John Stewart, Vice President, MathWorks EMEA
“Technology and innovation,” “environment and legislation,” “economy and competition,” and “society and customers” are four areas of megatrends in the automotive industry. To successfully embrace these trends and meet the challenges facing the industry, automotive OEMs and suppliers worldwide are modifying their development processes. Under the impetus of green initiatives, many projects, such as the electrification of powertrains, are evolving their core processes to adopt Model-Based Design throughout their development life cycle. In this talk, John Stewart discusses how early verification is a key approach within Model-Based Design that enables users and organizations to be well prepared to follow and benefit from industry megatrends.
Jonny Andersson, Scania, Sweden
Scania Driver Support is an on-board support system developed for heavy-duty trucks that went into production in the fall of 2009. It is designed to detect and analyze driving situations while driving. It identifies scenarios in which the driver’s actions are especially important for the driving economy, and it gives advice on how to act just after a situation has been evaluated. Rating the driver’s behavior is also a part of the method to create a positive influence on the driving style. Each situation is awarded up to five stars, and an average score also shows the overall progress.
Four categories define the criteria with which the driver is evaluated:
- Hill driving – Encourages the driver to adapt the speed to the terrain for better fuel economy.
- Brake use – Rewards smooth braking and encourages the use of auxiliary brakes for nonurgent braking.
- Anticipation – Focuses on the interplay between acceleration and braking and encourages a defensive driving style.
- Choice of gears – Gives advice on using the correct gears for fuel economy and performance.
This presentation focuses on the development process for the on-board Scania Driver Support system.
Dr. Martin Schmidt, AVL List GmbH, Austria
In the development process for hybrid powertrains, a lot of time and money can be saved by using simulation models instead of real vehicle components. The same model can be used to test basic functionalities in the lab and the behavior and impact to the unit under test on real components on engine or powertrain test beds.
MATLAB® and Simulink® enable such simulation models to be designed efficiently and support the code generation to a real-time target system.
This session presents engine starter simulation in hybrid powertrains. A starter simulation model has been developed with MATLAB and Simulink to test the start behavior of the hybrid engine.
In the model development process, AVL ARTE.Lab™ is used. Based on MATLAB, Simulink, and Real-Time Workshop®, AVL ARTE.Lab generates the real-time application. This real-time application is executed on the AVL test bed automation and control system. A generic Model Parameter Editor (MPE) allowing the smooth management of the model parameters at run time is provided. This is achieved by modifying MAT-files created in conjunction with the simulation model. In this way, parameter variations of the starter simulation model can be easily applied. A parameter updater has been created with MATLAB Compiler™ for automated parameter migration and preprocessing; updates of the starter simulation model are automatically considered in the parameters.
The results show the realistic start behavior by using the starter simulation model on a hybrid powertrain test bed. By using MATLAB and Simulink, different starter simulation models can be quickly exchanged, which leads to a shorter time to market.
Andrew Preece, Ricardo UK Ltd., United Kingdom
This session presents the Sentience project, which has pursued the application of information about a vehicle’s future route to improve fuel consumption. The project, a collaboration between Ricardo, Orange, TRL, Jaguar Land Rover, Ordnance Survey, and Innovits, has shown significant fuel consumption improvements on a demonstration hybrid vehicle.
During the initial phase, concepts were developed on the desktop using MATLAB and Simulink. Dynamic programming was used in MATLAB to provide a basis for strategies, which were developed in Simulink making use of a detailed vehicle model. Use of Real-Time Workshop coupled with the extensive use of the Embedded MATLAB® subset within the model allowed the rapid development of complex algorithms to the vehicle in the subsequent phase.
Three distinct fuel-saving strategies were developed:
- Enhanced acceleration/deceleration provides an advanced cruise control that matches the vehicle speed to the road ahead, avoiding fuel-wasting harsh braking and acceleration. This strategy is applicable to any vehicle.
- Intelligent air conditioning reduces the need to run the engine wastefully when the vehicle is stationary by anticipating stops and precooling the vehicle. This strategy is most applicable to mild hybrids and vehicles implementing stop/start.
- Optimized engine load manages the split of power between the engine and motors of a hybrid vehicle so that battery charging takes place at the optimum points along the route.
The demonstration vehicle was subjected to an extensive test program: on the road, on a test track, and in a climatic chamber. MATLAB was used to process the test results and derive calibration improvements.
