The 2016 SAECCE 2016 was held on October 26-28. The theme of the forum was “Technology Upgrade, Industrial Development, and Collaborative Innovation”. Together with more than 150 industry leaders and leaders of the society, Academician, more than 100 authoritative experts of "Thousand Talents Program" in automobile and related industries, and more than 800 domestic and foreign industry guests discuss the collaborative innovation and technology development path of auto industry depots and parts. The following is the speech of Xu Weiqing, chief engineer of the Active Safety System of Pan Asia Automotive Technology Co., Ltd. at the meeting:

I am very happy to share our development experience with you here. Today we mainly talk about simulation. In the research and development of the ADAS field, some help from the simulation system to us, including the ROAD-to-LAB to MATH transition, started as a real vehicle test, and later we got to LAB and how to get to MATH.

Specific to our R & D, first based on the road test and development, behind the vehicle we are doing in the ring development platform, and then to the hardware in the ring development platform to the model in the ring development platform. Through these four steps, we hope to further optimize control, enhance our experimental safety, and repeat test scenarios.

We did a lot of road-based testing and development. The active safety system was not a closed system. Many questions were asked in front. The driver was not satisfied with the brakes. The driver braked when the brakes were too late. The driver did not brake. These problems cannot be logically solved. In fact, we can only answer this question on the road. With the current technology on the current road, plus the current Chinese driver's habits, what kind of strategy is better because we can't solve the problem perfectly. problem.

Basically, we ran through the better-selling cities that the Ministry of Marketing had chosen to sell. We found that there is a significant difference between China and the United States. There are fences in our roads and we fear that pedestrians will wear roads, and the shape of different fences will be different. This shape has caused interference to our radar. Through such a road test, we have largely avoided the misdirection caused by the fence. Through this test, we choose active safety technologies that are suitable for China's working conditions. During the development process, I found that many real data do not know. When I finished the road test, I discovered that, maybe we didn't have time to do a lot of in-depth work. We found that the scene quality is very large, I really found a problem, and you solve the problem. It is difficult to duplicate the scene and find it hard to get to the local area and reproduce the local accident.

For example, if there is a problem with a fence, I can't go to Jiangxi every time. After that, I contacted the local supplier and bought the fence. This way, our Guangde testing plant directly tests the fence. It is still difficult to reproduce this issue in Guangde. Real car testing is always dangerous. In some cases, especially in the case of brakes just mentioned, we are very difficult to tell the driver in the real road test. You try not to brake, let me see the reaction of the system, which is obviously It is impossible. This has caused our public roads to be tested. It is time-consuming and inefficient.

On this point, the first thing we think of is that we should not take a balloon test on dangerous road conditions without trying on real roads.

This is a test scenario for a balloon car. Even if it is not very convenient to test with a balloon car, my first thought is whether or not the sensor can be virtualized. This way I don't need the balloon car. I used Prescan's simulation test software to use a PC. In the simulation test scenario, the radar module is used to tell our radar module the relative positions, relative speeds, and relative accelerations of obstacles in the virtual scene. Our ECU can judge the current working condition of the vehicle based on this virtual obstacle information. Need to brake, accelerate or how to, through the vehicle's CAN interface, the vehicle's operating conditions, such as the speed, speed, and steering IMU and other signals, feedback to our PC, forming a closed loop. In this closed loop, since the car is not virtual, only a simple vehicle kinematics model is needed in the virtual environment, so that the various reactions of the entire vehicle are real.

This is a new LaCrosse, we first built the radar model, the interaction interface between the simulation software and the real car, the environment of our real car, the right side is the human-computer interaction interface, our actual car map, the high-performance notebook in the upper right corner The computer mainly runs Prescan's simulation environment. In the lower left corner is an obstacle for the driver to see the virtual scene. In the lower right corner, the PAD is a human-machine interface. Various scenarios can be selected, and even different initial speeds and starting conditions. All kinds of scenes need the man-machine interface, which is implemented in the PAD in the lower right corner.

