Effective model based fault detection scheme for rudder servo system

The inherent nonlinearities of the rudder servo system（RSS） and the unknown external disturbances bring great challenges to the practical application of fault detection technology. Modeling of whole rudder system is a challenging and difficult task. Quite often, models are too inaccurate, especially in transient stages. In model based fault detection, these inaccuracies might cause wrong actions. An effective approach, which combines nonlinear unknown input observer（NUIO） with an adaptive threshold, is proposed. NUIO can estimate the states of RSS asymptotically without any knowledge of external disturbance. An adaptive threshold is used for decision making which helps to reduce the influence of model uncertainty. Actuator and sensor faults that occur in RSS are considered both by simulation and experimental tests. The observer performance, robustness and fault detection capability are verified. Simulation and experimental results show that the proposed fault detection scheme is efficient and can be used for on-line fault detection.     

模型数据相结合的电液伺服系统传感器早期故障检测与辨识

电液伺服系统中存在各类非线性因素和不确定干扰,传感器早期故障引起的系统状态量变化往往小于干扰引起的变化,因此很难发现。本文提出了一种模型和数据相结合的电液伺服系统传感器早期故障诊断方法,通过基于模型的方法对系统中难以准确测量的外干扰力进行有效解耦,通过建立观测器组对系统中常见的传感器故障进行辨识,针对系统中的建模误差、测量噪声等剩余不确定干扰,采用基于神经网络的方法进行模型补偿,通过建立补偿规则,可进一步提高对早期故障的敏感性。仿真和实验结果验证了所提出的混合式故障诊断方法的有效性。     

动态环境下基于改进几何与运动约束的机器人RGB-D SLAM算法

室内移动机器人使用传统视觉SLAM算法在动态场景下进行位姿估计时精度低、鲁棒性差,其主要原因是错误地将运动的特征点加入了相机位姿计算.为了解决这一问题,本文将特征点分为静态特征点、状态未知点、可疑静态特征点、动态特征点和错误匹配点.其中,静态特征点使用严格的几何约束进行筛选,并将状态未知点使用多帧的观测信息区分为可疑静态特征点、动态特征点以及错误匹配点,并进行卡尔曼滤波.最后,将静态特征点、可疑静态特征点和动态特征点一起加入位姿优化.利用TUM数据集,在室内存在运动物体的场景下进行实验.结果表明,所提出的算法在动态场景下的综合性能(包括精度、鲁棒性、运行速度)要优于其他动态场景下的SLAM算法.