Results showed an estimated 10% CO2 and fuel saving in real-world U.K. driving.
Renault’s Model-Based Design Powertrain Control Development Process: From Initial Requirements to Auto-Coded Blocks
Pedro Moreno Lahore, Renault, France
This session describes Renault’s powertrain software process, which is based on Renault EMS 2010 modular software, rules, and tools that are applied in a MATLAB and Simulink based simulation and software integration platform that helps users throughout development, from the first evaluation of the requirement to the HIL validation of the strategy. This platform helps the different teams with:
- Strategy requirement definition
- Strategy design and validation
- Strategy tuning
- Physical model development for offline strategy performance evaluation
- Software integration
The goal is to share and enrich future software-integrated control strategies, physical models, and clients stimuli throughout the development process in order to increase reliability and quality, thereby decreasing development time.
Developing Vehicle Models with MathWorks Physical Modeling Solutions and dSPACE Automotive Simulation Models
Tino Schulze, dSPACE, Germany
Steve Miller, MathWorks, Germany
When it comes to modeling and simulation, automotive engineers want it all. They want models of the vehicle, the components, and the environment. Engineers designing electric vehicles need to combine electrical, mechanical, and other domains in a single model. They want models that run in desktop simulation and in real time. And, no matter how complete the model provided by the software vendor is, they need the ability to customize it further. To meet all of these requirements, the best option can be the combination of two compatible solutions. The ability to create custom models of vehicles and vehicle components delivered by MathWorks physical modeling solutions combined with Automotive Simulation Models (ASM) from dSPACE enables engineers to create the exact model required for all tasks for Model-Based Design. This session covers the advantages of each environment and how they complement each other.
Lars Kristian Klauske and Christian Dziobek, Daimler AG, Germany
The descriptive quality and readability of models is considered among the major benefits of model-based software development over textual programming. While modeling languages like Simulink provide the necessary visual elements, the quality of a model's graphical representation is still heavily influenced by its layout. Significant effort is therefore spent on the layout's manual improvement: dragging blocks and rerouting lines.
This session describes current challenges in editing Simulink models using exemplary steps from the modeling workflow. It presents an approach supporting modelers by integrating specifically designed layout algorithms into the Simulink graphical development environment. Furthermore, it discusses aspects of user interaction during the layout process and shows that automatic layouting is an enabling technology for the further enhancement of usability and modeling efficiency. The session includes a live demonstration with examples.
Verification by Simulation Within the Model-Based Development Process at Continental Automotive Group’s Engine Systems Business Unit
Dr. Marco Kunze, Continental Automotive GmbH, Germany
The integrated development environment System Design Automation (SDA) at Continental Automotive Group’s Engine Systems business unit supports model-based function development (MBD) in the field of increasingly complex electronic control unit (ECU) structures and supports developers throughout the function development process. Model-based testing and test engineering have taken a larger and larger role, especially in the context of safety-relevant software and ISO 26262.
Based on PC simulation, models can be verified in an early step in addition to pure formal verification and before using further enhanced and hardware-based methods like rapid prototyping. An important point at this time in development is the traceability from requirement to the solution and its test and vice versa. The central topic for function and software engineers who are fulfilling all these tasks is to get guidance through the huge amount of functionalities offered in the commercial standard tools.
The Engine Systems business unit therefore integrated into the SDA environment an MBD test suite, which is based on MATLAB, Simulink, and Stateflow. The central part of the test suite is represented by the simulation manager. This central graphical user interface has been extended in-house and covers the generation of new test cases, the reuse of existing test cases, and test case execution.
This session explains the function of the MBD test suite as an XML-based test plan, specification, and report and shows how the model-based function development is supported by the simulation manager. It presents how the simulation manager makes full use of SDA’s closed workflow from the “golden model” representing the physical algorithm down to the generated code. All testing and validation tasks can be done based on PC simulation or by the means of different rapid prototyping methods and are seamlessly integrated in the development process. In addition, the generation of test vectors out of the golden model leads to test vector objects, which can be efficiently reused by downstream steps of the development process.
Using the simulation manager and the MBD test suite optimizes the efficiency of the daily tasks of the function development process in the Engine Systems business unit at Continental Automotive Group by giving a direct relationship between requirement engineering, test planning and specification, and test implementation, either based on Simulink or based on measurements and test execution and report. This results in a very high quality of modeled functionality. This integration makes it possible to further optimize the execution of the downstream activities in the verification cycle, such as production code generation, integration, and documentation.
