Search

Automatic assembly of elastic components is difficult because of the potential deformation of parts during the assembly process. Consequently, robots cannot adapt their manipulation to dynamic changes. Designing systems that learn assembly skills can help in alleviating the uncertain factor for industrial-grade assembly robots. This study proposes a skill acquisition method based on multi-modal information description to realize the assembly of systems with elastic components. This multi-modal information includes two-dimensional images, poses, forces/torques, and robot joint parameters. In this method, robots acquire searching, location determination, and pose adjustment skills using these multi-modal information parameters. As a result, robots can reach the assembly target by analyzing two-dimensional images with no position constraint. While acquiring pose adjustment skills, the reward function with depth and assembly steps is used to improve the learning efficiency. The deep deterministic policy gradient (DDPG) algorithm is applied for acquiring skills. Experiments using a KUKA iiwa robot demonstrated the effectiveness and conciseness of our method. Our results indicate that the robot acquired searching, location determination, and pose adjustment skills that allowed it to successfully complete elastic assembly.

The assembly process is a motion constrained by geometry and environment. The whole assembly process can be described as a series of transitions between contact states. There are many uncertain factors in the actual robot assembly environment, such as parts, robot motion and sensor information. The method with contact state recognition is widely used for assembly. At present, most work is independent for state recognition and action execution. On the one hand, the method of analysis and statistics is used to improve the recognition rate of state without the execution of assembly action. On the other hand, a variety of optimization methods are used to improve the control strategy. In this paper, a cognitive learning framework of “perception-action” mapping learning is proposed, which integrates contact state recognition and assembly action. The cognitive learning model of knowledge description of perception action mapping is constructed. The robot perceives and recognizes the contact state online, and updates the “state-action” experience knowledge base in time. The validity of the algorithm is verified by the example of low-voltage electrical appliance plastic shell assembly. The results show that the cognitive learning method based on “perception-action” mapping can sense the contact state of assembly online, which could accumulate and update experience knowledge base in time.

Tracked mobile robots (TMRs) will be affected by unknown interference force during operation such as interference forces of mobile robotic manipulator. The TMR dynamics model is difficult to be obtained accurately. In order to solve this problem, this paper uses a kind of dynamic steering resistance torque which can more accurately describe the force situation of the robot when turning. Based on this dynamic model, a dynamic adaptive sliding mode controller (DASMC) is proposed to track the desired speed and combine with kinematics controller (KC) to track desired trajectory. The simulation shows that this control method has strong robustness to the applied bounded external disturbance force. Compared with sliding mode controller (SMC), it can be found that the adaptive law can effectively estimate the disturbance and overcome the chattering problem of SMC.

The precision force sensing information of industrial robots to the environment is important for high-precision and flexible assembly operations. This paper proposed a calibration and calculation method to the contact force with the data from a six-dimensional force/torque sensor, which is installed between the end of the robot and the end-effector. In the proposed method, a multi-parameter coupled model was considered including the gravity of the end-effector, the zeropoint of the sensor and the installation angle of six-dimensional force/torque sensor. The contact force/torque could be precisely obtained using this model, and the influence of the gravity and zero drift of sensor can be compensated. Experiments were performed with KUKA iiwa. The results show that the force/torque information was more precise with the coupled model.

The perception of the vision and force modalities plays a crucial role in robot assembly. This paper describes a method to fasten assembly tasks based on vision and force sensing. It includes position estimation through vision and orientation adjustment through force perception. Position estimation is completed under visual guidance. Orientation adjustment is realized by distinguishing contact state between assembly parts. According to the state recognition, the fasten assembly could be completed. The proposed approach was demonstrated on the KUKA iiwa robot with the low-voltage apparatus assembly task. Results show that our system could accomplish the fasten assembly tasks including the robot position estimation and orientation adjustment using vision and force sensing.

Nowadays mobile manipulators are commonly used in assembly tasks, which can reach greater workspace but also cause more uncertainties. Uncertainty is one of the main factors affecting the quality and efficiency of assembly tasks. The framework of skill acquisition based on machine learning can make a good job of solving the uncertainties in mobile manipulation. In this paper, a method based on deep reinforcement learning is proposed to learn assembly skills. The Deep Deterministic Policy Gradient algorithm based on off-policy can scale to complex manipulation tasks. The safety constraint was set to the torque of each joint. The positive and negative reward function is designed to improve learning efficiency, which could be used in different assembly tasks. The deep neural network policies could be trained out efficiently on a real robot platform. This approach was performed on the KUKA iiwa robot arm equipped with 7-dof force/torque sensors. The results show that the robot could learn assembly skills properly with no prior knowledge.

In industrial robotic assembly process, the work surrounding environment is generally described by Contact State (CS). In order to solve the issue of contact state recognition, contact state classification based upon Extreme Learning Machine (ELM) is proposed in industrial robotic assembly process and extended the neural network to kernel learning in this paper. ELM has excellent performance in accuracy and speed of classification, owing to node generation being independent of training data, and ELM with kernel (ELM-kernel), based on the basic classifier ELM, is applied to classify contact state of the complex assembly process. Results have shown that contact state can be recognized by the proposed classification method and the classification performance of ELM-kernel is superior to ELM. In this way, more accurate information of contact states can be provided for the robot favoring the assembly tasks.

In industrial robotic assembly process, the work surrounding environment is generally described by contact state when vision-based systems fail for occluded parts. To solve contact state recognition problem, the paper builds an assembly process model based upon Support Vector Regression (SVR) and Particle Swarm Optimal (PSO), which maps the relationship between assembly contact state and robotic executive action. The established SVR model, whose parameters is optimized by PSO, used to predict the next motion of robot. The effectiveness and accuracy of the established model based on SVR and PSO are further demonstrated by experiments using fasten-assembly of circuit breaker in low-voltage apparatus automotive assembly. Results show that the proposed method is able to model the complex assembly process of low-voltage apparatus to construct the assembly rule base and lay the foundation for improving the flexibility and rapidity of small assembly.

In this paper, the stability and nonlinear vibration of an axially moving isotropic beam with simply supported boundary conditions are investigated. The equation of motion of the axially moving beam is established using the assumed mode method and Hamilton' s principle. For the linear system, the natural frequencies of the beam under different moving velocities are calculated by solving the generalized eigenvalue problem of the system to analyze the divergence and flutter velocities. For the nonlinear system, the bifurcation of the transverse displacement with respect to the moving velocity is analyzed and the vibration properties of the beam under harmonic excitation are presented. From the investigation, it can be seen that with the moving velocity increasing, the beam exhibits the divergence or the flutter types of instability. The beam may exhibit a transient stable state before flutter. The first order divergence velocity of the beam obtained from the linear analysis is accordance with the bifurcation velocity obtained from the nonlinear system. The nonlinear forced vibrations of the beam under different moving velocity are presented to show the effects of the moving velocity on the response amplitude. The existence of the transient stable state is also proved from the nonlinear forced vibration analysis.

Aiming at the non-stationary and time-varying characteristics of the diesel engine's fuel pressure wave signals, a method of extracting fault characteristic vectors by the energy of wavelet packet over the signals is presented. The nonlinear of neutral network makes fault diagnosis easier. According to the energy fault features and the kinds of faults, the network is constructed. The result of the simulation shows that this method can be used to classify faults of fuel oil system, and it is entirely practicable and reliable in the diesel fault diagnosis. In addition, it can also be applied to the faults diagnosis of other equipments.