Framework and basic parameters of a test bench for motor drive system of electric vehicle (EV) are illuminated. Two kinds of electric drive models, one was for the electric vehicle drived on real road, the other was f...Framework and basic parameters of a test bench for motor drive system of electric vehicle (EV) are illuminated. Two kinds of electric drive models, one was for the electric vehicle drived on real road, the other was for that on test bench, are put forward. Then, dynamic analysis of these models is made in detail. Inertia matching method of the test bench is researched and some useful formulas and graphs are brought forward. The experiment of an electric bus is introduced in order to explain the usage of this inertia matching method.展开更多
To improve the automation level of the vehicle drive axle test and better simulate a vehicle's actual operation, an advanced test machine has been developed. The load system of the machine consists of hand brakes and...To improve the automation level of the vehicle drive axle test and better simulate a vehicle's actual operation, an advanced test machine has been developed. The load system of the machine consists of hand brakes and electric cylinders. It is simple-structured and low-cost. The major motor of the machine is controlled by a transducer and its speed can be adjusted easily. In addition, the programmed machine can automatically test such parameters as the grinding condition, the differential speed, the noise level, etc. It can also adjust the test procedures according to different requirements. Detailed discussion of the structure and mechanism of the test machine is given in this paper.展开更多
人机共驾阶段人类驾驶员对驾驶环境保持较高的风险感知水平是保证及时有效、稳定安全接管的核心。本研究通过开展风险感知模拟驾驶试验,获取了驾驶员在典型汽车-动力两轮车碰撞场景下的驾驶行为及脑电响应数据。从驾驶行为层面以制动TTC...人机共驾阶段人类驾驶员对驾驶环境保持较高的风险感知水平是保证及时有效、稳定安全接管的核心。本研究通过开展风险感知模拟驾驶试验,获取了驾驶员在典型汽车-动力两轮车碰撞场景下的驾驶行为及脑电响应数据。从驾驶行为层面以制动TTC(time to collision)和平均加速度为评价指标,利用分位数回归构建了驾驶员风险感知量化模型,通过独立样本检验发现驾驶经验、碰撞场景类型对驾驶员风险感知存在显著影响。在脑电响应层面,通过双独立样本检验及FDR校正发现Alpha频段与驾驶员风险感知显著相关。此外,提出了驾驶员风险感知神经机理,包括视觉感知与认知加工两个阶段。研究结果有助于提升人机共驾汽车的安全性。展开更多
以深度神经网络(deep neural network,DNN)为基础构建的自动驾驶软件已成为最常见的自动驾驶软件解决方案.与传统软件一样,DNN也会产生不正确输出或意想不到的行为,基于DNN的自动驾驶软件已经导致多起严重事故,严重威胁生命和财产安全....以深度神经网络(deep neural network,DNN)为基础构建的自动驾驶软件已成为最常见的自动驾驶软件解决方案.与传统软件一样,DNN也会产生不正确输出或意想不到的行为,基于DNN的自动驾驶软件已经导致多起严重事故,严重威胁生命和财产安全.如何有效测试基于DNN的自动驾驶软件已成为亟需解决的问题.由于DNN的行为难以预测和被人类理解,传统的软件测试方法难以适用.现有的自动驾驶软件测试方法通常对原始图片加入像素级的扰动或对图片整体进行修改来生成测试数据,所生成的测试数据通常与现实世界差异较大,所进行扰动的方式也难以被人类理解.为解决上述问题,提出测试数据生成方法IATG(interpretability-analysis-based test data generation),使用DNN的解释方法获取自动驾驶软件所做出决策的视觉解释,选择原始图像中对决策产生重要影响的物体,通过将其替换为语义相同的其他物体来生成测试数据,使生成的测试数据更加接近真实图像,其过程也更易于理解.转向角预测模型是自动驾驶软件决策模块重要组成部分,以此类模型为例进行实验,结果表明解释方法的引入有效增强IATG对转向角预测模型的误导能力.此外,在误导角度相同时IATG所生成测试数据比DeepTest更加接近真实图像;与semSensFuzz相比,IATG具有更高误导能力,且IATG中基于解释分析的重要物体选择技术可有效提高semSensFuzz的误导能力.展开更多
文摘Framework and basic parameters of a test bench for motor drive system of electric vehicle (EV) are illuminated. Two kinds of electric drive models, one was for the electric vehicle drived on real road, the other was for that on test bench, are put forward. Then, dynamic analysis of these models is made in detail. Inertia matching method of the test bench is researched and some useful formulas and graphs are brought forward. The experiment of an electric bus is introduced in order to explain the usage of this inertia matching method.
文摘To improve the automation level of the vehicle drive axle test and better simulate a vehicle's actual operation, an advanced test machine has been developed. The load system of the machine consists of hand brakes and electric cylinders. It is simple-structured and low-cost. The major motor of the machine is controlled by a transducer and its speed can be adjusted easily. In addition, the programmed machine can automatically test such parameters as the grinding condition, the differential speed, the noise level, etc. It can also adjust the test procedures according to different requirements. Detailed discussion of the structure and mechanism of the test machine is given in this paper.
文摘人机共驾阶段人类驾驶员对驾驶环境保持较高的风险感知水平是保证及时有效、稳定安全接管的核心。本研究通过开展风险感知模拟驾驶试验,获取了驾驶员在典型汽车-动力两轮车碰撞场景下的驾驶行为及脑电响应数据。从驾驶行为层面以制动TTC(time to collision)和平均加速度为评价指标,利用分位数回归构建了驾驶员风险感知量化模型,通过独立样本检验发现驾驶经验、碰撞场景类型对驾驶员风险感知存在显著影响。在脑电响应层面,通过双独立样本检验及FDR校正发现Alpha频段与驾驶员风险感知显著相关。此外,提出了驾驶员风险感知神经机理,包括视觉感知与认知加工两个阶段。研究结果有助于提升人机共驾汽车的安全性。
文摘以深度神经网络(deep neural network,DNN)为基础构建的自动驾驶软件已成为最常见的自动驾驶软件解决方案.与传统软件一样,DNN也会产生不正确输出或意想不到的行为,基于DNN的自动驾驶软件已经导致多起严重事故,严重威胁生命和财产安全.如何有效测试基于DNN的自动驾驶软件已成为亟需解决的问题.由于DNN的行为难以预测和被人类理解,传统的软件测试方法难以适用.现有的自动驾驶软件测试方法通常对原始图片加入像素级的扰动或对图片整体进行修改来生成测试数据,所生成的测试数据通常与现实世界差异较大,所进行扰动的方式也难以被人类理解.为解决上述问题,提出测试数据生成方法IATG(interpretability-analysis-based test data generation),使用DNN的解释方法获取自动驾驶软件所做出决策的视觉解释,选择原始图像中对决策产生重要影响的物体,通过将其替换为语义相同的其他物体来生成测试数据,使生成的测试数据更加接近真实图像,其过程也更易于理解.转向角预测模型是自动驾驶软件决策模块重要组成部分,以此类模型为例进行实验,结果表明解释方法的引入有效增强IATG对转向角预测模型的误导能力.此外,在误导角度相同时IATG所生成测试数据比DeepTest更加接近真实图像;与semSensFuzz相比,IATG具有更高误导能力,且IATG中基于解释分析的重要物体选择技术可有效提高semSensFuzz的误导能力.