工业机械臂作为工业4.0时代的重要设备，已广泛应用于诸多领域。工业机械臂的系统故障不仅会造成工业生产的经济损失，而且会对相关操作人员及设备的安全造成威胁，因此针对工业机械臂的健康检测技术展开研究，及时的发现其故障具有重要的工程意义。但是工业机械臂作为复杂的机电一体化系统，对其进行精确的建模相对困难，目前对该类系统的诊断多利用基于数据驱动的故障诊断方法，主要集中于单一部件单一工况下的故障诊断问题。为了实现工程应用，系统级、变工况下的诊断问题需要进一步讨论。针对该问题，本文开展了下述工作: （1）进行了工业机械臂建模与故障仿真。以典型的工业机械臂UR5作为具体的研究对象， 构建了MATLAB 与 CoppeliaSim的联合仿真模型。在此基础上，深入分析了工业机械臂的5种不同关节故障形式及原因，并对不同关节故障形式进行了仿真和分析。最后，利用仿真数据构建了机械臂关节仿真故障数据集，作为进一步分析的数据基础。 （2）提出了工业机械臂关节故障检测方法。分别实现了单一工况下和变化工况下的工业机械臂关节故障检测方法研究。针对单一工况下的关节故障检测问题，建立了以深度神经网络为基础的关节故障检测方法。在此方法基础上，针对变化工况下的跨域关节故障检测问题，通过引入领域适配策略，实现源工况与目标工况的适配。 （3）提出了工业机械臂故障关节定位方法。分别实现了单一工况下和变化工况下的工业机械臂故障关节定位方法研究。针对单一工况下的故障关节定位问题，建立了以深度神经网络与支持向量机相结合故障关节定位方法。在此方法基础上，针对变化工况下的跨域故障关节定位问题，通过引入领域适配策略，学习不同工况同类别数据的不变性特征，以及不同类别数据的区别性特征。
As an important equipment in industry 4.0 era, industrial manipulator has been widely used in many fields. The system failure of industrial manipulator will not only cause economic losses in industrial production, but also pose a threat to the safety of relevant operators and equipment. Therefore, it is of great engineering significance to carry out research on the health detection technology of industrial manipulator and find out its fault in time. However, as a complex electromechanical integration system, it is relatively difficult to conduct accurate modeling for industrial manipulator. At present, the diagnosis of this kind of system mostly uses the fault diagnosis method based on data drive, which mainly focuses on the fault diagnosis of a single component under a single working condition. In order to realize engineering application, the diagnosis problem of system level and variable working condition needs further discussion. To solve this problem, the following work is carried out in this paper :(1) Modeling and fault simulation of industrial manipulator are carried out. Taking UR5, a typical industrial manipulator, as the specific research object, the co-simulation model of MATLAB and CoppeliaSim is constructed. On the basis of this, five different joint failure forms and causes of industrial manipulator are analyzed, and the different joint failure forms are simulated and analyzed. Finally, the simulation failure data set of the manipulator joint was constructed by using the simulation data, which was used as the data basis for further analysis. (2) Proposed the joint fault detection method of industrial manipulator. The joint fault detection methods of industrial manipulator under single working condition and varying working condition were studied respectively. Aiming at the problem of joint fault detection under single working condition, a joint fault detection method based on deep neural network was established. On the basis of this method, aiming at cross-domain joint fault detection under varying working conditions, domain adaptation strategy is introduced to achieve the adaptation of source working conditions and target working conditions. (3) A joint location method for industrial manipulator faults is proposed. The method of locating the fault joint of industrial manipulator under single working condition and varying working condition is studied respectively. Aiming at the fault joint location problem under single working condition, a fault joint location method combining deep neural network and support vector machine was established. On the basis of this method, aiming at the cross-domain fault joint location problem under changing working conditions, domain adaptation strategy was introduced to learn the invariance characteristics of data of the same category under different working conditions and the distinguishing characteristics of different categories of data.