Production Code Generation Time Machine: A Guided Journey from Rapid Control Prototyping to E-Mobility
Tom Erkkinen, MathWorks, United States
Code generation experts and novices alike should enjoy this trip back in time, which begins with the roots of Model-Based Design and ends with its latest capabilities. Topics discussed include large-scale modeling, fixed-point design, code optimization, software integration, and code verification. Emerging automotive trends involving certification and standards support are also covered.
Automatic Code Generation of AUTOSAR Software Components for Mass Production Application of Engine Management Systems: Process and Benefits
Franck Narcisse, VALEO Engine and Electrical Systems, France
This session presents the VALEO Engine and Electrical Systems (E.E.S.) development cycle, which is optimized around MATLAB, Simulink, and Real-Time Workshop Embedded Coder.
In the context of Model-Based Design for engine management control laws, VALEO E.E.S. deployed the latest engineering technologies to optimize the design, simulation, validation, and generation of AUTOSAR software components. The close collaboration of the various actors of the development cycle made it possible to bring the requirements for each skill into a consistent and robust workflow of deliverables at each stage of the cycle.
The combination of this global approach for the development cycle and the suppliers’ integration enabled the successful deployment of these processes for a new project with mass production development constraints.
One System, Dozens of Devices and Controllers, Thousands of Parameters — Parallel Computing as an Enabler for System-Level Optimization
Loren Dean, MathWorks, United States
Designing e-mobility systems means multiple devices, batteries, complex machines, multifaceted controllers, and thousands of parameters. And only some combinations will provide the energy efficiency and the dynamic that the vehicles’ customers want.
So engineers need to develop and especially optimize on the system level, meaning hundreds and thousands of simulation and calculation runs. Parallel computing offers a way to realistically unleash this optimization potential while keeping tight schedules. The increased availability of multicore machines and clusters offers the potential to perform analyses on or from the engineer’s computer. The term “stochastic methods” should no longer be an excuse for insufficient coverage of the search space. This session shows how parallel computing can be applied with MATLAB and Simulink to enable engineers to solve large problems more quickly and efficiently using multicore computers, clusters, and beyond.
Dr. Sven Semmelrodt, Continental Automotive GmbH, Germany
Software development for vehicle electronics has evolved from first-generation approaches based on textual specifications and manual coding, through second-generation techniques using model-based specifications, to third-generation methodologies applying automatic code generation. Nevertheless, integration of those model-based components and traditionally programmed parts, such as the basic software, is still done manually, which is an extensive and error-prone process.
Within the product field of instrument clusters and control/network electronics for commercial vehicles (CVs), Continental Automotive provides the Model-Based Development System (MBDS), a closed tool chain for software development, integration, and test that has successfully been applied in several customer projects. In contrast to existing approaches, the MBDS environment provides a fully automated integration and build process, and is hence classified as a fourth-generation integrated model-based development platform.
MBDS is based on MATLAB and Simulink from The MathWorks as Model-Based Design tools and integrates several components for design (MSR blockset, CV blockset, design patterns, guidelines, help), test (model rule check, test coverage analysis, MIL-, SIL-, PIL-, back-to-back, and regression test), documentation (document generator), and integration (project structure, handling, and library concept) as well as interfaces to external tools to ensure a reliable software development process in accordance with the CV product life-cycle definition based on CMMI. A user-friendly graphical user interface summarizes and extends useful features of the MATLAB and Simulink platform.
The session describes the MBDS approach, its advantages, and the resulting business models in partitioning software development between OEM and supplier. Furthermore, the MBDS environment and its main features based on MathWorks products are introduced and the workflow is explained. Finally, the benefits of using MBDS are discussed.
Best Practice for Software Quality — Defining and Measuring Software Quality Objectives for Source Code
Philippe Spozio, Renault SA, France
Thierry Cambois, PSA Peugeot Citroën, France
When verifying and validating code, automotive manufacturers and their suppliers often share the objective of producing safe code with the right timing and cost. However, they have different ways of meeting this objective. The supplier may ensure quality through verification and validation in the development process; the manufacturer may check quality through verification and validation in the final product. Without a common approach on both sides, measuring the real quality of the end code can be a very complicated task.
