Andreas Abel, ITI
Designing a holistic validation strategy for the embedded systems in an armored multipurpose vehicle (AMPV).
Designing a series of tests using real-time testing tools built with NI VeriStand software and TraceTronic ECU-TEST automation software to create a hardware-in-the-loop (HIL) test bench to validate embedded systems more quickly and completely.
Andreas Abel - ITI
René Müller - TraceTronic
ITI is one of the leading software and engineering companies for system simulation. SimulationX is a standard tool that evaluates the interaction of all components in technical systems with full support for the Modelica language. ITI works with subsidiaries, distributors, and partners worldwide and is also a National Instruments Alliance Partner .
TraceTronic provides innovative solutions, services, and software products designed to develop and validate complex embedded systems. The company’s services range from developing and testing software functions for electronic control units (ECUs) to developing HIL systems from conception to realization.
To equip defense units as well as police and security forces with new levels of mobile, modular, and protective technologies for their current operations, Krauss-Maffei Wegmann (KMW) and a number of other companies took on the challenge of developing a new generation of AMPVs that have high mobility and provide maximum protection at the same time. They also created a self-supporting safety cell made from armored steel and a composite armor that set new benchmarks for these vehicles. The vehicles exceed the current protection standards and achieve significant weight optimizations. Simple vehicle handling and optimized human-machine interfaces (HMIs) inside the vehicle further contribute to the high protection level because the driver and crew can focus on mission-related tasks. The simpler it is to drive the AMPV, the safer it is for vehicle occupants and the equipment. In cooperation with experienced software and hardware manufacturers, we designed a holistic validation strategy for the embedded systems in the vehicle.
The project started with implementing the HIL test bench. First, we analyzed the customer requirements and the electronic control units (ECUs). The resulting analysis formed the foundation for the technical concept and the test bench specification. Market research on existing HIL simulators quickly revealed that there is not a standard solution that meets the specific project requirements concerning flexibility, degree of integration, and price, so we developed a custom system based on both off-the-shelf and specialized components.
We selected NI VeriStand as the real-time platform. This NI solution is based on industry standard hardware, which helps us implement a high-performance system at a very reasonable cost. Also, we can scale the system’s computational power with growing testing requirements in a flexible and cost-effective manner.
To quickly compute real-time models, we selected a standard server with two Intel Xeon processors, both clocked at 2.53 GHz. The two processors have eight total cores. The comparatively low load caused by the current real-time models provides sufficient capacity for future extensions, even without hardware upgrades.
The I/O hardware is connected to the PC through a PXI expansion chassis. This occupies just one PCI Express slot, and the PXI chassis offers a sufficient number of free slots for additional I/O boards. The test bench uses NI PXI boards for controller area network (CAN) communication as well as analog and digital I/O. For certain time-critical signals, such as emulating speed sensor signals, we added an NI PXI-R Series field-programmable gate array (FPGA) module. We developed an FPGA program using NI LabVIEW FPGA software.
We also chose a signal conditioning unit with integrated fault simulation. This reduces the wiring complexity in the test bench without unnecessary signal quality degradation. To meet the requirements of a vehicle with two onboard voltage levels, we integrated two controllable power supplies into the test bench. A display shows the current load of the processor cores as well as relevant messages of the real-time system and the real-time models.
All components and the wiring of the combination HIL test bench were fully integrated into a 19-in. rack. In addition to validating ECU software, we can also use the arrangement of the test bench to test small-batch series modules such as carriers with ECUs. This is possible because we can connect the vehicle wiring harness directly to the test bench.
Requirements
The increasing complexity of controller functions also leads to increasing requirements on real-time plant models with respect to their capabilities and the modeled degree of detail. In particular, actuators in modern vehicles are increasingly operated in a controlled way rather than just in an on/off fashion. For this reason, we chose SimulationX from ITI.
In this project, we modeled all physical components interacting with vehicle controllers in SimulationX, including the following:
In contrast to preconfigured black-box solutions that are designed for real-time capabilities, physical models that are tailored for a particular task or derived from other real-time models are not generally capable of performing real-time tasks. Instead, their real-time capabilities are ensured by the modeler during model development.
The real-time capability of the models is achieved based on two main mechanisms. In one instance, a unique and thorough symbolic preprocessing is used. During code generation, SimulationX automatically preprocesses the physical and mathematical equations of the complete system model. It simplifies the system by resolving and substituting equations, reducing expressions that occur multiple times to one computation, and completely removing the computation of quantities that do not affect the specified interface signals (such as internal result variables). All this takes place without requiring user interaction and, in combination with further code optimization measures, results in very efficient real-time code. On the other hand, a number of analysis methods such as natural frequencies and vibration modes as well as energy distribution and performance analysis, assist the user in the model-performance optimization process and thus contribute to the fulfillment of all computation-time requirements.
