TU Darmstadt. During their research work, they recognized the high optimization potential that still remains unused in vehicle engineering. Insights from the use of in-vehicle sensor data allow cost savings and reduction of CO2 emissions for manufacturers, fleet operators and consumers: through precisely right-sized components, longer durability, predictive maintenance and optimized fleet management until resale.
Current use of sensors data in automotive industry
One has to distinguish between image-based sensors for advanced driver assistance systems (ADAS) like for autonomous cars, and time-based sensors (time series like wheel speed sensor or GPS tracking). Examples for ADAS functions are ACC (Adaptive Cruise Control) and AEB (Automatic Emergency Braking). In the area of heavy commercial vehicles like long haul trucks, platooning consists of linking of two or more trucks in a convoy, using connectivity technology and such automated driving support functions. Because of the market demand and market growth in AD/ADAS, the research in this field is acting as a precursor regarding sensor fusion.
Which sensor is used to detect other vehicles, pedestrians, etc.?
The adoption of advanced driver assistance systems and autonomous vehicles make extensive use of imaging data. The type of sensors that they use are typically cameras in the visible light range, Lidar technology and radars alongside with communication to the traffic infrastructure (V2X) and to other vehicles (V2V). All major players in the automotive industry work on sensor fusion in order to generate meaningful information out of the imaging data. One well-known challenge consists of identifying objects on the road in the backlight at sunset. Cameras struggle in this kind of situation, where a Lidar can be expected to be more reliable. Sensor fusion consists of taking the best out of the information from each sensor like cameras, Lidars or radars, depending on the situation and environmental factors.
Sensor fusion of time based signals
Other areas in automotive R&D already use simulation, front-loading and digital twins for product development. Sensor fusion of time based signals is less commonly used. The percentage of connected vehicles rolling off production is ever-increasing, and so is the volume of data they produce and feed back to the car manufacturers.
A recent article, Forbes cited a prediction that “[…] the market for data from cars in motion will become five times bigger than the market for cars themselves.” (Peter Cohan, Forbes.com, August 18, 2021). Whatever the exact figures will be, industry experts agree on market growth: the market size for in-car data will become a multiple of the market size for cars.
However, when currently talking to automotive industry players, it seems that the adoption of methodologies to make use of vehicle data is mostly still at an early stage. In many cases, large amounts of data are generated, but then hardly used. There is a demand for clear, powerful, and privacy-compliant tools that can be used across departments and easily rolled out. This is where we step in. At COMPREDICT, we develop virtual sensors that enable such an evaluation of vehicle data at large scale, as they generate unique vehicle insights from existing sensor data.
COMPREDICT’s areas of activity
Our first area of activity right since 2016 was automotive R&D. Stéphane Foulard has written his PhD thesis on how to use in-vehicle data for improved vehicle reliability and durability calculations. Today, vehicle components are often over-designed, in order to be “on the safe side” so that they do not break over a whole vehicle lifetime, even for the most demanding 1% of the drivers. But this does not make sense for the 99% of other drivers: over-design increases weight, cost, fuel consumption and CO2 emissions. Our target is to contribute to CO2 reduction with precise determination of loads and usage profiles from prototype vehicles or series cars, so that manufacturers can reduce component weight, cost and also save development time.
The second area of activity which we further developed this year is condition monitoring of series vehicles. We either cooperate with companies providing telematics solutions to fleet managers and consumers, or with fleet operators directly, or with OEMs. Our virtual sensors enhance existing fleet management solutions with precise data for predictive maintenance and fleet usage optimization. We observe a clear market demand for tools to reduce maintenance cost and to increase fleet efficiency.
We are also dealing with insurance companies regarding driving profiles’ analysis and the estimation of the remaining value of vehicles.
Recent R&D activities on virtual sensors
We are working on new products and solution to support the industry shift to hybrid electric vehicles (HEV/PHEV) and battery electric vehicles (BEV). Electrified cars behave differently and are driven differently from conventional vehicles.
For instance, we are now working on a project with an European manufacturer, where we calculate the impact of a new electrified powertrain on their durability requirements. We even deployed our virtual sensors to a fleet of prototype vehicles in order to gather reliability data over large distance in short time, with software-based virtual sensors.