In order to share their experience in using software verification tools such as Polyspace® code verifiers, automotive manufacturers Renault SA and PSA Peugeot Citroën, automotive suppliers Valeo and Delphi Diesel Systems Power Train, and MathWorks decided in 2007 to create a working group. This working group focused on defining software quality objectives for source code and a common approach to implement and measure them. This session presents their results.
Incremental Quality Objectives have been defined along the software life-cycle process, from the first code version to the ultimate code delivery. Along that path, we describe verification milestones, such as the absence of coding rules violations or the absence of run-time errors, and propose to associate different quality levels with different modules and different deliveries. The quality therefore relies on a modular verification approach based on a typical development process.The final result is a tool-independent document that can be used as a step in a better formalization of relationships between car manufacturers and suppliers with regards to software quality objectives. As a consequence of this work, Renault and PSA integrated the document in their software requirement plan. Additionally, the document proposes a pragmatic way of using verification tools such as Polyspace code verifiers for the purpose of detecting coding rules violations or proving absence of run-time errors.
Dr. André Stork, Fraunhofer Institute, Germany
Digital Mock-Up (DMU) is a widely used technology to virtually investigate geometrical and mechanical product properties. Functional Digital Mock-Up (FDMU) is a combination of traditional DMU with behavioral simulation in mechatronics. Enhanced with functional aspects, this technology enables considerably more insight in product properties to be achieved. To enable FDMU, two main tasks have to be solved: First, established simulation approaches of the areas of mechanics, electronics, and software simulation must interact. This implies solving the task of simulator coupling. Second, the simulation results must be visualized using the geometric models of DMU.
This session presents the development of a prototypic FDMU framework. Given geometric models (e.g., VRML files) and physical models are combined into so-called functional building blocks (FBBs). The FBB interface is prepared to be connected to a framework to build the FDMU simulation model (FSM). During simulation, a simulator coupling algorithm controls the simulation processes of each FBB depending on the FBB’s physical wrappers for multiphysics, electronic, software, multibody, and FEM simulation tools. The visualization enables user interactions, for example, pushing buttons. If it is prepared, parameters can be changed before each simulation run. The session introduces the components of the FDMU framework and illustrates the approach with an application example.
Dr. David Sampson, MathWorks
Steadily rising oil prices as well as ever tougher emissions regulations are strong drivers to reduce the fuel consumption of modern vehicles. However, these customer and legislative requirements, together with the increased complexity of modern engine hardware and control systems, have led to the calibration process becoming significantly more difficult and time consuming. Offline application of Design of Experiments, modeling, and optimization has delivered considerable performance and process efficiency gains. Now, there is interest in whether further improvements are possible through appropriate use of these techniques in the test cell.
This presentation explores the automation of offline analysis using expert scripts, to reduce overall calibration time, and the online estimation of model quality and accuracy, to ensure that the data collected is sufficient but not excessive.
Verification and Validation in a Collaborative, Layered Design Environment for Embedded Software-Intensive Systems
Raymond Tinsel, DAF Trucks, Netherlands
Co Melissant, MonkeyProof Solutions BV, Netherlands
Confidence in correctness and fit-for-purpose of models of both the system-under-design and its environment are key to the success of any model-based approach to systems engineering. With the growing need for complex safety-critical systems in automotive, bidirectional traceability between implementation and tests, and related requirement(s) or change requests, is of growing importance, as is the ability to perform regression testing.
This session demonstrates a lean but scalable, engineer-friendly (single environment, easy navigation) and management-friendly (automated data integrity checking and reporting, role-based permissions) approach to get to a fully verified and validated model-based system design. We take a close look at a pragmatic and manageable, layered, top-down Model-Based Design and engineering process as well as the tool suite and configuration that support and control it.
The process and tool suite are MATLAB-, Simulink-, and code generation–centric and build on top of the available verification and validation capabilities of these tools. Database functionality is used to establish and maintain a central truth and to control permissions as required in a collaborative environment. Version control and audit trails are applied throughout the process. A controlled environment is created without loss of flexibility required in innovation.
The presented combination of process and tool suite has been successfully tailored and implemented in production environments in joint efforts between MonkeyProof Solutions BV and its automotive OEM and tier-1 customers. The level of adoption of the approach is flexible, ranging from creating executable specs for a supplier to in-house production code generation. The discussed approach is usually implemented in stages to enable a smooth transition and to minimize risk.