One result of our test.

The balloon car I mentioned earlier, these balloon vehicles are more expensive and they are hundreds of thousands of yuan, and they do not look like they have crashed. Occasionally, they crash, and they are expensive to repair, so it is impossible to do it with balloons. Test, especially for the balloon car to make UNcake scene, the front rail car is also easy to hit the deformation.

Therefore, with the use of virtual scenes, we can avoid the problem of frequent use of a balloon car. In the above scenario, after a car comes over, we urgently brake. In the same scene, we can test on the road. The following are tests conducted by the Guangde Testing Machine Factory. Both cars are due to Prescan's simulation model, telling me the virtual obstacle information, and then the radar makes the vehicle move according to the virtual obstacle information. In addition to the falsehood of the obstacle, the performance of the whole vehicle is real, especially the above figure. It can be tested in our Pan-Asian Technology Center. We can start to improve our efficiency in the room without leaving the room. The entire scene can be repeated. Our various algorithms are more convenient. If we test on the real road, we can Guangde two cars took the intercom, and said that in front of you a car brake, a car in front of you is a human operation, certainly can not repeat the scene, because everyone reacts differently. If the front is to take a robot test, but the robot is more expensive, it is more troublesome to set up, generally if the test is two or three hours, and then the robot is removed, the overall efficiency is very low.

The following test differs from the above. The following test on real roads shows that the entire tire attachment technology to the ground is real. The driver is seated inside and all the feelings of pushing back are real. The following test can know the reaction of the entire driver, and feel comfortable and uncomfortable. Some of the questions can be answered just now. I brake early and the driver feels that he shouldn't brake early. He should play the steering wheel or something. This is a little bit missing. What we did a little bit unreal is that the obstacles are virtual. The screen in the upper left corner is not as realistic as the balloon car, but I believe that if VR and other technologies are available in the future, Through the breakthrough, the driver can certainly feel the virtual scene more realistically.

After this step, the next hardware obstacle platform, there are two in-car video. The road ahead is actually blank. We can see the car on the screen. We set 30 km/h to see the obstacles before we can brake. The feeling inside the car is very real.

Our ACC corners follow the car, in the virtual scene there are different curvature and follow the car environment, there is no lane in the square, the driver to open according to the lane in the screen, so as to develop comfortable features. If tested on the real road, the real driver control of the front two cars, three cars interact, and the walkie-talkie must be indiscriminate.

Just test the test under the virtual scene, test the data, for example, the vehicle speed of the target in the virtual scene, the speed of the car, and the CAN signal of the acceleration are completely known. This can do in-depth qualitative analysis, unlike the test on the road. The car signal is true, but the signals of the other car are often unknown. In many cases, very qualitative tests cannot be done.

The front measurement is a static one. The second vehicle in front is running at 10kph and the main vehicle is running at 33kph. This will result in the situation where the first scene is successfully braked. The second scene, though braked, unfortunately crashes into it. , Obviously there is an error with intuition, intuition on the car 33KPH, the car opened 10KPH should be easier to avoid collision, the car is easy to avoid collision static, intuitively useless, R & D process, in many cases is based on the actual test Looking at the R&D calibration and the algorithm itself, is there an unsatisfactory situation? It is very convenient to reproduce the conditions in the virtual test to see if there are such problems. Before this system was installed, the vehicle was tested on the road. Occasionally, we found that we solved the problem. With a large amount of data, there are no BUGs if we open tens of thousands or hundreds of thousands of kilometers. If we open a thousand kilometers, the problem is solved, and other problems may be caused. The BUG does not know.

In the car virtual scene, the typical scene is determined. Every time a relatively large set of targets or algorithms is changed, each test can be repeated. Perhaps the new problem is solved, but the old scene does not appear new. problem.