Generally, the SimulationX models developed for this project have shown excellent performance. For example, the complete driveline models require only 20 percent of the computational capacity on one processor core, even though the model is implementing a comparatively high sampling rate.
The component models in the driveline are implemented in different degrees of detail, which correspond to the I/O requirements of the involved ECUs. From the engine perspective, a map-based model is sufficient for describing the engine behavior in a precise way. However, the injection system actuator requirements necessitate accurate plant modeling, from control input up to position sensor, and parameterization.
In this project, we validated this model part against the real injection control system. The gearbox and the torque converter are physically modeled and contain clutch and brake models that are parameterized through their friction characteristics. This permits the modeling of not only gear changes, but also the transient behavior during gearshifts such as speed gradients and gear change times. This is meaningful because the gearbox actuators are not only operated in an on/off fashion, but also in intermediate steps, that is, with varying brake and clutch torques. The remaining driveline model includes drive shaft elasticities, so it can exercise typical driveline-jerking vibrations. Depending on the steering angle, the curve radius is varied for each wheel so that the sensors detect differing wheel speeds during cornering.
In addition to controller output signals, the driveline model also processes brake torque provided by the brake system model and applies it to the wheels. The speed sensor outputs along the driveline, which have to be provided to various ECUs, cannot be generated by the real-time model as their signal frequencies are too high. Because of that, these signals are generated by an FPGA. The model only delivers the pulse frequencies of the wheel teeth passing the sensors.
The model shown runs on one processor core of the real-time system with a cycle time of 0.1 ms. Thereby it uses less than 20 percent of the core’s computational resources.
To fully take advantage of a HIL test bench, we needed a flexible test automation environment. Due to the extensive regression tests required for KMW’s in-house development, automated tests are indispensable for quality and cost reasons.
For this application, we used the test automation environment from TraceTronic, ECU-TEST. This tool is used to specify, implement, execute, and document the test case results.
The reusability of test cases saves valuable time for the user and is achieved by altering signal mappings for different development stages in the respective test environment. Tests are designed graphically without editing any source code manually.
Regression tests implemented in ECU-TEST cover the full bandwidth of required validation levels, ranging from low-level tests such as stimulating an ECU input and observing the respective response on the CAN, up to testing heavily interacting and complex functions such as fault management and fault recognition. This helped reduce the test efforts to just 15 percent of the previously required efforts. At the same time, the test depth increased significantly.
Producing state-of-the-art, highly protected, and comparably lightweight multipurpose vehicles with a lot of new functionality was only possible when using complex networked ECUs. The vehicle manufacturer bears the responsibility for the overall system, which consists of the vehicle, ECUs developed in-house, and ECUs obtained from external suppliers. In order to fully master this responsibility, all ECUs must be integrated and tested in combination so that they can be installed to the vehicle correctly the first time.
The novel HIL test bench is a unique combination of internationally established standard hardware and software components. As a result, the customer receives an optimally priced, highly scalable validation framework composed of the HIL test bench, tailored real-time models, and a highly automated test environment. This combination helps the manufacturer integrate the different vehicle ECUs in an optimal and cost-efficient way. Thus, the customer can fully exploit the scalability and I/O flexibility advantages. With real-time models in-the-loop, the AMPV’s ECU network can be validated quickly, providing an integrated approach to optimize the whole system. In this project, the test effort was reduced by 85 percent compared to non-HIL testing methods and simultaneously achieved a significantly higher test depth.
Using NI real-time hardware and NI VeriStand software, we performed the model development and HIL test bench integration very efficiently. We used the well-defined interfaces between models, test bench software, and hardware to develop activities in parallel on all three fields. The short learning curve of NI VeriStand helped us get our HIL test system up and running very quickly. The extensible environment provides assurance that we can scale our HIL test system to meet future needs. NI VeriStand is easily configurable, which allows the configuration to be modified as testing requirements changed, for example, when signals and models needed to be rerouted for debugging purposes. The native integration of NI VeriStand with real-time and FPGA hardware enabled the test system to meet necessary timing requirements and allows for future test expansion.
Andreas Abel
ITI
Webergasse 1 01067 Dresden
Dresden
Germany
Tel: T +49 (0) 351 – 260 50 0
Fax: F +49 (0) 351 – 260 50 0
abel@itisim.com