In the aerospace field, electromechanical actuators are increasingly being implemented in place of conventional hydraulic actuators. For safety-critical embedded actuation applications like flight controls, the use of...In the aerospace field, electromechanical actuators are increasingly being implemented in place of conventional hydraulic actuators. For safety-critical embedded actuation applications like flight controls, the use of electromechanical actuators introduces specific issues related to thermal balance, reflected inertia, parasitic motion due to compliance and response to failure. Unfortunately, the physical effects governing the actuator behaviour are multidisciplinary, coupled and nonlinear. Although numerous multi-domain and system-level simulation packages are now available on the market, these effects are rarely addressed as a whole because of a lack of scientific approaches for model architecting, multi-purpose incremental modelling and judicious model implementation. In this publication, virtual prototyping of electromechanical actuators is addressed using the Bond-Graph formalism. New approaches are proposed to enable incremental modelling,thermal balance analysis, response to free-run or jamming faults, impact of compliance on parasitic motion, and influence of temperature. A special focus is placed on friction and compliance of the mechanical transmission with fault injection and temperature dependence. Aileron actuation is used to highlight the proposals for control design, energy consumption and thermal analysis, power network pollution analysis and fault response.展开更多
电力设备在运行维护中积累了大量包含重要实体信息的故障文本,然而文本实体边界模糊、术语较多等特点导致传统实体识别方法训练效率低下,效果难以提升。为此,该文提出一种新的实体识别方法I-BRC(integrated algorithm of BERT based BiR...电力设备在运行维护中积累了大量包含重要实体信息的故障文本,然而文本实体边界模糊、术语较多等特点导致传统实体识别方法训练效率低下,效果难以提升。为此,该文提出一种新的实体识别方法I-BRC(integrated algorithm of BERT based BiRNN with CRF)。该方法采用字嵌入模型将文本逐字转化为字向量序列以避免分词处理带来的误差累积;利用循环神经网络与概率图模型对文本的序列特征信息进行抽取;集成多个单一类型实体识别器分别独立学习不同类型实体的特征并采用并行预训练机制提升算法训练效率;最后利用多类型识别器对识别结果进行整合。此外,通过调整单一类型实体识别器可以灵活机动地应对不同电力设备的实体识别任务,避免重复训练,节省计算资源。实验表明,所提出的I-BRC仅需3次迭代就可收敛,训练效率大幅度提升;且该模型的F1值、精确率、召回率分别达到了88.0%、86.8%与89.2%,相比传统模型性能提升了7.5%~29.3%,验证了所提模型的有效性与可行性。展开更多
基金supported by the Aeronautical Science Foundation of China (No. 2012ZD51)the support of the China Scholarship Council (CSC)
文摘In the aerospace field, electromechanical actuators are increasingly being implemented in place of conventional hydraulic actuators. For safety-critical embedded actuation applications like flight controls, the use of electromechanical actuators introduces specific issues related to thermal balance, reflected inertia, parasitic motion due to compliance and response to failure. Unfortunately, the physical effects governing the actuator behaviour are multidisciplinary, coupled and nonlinear. Although numerous multi-domain and system-level simulation packages are now available on the market, these effects are rarely addressed as a whole because of a lack of scientific approaches for model architecting, multi-purpose incremental modelling and judicious model implementation. In this publication, virtual prototyping of electromechanical actuators is addressed using the Bond-Graph formalism. New approaches are proposed to enable incremental modelling,thermal balance analysis, response to free-run or jamming faults, impact of compliance on parasitic motion, and influence of temperature. A special focus is placed on friction and compliance of the mechanical transmission with fault injection and temperature dependence. Aileron actuation is used to highlight the proposals for control design, energy consumption and thermal analysis, power network pollution analysis and fault response.
文摘电力设备在运行维护中积累了大量包含重要实体信息的故障文本,然而文本实体边界模糊、术语较多等特点导致传统实体识别方法训练效率低下,效果难以提升。为此,该文提出一种新的实体识别方法I-BRC(integrated algorithm of BERT based BiRNN with CRF)。该方法采用字嵌入模型将文本逐字转化为字向量序列以避免分词处理带来的误差累积;利用循环神经网络与概率图模型对文本的序列特征信息进行抽取;集成多个单一类型实体识别器分别独立学习不同类型实体的特征并采用并行预训练机制提升算法训练效率;最后利用多类型识别器对识别结果进行整合。此外,通过调整单一类型实体识别器可以灵活机动地应对不同电力设备的实体识别任务,避免重复训练,节省计算资源。实验表明,所提出的I-BRC仅需3次迭代就可收敛,训练效率大幅度提升;且该模型的F1值、精确率、召回率分别达到了88.0%、86.8%与89.2%,相比传统模型性能提升了7.5%~29.3%,验证了所提模型的有效性与可行性。