Richard Backman, AVL Södertälje Powertrain Engineering AB, Sweden
The engine controller has become the most important system that connects the engine with the vehicle and driver. The requirements on an engine controller are among many customer and legal demands; others include:
- Emission legislation
- Fuel efficiency
- Low cost (minimizing the number of sensors and actuators for piece cost and warranty)
To handle these requirements, production engine controllers have become very complex and consist of thousands of parameters and models that have to be optimized. This task normally takes up to 12 months in the test cell to calibrate, not including vehicle calibration in climate and emission tests.
This time-consuming procedure is acceptable for an engine controller that is optimized for mass production, but not for early development and prototyping. The process is so lengthy because of long calibration time, limited I/O, and the inherent complexity, which makes it difficult to adapt the controller to new engine or combustion concepts.
AVL Raptor is a complete control system for rapid prototyping as well as development of production code. Currently it is based on modular and flexible hardware from dSPACE using MATLAB, Simulink, and Real-Time Workshop for the software following the AUTOSAR standard. The model-based software can control any type of engine, including SI, SIDI, CI, and HCCI, with any type of actuator, such as multiple injections, fully flexible valve trains, or complex turbo configurations.
The software uses information from additional sensors, such as cylinder pressure and exhaust temperature, compared with a production system, but can be kept very compact and optimized for fast calibration. The system can normally be calibrated in a fraction of the time it takes to calibrate a production controller.
To further increase the speed of function development, there is a complete model-in-the-loop environment running faster than real time. This environment includes:
- Mean value engine model
- Combustion model for cylinder pressure generation
- Vehicle model
- Driver model
The plant can be used to validate new functionality before running the algorithm on the actual engine, and since the controller used in simulation is identical when running in the test cell or in demonstrator vehicle, most of the bugs in the new algorithm are captured at offline simulation.
The Raptor environment is used for production algorithm development as well as rapid prototyping and advanced algorithm development; the model-in-the-loop environment combined with the complete controller is particularly well-suited as a plant model for developing the algorithm offline and can be validated on an actual engine, using the dSPACE hardware before the production controller is available.
Dr. Angela Bernardini, CITEAN (Centro de Innovación Tecnológica de Automoción de Navarra), Spain
Virtual engineering technology has undergone rapid progress in recent years and has been widely accepted for commercial product development. Product design and manufacturing organizations are moving from the traditional multiple and serial test cycle approach to simulation, which solves problems and validates performances using CAE and CAD tools.
For an efficient process, it is essential that design variants can be done within a short time frame. This generally leads to a challenge when the system under study exhibits nonlinear behavior. This session introduces a new methodology based on neural networks (NNs) and genetic algorithms (GAs), which “put data to work” and provide the best possible solution for a given design based on the available data. The goal of this methodology is to provide designers with a tool that can be used to select the optimum design for a given product. This is possible thanks to the optimization of the NN itself through GA implementation based on the available training data. Genetic algorithms have been used for neural networks in two main ways: to optimize the network architecture and to train the weights of a fixed architecture.
The performance of a NN is critically dependent on, among other variables, the choice of the processing elements (neurons), the architecture, and the learning algorithm. In particular, the connection density (among neurons) determines its ability to store information and learn from it. On one hand, a reduced number of connections may disable the network to approximate the function. On the other hand, dense connections may cause overfitting. NNs are usually seen as a method to implement complex nonlinear functions using simple elementary units connected with adaptive weights. We focus on optimizing the structure of connectivity for these networks using GAs to reduce learning time and avoid CAD/CAE loops. Indeed, this implementation provides neural network topologies that, in general, perform better than random or fully connected topologies when they learn and classify new data.
Genetic operators, such as mutation and cross-over, introduce variety into the initial randomly connected population, modifying the network’s architecture and testing candidate solutions. Once the most effective NN is trained, it is possible to adjust the design parameters, with the same accuracy as FEA or testing data, but sharply reducing the simulation time: The approximate hour and an half needed to analyze critical points by FEA is reduced to few seconds using neural networks. A MATLAB graphical user interface (GUI) works as a quick design guide, where the training data for the NN is obtained from a set of automatically generated FEA analyses. To assess the effectiveness of this methodology, several practical applications are shown. As an example, the optimal preload for bolted joints is returned in a few seconds starting from bolt’s geometry, friction coefficient, and applied torque.