Since 2021, we are involved in a project called De4LoRa (Double E-drive for Long Range), where a consortium led by Vitesco Technologies and TU Darmstadt is developing an innovative electrified powertrain with a specific architecture to reduce the cost of electrified vehicles for consumers while still providing sufficient range autonomy.
We also participate at the self-driving fleet project named Campus Free City which was officially launched in November 2021. The project is conducted by a consortium of eight industry and reasearch partners and funded by the German Federal Ministry of Transport and Digital Infrastructure (BMVI). It aims to study a fleet of self-driving robot vehicles that are connected to the Internet in a real environment. COMPREDICT delivers AI-based algorithms to predict and avoid vehicle component failures.
See also our press/media section.
Portfolio of virtual sensors for motor vehicles
In the market segment of telematics solutions and fleet management, we offer various virtual sensors for vehicle mass, battery condition monitoring, brake pad wear, brake anomaly and tire tread depth. Integration and interfacing into existing fleet management tool takes place through flexible REST API and SDK. Our virtual sensors can provide a clear competitive advantage to fleets through improved fleet efficiency, true predictive maintenance and reduced operating costs.
In the market segment of OEMs and vehicle development, we provide specific virtual sensors that measure component fatigue and durability, wheel forces and torques, steering forces or high voltage battery failures. Our virtual sensors help to reduce R&D times and costs, and to increase product quality. They allow scaling of very expensive test campaigns to larger number of prototype vehicles and longer mileage, without any risk of damaging sensitive laboratory equipment.
Read more about our use cases of virtual sensors.
How does a virtual vehicle sensor work?
Virtual sensors are software-based. They make use of various existing sensors in a vehicle, like for example a temperature sensor or a speed sensor. Through sensor fusion, virtual sensors compute a target signal or target information that would be otherwise too costly or too complicated to measure. Read more about the functioning of our CAN-bus based virtual sensors.
What’s unique about COMPREDICT’s virtual car sensor technology?
What distinguishes us from other providers of AI-based algorithms and analytics are our deep vehicle knowledge and our patented algorithms. We provide, for example, a unique virtual sensor vehicle mass. It is capable to monitor the loaded mass throughout a working day with a precision of +/-5%, and it works off the shelf with built-in automatic algorithm training for each car.
The virtual sensor for vehicle mass is of major interest for precise range estimation of battery electric vehicles. Efficient last mile logistics have a strong impact on costs and customer satisfaction of the overall supply chain. When you consider that the driving range of an electric delivery vehicle depends heavily on the current mass, which changes all over the day, the value of our innovation for efficient charging management is obvious.
COMPREDICT’S outlook to 2022
Data science is a rather young discipline in automotive industry. Currently, data volumes generated by connected vehicles increase. The industry still lacks adoption of safe and convenient tools for large-scale exploitation of these in-car data. At COMPREDICT, we offer a mature solution to a market that is now becoming aware of the value in their data – a domain where considerable market growth is expected in the upcoming years. Manufacturers and fleet managers estimate that big data analytics is the next big thing to drive innovation.
We are right now scaling up and working on large vehicle fleets with automakers and telematics companies. As well, we are running our third financing round (Series A) to support our growth. In 2022, we plan to showcase our solution at major industry events and to expand abroad (e.g in North America or South Korea). Contact us for more information.
Further reading: What is the significance of automotive sensors and car driving sensors?
Sensors in general are physical hardware devices that make use of physical or chemical effects in order to measure a physical quantity like temperature, pressure, weight, distance, force, torque, speed and many others. In a vehicle, sensors are used for instance to display information like current speed, gear, engine oil temperature, outside temperature etc. to the driver. Sensor data are also used by the various sub-systems in a car: the ECU (engine control unit) or MCU (electric motor control unit), BMS (battery management system).
In some cases, such sensors are too complex to implement and operate, or too costly in larger series. Many vehicles already have tire pressure sensors on board, but measuring the tire tread depth is still a manual operation. Our virtual sensors provide a convenient, software-based solution for fully automatic monitoring of tire wear, brake pad wear and battery condition.
+49 (0)6151 3844614
64283 Darmstadt, Germany