Xincheng Tian

OOD detection mechanism is important when en-countering new (out-of-distribution, OOD) faults. Current OOD detection methods are often coupled with diagnostic models, lim-iting their actual deployment. Moreover, they usually assume that OOD faults exhibit significant distributional differences from the in-distribution (ID) data, and the ability to magnify differences is insufficient, affecting the detection ability. There-fore, we propose a reconstruction-based method using deep com-pressed sensing (DCS) and contrastive learning (CL), which is plug-and-play and can magnify the differences without OOD fault samples during training. The frequency-domain signal first (1) passes through the measurement module initialized by CL to align ID semantics, then (2) is reconstructed differently by the reconstruction module centered on global-local feature recon-struction blocks with local feature frequency-domain attention (LFA-GLFR blocks). (3) With the optimization objective of joint reconstruction accuracy, sparsity, and ID error consistency, the parameter training is guided to move in the direction of increas-ing differences to achieve better discrimination. We conduct ex-periments on the Shandong University (SDU) bearing dataset and the University of Connecticut (UoC) gearbox dataset. On the former, its average AUROC, AUPRC, F1-score, and F0.5-score are 0.1303, 0.1645, 0.0850, and 0.1156 higher than those of the compared state-of-the-art (SOTA) methods, demonstrating its advancement.

Within the domain of robot assembly manipulation, contemporary skill-learning approaches exhibit significant generalization capabilities, enabling them to adapt to a broad spectrum of tasks and dynamic environmental conditions. However, these methods suffer from catastrophic forgetting when applied to various robot manipulation tasks, leading to suboptimal performance and inadequate adaptability in real-world scenarios. To this end, a continual learning strategy that incorporates both knowledge sharing and model acceleration is proposed for multiple robot assembly tasks. First, a common experience pool is established that enhances data efficiency by selectively revisiting past task experiences during the learning process. This mechanism mitigates forgetting by preserving critical knowledge from previous tasks. Second, incorporating the effects of the old model, the new model can converge more rapidly. The weights of the old and new models are learned through interaction with the environment. Experiments on various assembly tasks are conducted to verify the continuous memory retention and efficient learning capabilities of the proposed method in both simulation and real-world environments.

As the core component in intelligent manufacturing systems, the precise control of the welding clamp’s electrode pressure plays a decisive role in ensuring the quality of spot welding. This paper proposes a novel pressure control strategy for robotic welding clamp based on partitioned adaptive RBF neural networks: (1) Deformation of the clamp body can lead to deviations in workpiece positioning. To address this issue, a deflection compensation method for robot welding clamp based on the PSO-RBF neural network is proposed. By leveraging pre-calibrated empirical data, the intrinsic mapping relationships are identified, and the derived deflection compensation value is integrated into the real-time position command of the robot end-effector. (2) During electrode motion, the system is subjected to external disturbances such as friction and gravitational forces. So, a sliding mode control strategy incorporating adaptive RBF disturbance compensation is proposed to achieve robust speed regulation. Furthermore, the electrode’s reference velocity is dynamically adjusted based on the welding force error and improved admittance control algorithm, enabling indirect regulation of the welding force to reach the desired set value. The results demonstrate that the proposed composite control strategy reduces electrode pressure overshoot to less than 5% and enhances steady-state control accuracy to ±1.5%.

The compressed sensing (CS) methods have two issues when processing vibration signals of rotating machinery: (1) the data-driven method and signal priors (e.g., global/local and structure priors) are not combined well, resulting in the reconstruction not converging to high-precision results; (2) the current metrics only focus on numerical accuracy and do not consider PHM scene fitness. We propose a deep compressed sensing framework (RMVS-DCS): (1) Global-Local Feature Reconstruction (GLFR) blocks alternately reconstruct features at global and local scales. Pre-train the output layer by dictionary learning, introducing the signal structure prior to guiding the model to converge to a better one. (2) A dual evaluation system, combining the numerical indicators and prognostics and health management (PHM) task accuracy, helps the model selection in PHM scenarios. On the rotor unbalance data set, its reconstruction and prediction error are reduced by 82.60% and 89.90% (compression ratio is 0.5), demonstrating the advanced reconstruction precision and PHM compatibility of the framework.

For cylinder production, visual inspection of surface defects on cylinder joints holds great significance, yet the task is challenged by the high reflectivity on joint surfaces and the presence of numerous small defects. To address this issue, a metal surface defect detection network, named Information Augmentation and Multi-scale Feature fusion with YOLO (IAMF-YOLO) based on YOLOv8, is proposed in this paper, aiming to enhance the accuracy of joint surface defect detection. The IAMF-YOLO detection network incorporates an information augmentation strategy into the backbone network and introduces a novel backbone network called Information Augmentation Network (IAN). This IAN reduces the information loss during feature down-sampling extraction. Furthermore, to boost the detection performance of small defects on joint surfaces, improvements have been made to the Path Aggregation Network (PAN), and a Parallel Multi-scale Feature Pyramid Network (PMFPN) is introduced. PMFPN is designed to separate feature fusion into two stages: an initial feature fusion stage and an incremental feature fusion stage, of which the latter one further merges the features of the former one. Through ablation studies and comparative experiments on IAMF-YOLO, it is demonstrated that this model surpasses YOLOv8 in terms of detection accuracy and precision. Specifically, it achieves an accuracy of 88.6% on a self-constructed dataset, 91.0% on the NEU-DET dataset, and 84.2% on the GC10-DET dataset. Furthermore, this study demonstrates the robustness of the proposed model against environmental disturbances through actual defect detection under varying lighting conditions.

Traditional assembly tasks often require robots to transfer the acquired skills to new tasks. However, previous transfer reinforcement learning methods typically ignore the inherent relationship between the source and the target domain tasks. This requires a substantial amount of interaction data to compensate for this deficiency, and generally results in poor transfer effects. To address this issue, a strategy transfer method that establishes a shared feature space between the source domain and the target domain is proposed to enhance the efficiency of strategy learning on peg-in-hole assembly. Initially, by calculating the distance between each feature in the source and target domains, the features with small distance are selected as shared features. Subsequently, in order to determine the successful search state, this paper uses the jump state of contact force and the relative position between the peg and the hole as the judgment criterion. Lastly, search and insertion peg-in-hole assembly experiments are conducted to validate the generalization of the proposed strategy, demonstrating its capability to transfer from simulation to the real world.