Before the development of performance, care about whether this function can be achieved, the function is realized that there is no BUG between each module between the respective software algorithms, there is no BUG between the respective modules.

Obstacle information was simulated with Prescan, but the controlled object vehicle was moved to the laboratory. In order to do this well, the vehicle dynamics model was used to simulate the actual vehicle. The more important is the Bosch, Bosch, circled in the lower right corner. The EPS and the engine control module can directly perform the closed-loop operation, directly affecting the performance of our BANH. At present, our key work is to open the system. The process of doing the simulation calibration should try to make this system close to the previous response. The advantage of this system over the previous system is that the R&D process is relatively early, and the system can be verified before we build a vehicle; whether there are BUGs between the two modules in the laboratory environment, and whether the communication between the two modules will be lost. This is a test of the benefits of this system.

The model in the ring development platform is equivalent to the controller and actuator are virtualized, and the control algorithm is directly measured in the simulation environment. This start is later than the previous two. Before the general Pan-Asian development of the car, the ADAS system was mainly used to select the world famous Delphi, Bosch these suppliers. We stand at the perspective of the entire vehicle and build the system of the entire vehicle. After the specifications of the sub-components are designed, they are outsourced to the component suppliers. In the future, there will be no driver’s fire in automatic driving. There are many sensors throughout the driverless vehicle and the control objects are very complicated. Volvo CEOs claim that the production of self-driving cars is in full responsibility. This makes it difficult to package one system for the same supplier. If outsourced to the same supplier, this sensor is outsourced to supplier A and another sensor is packaged to the supplier. B, this control algorithm basically requires the OEM to pay for itself, so we now build a model in the ring development platform. This will enable us to develop better control algorithms on the premise that components have not yet been selected and sensors have not yet been selected. The control algorithm selects what kind of sensor, which can meet the driverlessness of our control algorithm time, and discusses with the supplier to develop a better driverless system.

The summary that was just made now is basically based on the collection of Chinese roads and the setting up of scenes so that we know that there are many differences between Chinese scenes and foreign scenes, especially the driving habits of Chinese drivers. At the same time, a lab-in-the-loop simulation system can greatly improve the accuracy and repeatability of our algorithm development. Finally, we have also done a lot of smart driving vehicle road calibration, and the three roads promote each other, many smart driving roads The calibration can verify whether the laboratory's simulation system in the ring is true. It can be done at the beginning of the calibration work and then on the real road. This can relieve a large part of the initial calibration.

We have seen various strange scenes on the road, we can also build in the virtual scene library just now. Finally, we said the reverse of the verification, previously said Road to Lab to Math, behind the time of validation is definitely from Math to Lab to Road. Initially, simulation tests emphasized the importance of automated testing, verification algorithms, and parameter calibration, and the various scenarios can be quite complex and qualitative. But doing laboratory tests can be done more objectively. Finally, it is the venue. Testing at the Guangde testing plant is a subjective feeling and an objective evaluation. Just now that the limitations of the test, it is impossible to get the scenario like the previous two tests. So complex and qualitative, the last large-scale road test can not be avoided, telling us the situation of misbehavior under real conditions, as well as the user experience.

The above is our simulation test method and development method. thank you all!

Moderator: Thank you Mr. Xu Weiqing, do you have any questions?

Question: Xugong Hello, I would like to ask three questions, the simulation of the vehicle inside the ring, the main purpose of the simulation test is to measure the strategy or the driver's feelings?

Xu Weiqing: Both have to test, the essence of the vehicle in the ring test is repeated after the virtual obstacles in front of the test, this time the algorithm must be tested, since the vehicle is true, if I have to go to the Guangde test car factory to test the driver's experience , but if in the hub laboratory, that convenience, I do not have to go there, when the driver feels no way to measure, it pays more attention to algorithm testing.

Question: Is it a high demand for various scenarios?

Xu Weiqing: Analysing C-NCAP and Euro-NCAP, we found that there are BUG built-in scene libraries. Every time we do a major algorithm and calibration improvement, we must take the scenario again and see if there are any unexpected results.

