Using big data in automotive engineering? ETAS GmbH Borsigstraße 14 70469 Stuttgart, Germany Phone +49 711 3423-2240 Commentary by Friedhelm Pickhard, Chairman of the ETAS Board of Management, translated from Elektronik automotive 1/2016 Press and Public Relations: Anja Krahl anja.krahl@etas.com www.etas.com Big data has long since taken hold in the automotive industry, serving as the basis for new business models and value chains. But what are the benefits of big data for automotive engineering? Virtualization has dramatically changed new vehicle development in recent years. Simulations on computers can increasingly be used in place of test benches and on-road testing. Tasks that previously had to be completed in sequence can now be performed in parallel. Cost and time expenditures, in addition to failure rates, are on the decline. The next logical step towards achieving increased efficiency in development processes while managing system complexity is big data. The task at hand is to use measurement data information more effectively now than in the past.
Big data is the hot topic of the moment in the automotive industry. The number of conferences and publications on the subject has increased dramatically as of late. Connected vehicles, the increasing availability of diagnostic, sensor, and traffic data, and digitalized planning, production, and sales processes all give rise to enormous amounts of data. The analysis of this data promises novel possibilities for process optimization, in addition to incredibly accurate insights into the behavior of vehicles and drivers, which can be very beneficial in areas such as the development of new vehicle models and variants. Big data refers to a statistical approach that seeks to obtain new information from huge amounts of data. The term is also used as a synonym for the efficient processing, management, and evaluation of complex data that exists in large quantities in a variety of forms and formats. Just as they yield benefits in marketing, product planning, production and logistics, quality assurance, sales, fleet management, and Car-2-X services, big data methods can also be a useful tool in automotive engineering. To measure or to simulate? Intense global competition and stringent requirements for automotive safety, emissions, and fuel consumption all necessitate the use of advanced electronic systems. These systems must be developed, tested, and adjusted within specified time periods and on limited budgets while taking into account hundreds of possible driving and operating scenarios. Managing the complexity and elevated requirements for efficiency and quality in automotive engineering means that many tasks can only be effectively carried out through extensive use of computer simulations and virtual models. Simulations are replacing testbench and road-test trials long before physical prototypes are available. These simulations also allow the investigation of extreme conditions that cannot be replicated using real-world testing, or can be replicated only at great expense. However, it is not possible to completely eliminate test-bench, on-road, and hardware-in-the-loop testing. One reason is because the results from tests performed on real-world objects remain the gold standard during the development process, in addition to being an essential input for approving releases. Another is that measurement data from such testing is very beneficial in the construction of simulations and models, as these then typically depict and predict system behavior far more accurately than physics-based calculations.
Aside from this, the only way to find and analyze those errors in the behavior of the system during validation that cannot be traced back to deviations from specifications during implementation is by employing statistical methods using measurement data. Measure everything In the past, isolated measurements were often carried out using spot checks in order to answer a single question. However, this approach falls short when it comes to predicting and managing the behavior of complex systems with the proper degree of certainty, such as modern high-tech IC engines, hybrid powertrains, and advanced braking and driver assistance systems. Here it is a big advantage to amalgamate all measurement tasks as far as possible; in other words, to measure everything at once in a set of tests carried out on the test bench or in the vehicle. This approach enables users to systematically reuse data, easily correlate different measurement variables, optimize dependent functions, and understand complex error patterns. At the same time, it facilitates the complete documentation of measurement results, such as with OBD inspections. There is also a compelling economic argument in favor of getting the maximum possible use out of test benches and test vehicles by integrating as many measurement tasks as possible into one set of tests. The challenge this poses for in-vehicle measurement is how to ensure complete and time-synchronized recording of measurement signals from different sources. In the case of advanced driver assistance systems, for example, these sources include sensors such as stereo cameras and radar units, which supply a large amount of raw data in an uninterrupted stream during the journey data that has to be processed by the ECUs in real time. Increasing capabilities for data collection and analysis In its role as a vendor of both hardware and software for measurement, calibration, and diagnostics, ETAS has played a major role in the steady improvement of data collection efficiency in testing over the last several years. Tools such as the FETK ECU interface or the ECU and bus interface modules of the new ES800 hardware can work in concert with powerful processors, data buses, and transfer protocols to record ECU data at rates of several gigabits per second while time stamping this data with exacting precision. This is the
principal reason why testing engineers can increase the scope of measurement tasks to go beyond the satisfaction of their own inquiries. To save time and expense, the goal must be to record all signals from the vehicle electronics on as few testing days as possible without interruption. In this way, data volumes measuring in the hundreds of terabytes can be accumulated in short amounts of time. The basis for this big data approach lies in the collection and processing of very large volumes of data. These measurement data contain several tens of thousands of signals which are recorded in several thousand time grids. In addition to the data collection instruments, solutions for analyzing these measurement data are also on the horizon. The MDA 8 measurement data analysis tool developed by ETAS enables development, testing, and calibration engineers to handle measurements involving extremely large data volumes. MDA 8 offers high processing speeds, in addition to new user interface concepts such as the new oscilloscope scroll and zoom functions, which were developed together with users within a scrum framework. These allow very short time segments to be intuitively extracted from very long measurement sequences. Picture: A test vehicle is equipped with the ETAS measurement, calibration, and prototyping system of the future. The ES800 system communicates with the ECUs via FETK or serial ECU interfaces, and collects measurement data from the vehicle buses, as well as from other digital or analog signal sources in the vehicle.
The challenge of data management In principle, existing data arising from simulations and testing during the course of development can be used again and again for a wide range of purposes. But for other engineers, teams, departments, or companies to effectively utilize this existing data, it must first be efficiently managed and made accessible using powerful databases, search algorithms, and visualizations, in addition to navigation and selection mechanisms. Of course, one vendor cannot cover every single facet of the big data chain, with solutions for the generation, transfer, management, analysis, and visualization of data. Open, standardized approaches are needed so that customers can utilize specific tools from various vendors. One thing that is clear is the need for new, scalable solutions that provide more flexible and faster access to the large pool of stored data. With its scalable Enterprise Automotive Data Management (EADM) solution, ETAS is pursuing a scalable approach that is open and ASAM-ODS compatible from the start. Outlook: efficiency counts The goal of reducing testing days and measurement campaigns through the use of big data methods is ambitious, but realistic. The work of calibration engineers will increasingly transition towards virtual environments, and to the desktop. On-road testing will only be used to certify validations that were carried out using simulations. And even this need not be done during testing conducted by one s own team; rather it can be based on measurement data collected by others. Efficient, scalable tools are essential in order to access this data. Hardware, too, plays a crucial role when it comes to processing large volumes of data. A popular rule of thumb in this regard is that twice the number of processor cores is necessary in order to process twice the volume of data at the same speed. In the past, databases could only offer this type of scalability within certain limits, but now with big data technologies these limitations no longer exist. Despite the growing complexity of powertrain and assistance systems, and despite an equally growing number of sensors and ECUs, a big data approach will make it possible to further shorten the testing phase while once again
significantly reducing the number of prototypes and test vehicles used. The key lies in the collection of a wider scope of measurement data, as well as the systematic, intelligent management of these data in powerful databases. Not only will engineers selectively search for and quickly find data in these databases, but they will also be able to supplement and refine the data during the course of the project. And last but not least, they will be able to utilize this treasure trove of data far more effectively during subsequent projects than in the past. The use of big data methods is the next logical step towards increasing the efficiency of the development process while managing the complexity of automotive electronic systems.