Question: During the Lab test, you opened both the EPS and the BCN. The mechanical components, the straightener, the steering system, and the brake system all use real machines.

Xu Weiqing: Yes, especially the EPS is very large. Our test is still quite large. In order to support the EPS and make the straightener behave like a real one, the hydraulic pressure and piping output from the straightener are as real as the real car.

Question: Full virtual simulation, vehicle dynamics, how do you consider CARCIM here?

Xu Weiqing: We currently use CARCIM, and North America also uses CARCIM. There is also communication between these softwares. In particular, the active security scenario of CARCIM is established. It may not be necessary to cooperate with Prescan. CARCIM itself can establish these things. At present, we think of Prescan. Scene, sensor modeling may be better and more realistic.

Question: In your software testing, how is the benchmark between software testing and real vehicle testing considered?

Xu Weiqing: Everyone hopes that the simulation environment, like the hacking empire, is exactly the same as the real environment, but in fact it cannot. We basically do the real cars and the simulations go into simulations. Naturally, we find two different places. Although these are not the same places, it makes us sad, but there is no way to do that. We basically can't make up for it. To such a balance, there is no way behind the defects. We can only overcome it. We have three paths that we have been walking on and we have found that some scene libraries that we should build are built into it. The error can't basically solve this problem. That kind of BUG still needs time and labor to be reproduced on the real road. In some scenes, this problem is perfectly solved. Of course, with this method, the fuel consumption test is basically completely related to the real road fuel consumption. The test is exactly the same, it certainly will not go to the real road.

Question: What are the suppliers doing during the process of building the platform?

Xu Weiqing: Work supported by major suppliers, radar sensor modeling.

Question: Is this piece of VIL a piece of vendor perception and interface?

Xu Weiqing: Yes.

Question: During the development of the HIL platform, is it a stand-alone radar HIL or a fusion of the two?

Xu Weiqing: Fusion.

Question: The camera and radar are integrated programs?

Xu Weiqing: Yes, it is integrated into the algorithm. The radar and camera fusion algorithm is a supplier. The supplier algorithm is in the module. In this module, the HIL is real. The only thing that is not true is the camera. Based on the mobile solution, the mobile solution does not communicate with Prescan. , Reluctant to open the Prescan program, we do the mainstream method only the camera to the screen. The degree of authenticity in this area has been greatly discounted.

Question: This part of the hub laboratory is a horizontal control strategy?

Xu Weiqing: We are trying to do this, we are currently doing the wind hub, most of the hub is vertical without any horizontal, we will consider the horizontal next, consider the horizontal there is the pressure of cost, once there is a good In the cost plan, the tires are removed and the vehicle is lifted up with the vehicle. The entire vehicle torque is used. This cost is too high. This is only used by the general power department. The following turning hubs have been greatly improved. Some of the hubs are like tank tracks. Below, a large amount of wind is used to blow up the car through the track. The track is flat, not a feeling of pressure, and it guarantees that there is a very real road surface. A large coincidence, but this program is also very expensive, and there are currently such considerations.


This product(Concrete Batching Plant) is a fully automatic concrete mixing equipment consisting of five systems of mixing, electrical control, storage, conveying and weighing.
Features of concrete batching plant
1.It is assembled by building blocks, which is short, quick and convenient.
2.It is convenient and quick to use the cylinder discharge or electric push rod to discharge.
3.A variety of mixing processes, highly automated, high production efficiency,
4.Environmental protection and energy saving, using pulse dust removal, recycling waste-water for reuse

Concrete Batching Plant

Concrete Batching Plant,Ready Mix Concrete Batching Plant,Stetter Concrete Batching Plant,Portable Concrete Batching Plant

Shandong Zeyu Heavy Industry Science and Technology Co.,Ltd. , https://www.zeyuconcretemachine.com