Generalized zero-shot fault diagnosis (GZS-FD) is a challenging and commonly encountered issue due to the coexistence of seen and unseen fault types. To deal with single and compound unseen faults and generate deceptive synthetic unseen fault samples, the bidirectional sensitive feature generation network (BSFGN) for GZS-FD framework is proposed, which operates through three sequential modules. First, the multidomain sensitive feature selection module selects the informative features through the dependencies of features from multidomain, which are then input to the BSFGN module to generate deceptive synthetic features for diverse unseen faults. Finally, the GZS-FD module leverages both synthetic unseen fault samples and real seen fault samples for fault diagnosis. Experiments on the designed feed drive system testbed validate the effectiveness of the framework, demonstrating superior diagnostic accuracy over state-of-the-art methods when considering single and compound faults.

Lithium-ion batteries in cold climates suffer from significant performance degradation, such as reduced available power and life cycle deterioration. To address this problem, a reconfigurable battery system (RBS) based self-heating method is proposed in this article. This innovative approach leverages a switch array integrated within the RBS, achieving self-heating with a fast temperature rise and low energy loss. Additionally, employing ac heating current with high frequency further mitigates damage to the battery. The modularized three-switch reconfigurable topology is proposed, and the ac-heating principle is designed. Also, based on the frequency-dependent characteristics of the battery impedance, the heating strategy is developed to accurately control the heating current. The experimental results demonstrate the efficient heating capability: the battery can be heated from −30 °C to 0 °C within 237 s by consuming only 6.27% of nominal capacity, and the battery capacity fade rate is only 0.47% after 200 heating cycles.

The job shop scheduling problem (JSSP) is a classic NP-hard problem. This article focuses on a realistic variant of the JSSP incorporating fuzzy processing times, with the objective of minimizing the maximum completion time. We propose a proximal policy optimization with graph transformer (GT-PPO) algorithm, which leverages proximal policy optimization (PPO) as the foundational framework, to address this problem for the first time. First, the intricate variability in states and actions often leads to suboptimal scheduling outcomes. To address this, we refine the representation of states and actions for improved performance. Second, to overcome inherent limitations of conventional graph neural networks (GNNs)-including difficulty in handling heterogeneity, over-squashing, and limited ability to capture long-range dependencies-we employ a graph transformer (GT) architecture for the first time in this study. These transformers effectively capture both the topological relationships in fuzzy disjunctive graph models and the long-range dependencies in large-scale JSSP instances. Additionally, we also reduce the computational complexity of the GT to $O(n)$ , enabling the agent to derive optimal scheduling solutions for large disjunctive graphs more efficiently, with reduced memory usage. Finally, the testing results demonstrate the strong robustness of our model across various scales of generated instances and public datasets after a single training session. Notably, on large-scale DMU and Taillard public datasets, the model exhibited exceptional robustness, further validating its effectiveness in addressing large-scale fuzzy JSSP.

To achieve fine control of multilayer temperature uniformity and energy consumption in a battery thermal management system (BTMS), a model predictive control (MPC) based on the reduced-order model and the heat generation previewer is proposed in this work. A direct contact liquid cooling battery pack is adopted to verify the control strategy. The control-oriented reduced-order model is developed for online multilayer temperature distribution acquisition. A heat generation predictor coupling with a dual neural network is integrated into the MPC controller to provide accurate future disturbances preview. The results indicate that the BTMS can be controlled to the target temperature with less overshoot. Besides, the temperature difference of the cell, module, and pack level can be limited to 0.8 °C, 1 °C, and 2 °C, respectively, decreasing the state of health difference among the cells. For energy consumption, the proposed method improves up to 56.48%.

Surface roughness is one of the determining factors for evaluating the quality of machined parts. However, the inevitable time-varying and uncertain characteristics in the actual machining process brings challenges to the construction of virtual metrology model. To address the problems of time-consuming training and low prediction accuracy in conventional virtual metrology models for surface roughness, a novel fuzzy echo state broad learning system (FESBLS) is proposed by introducing a reservoir with echo state properties to capture the dynamics of the machining process and then by employing incremental learning to reduce computational complexity and improve prediction accuracy. Besides, the effectiveness of the proposed method is validated by a grooving experiment and compared with benchmark approaches. Herein, the force signal collected during the grooving process and its fusion with cutting parameters are input into the FESBLS. The results show that the proposed FESBLS outperforms other models in improving the prediction performance. All in all, FESBLS is a promising technique for virtual metrology in machining processes.

In the large-area multidevice linkage-based laser processing, using the traditional control method cannot guarantee the processing accuracy of a permanent magnet linear motor (PMLM) system for a high-speed subject, due to unknown model uncertainties, backlash-like hysteresis, states constraints, and external disturbance. To address this shortcoming, this article develops a disturbance observe-based adaptive neural control for the PMLM system. First, aiming at the problem of unknown and uncertain system parameters, this study uses a radial basis neural network to estimate unknown functions of the system. Then, an adaptive variable is introduced to compensate for the effect of unknown backlash-like hysteresis, and the barrier Lyapunov function is used to restrict the motor from operating in a specified area. In addition, a nonlinear disturbance observer is constructed to adapt to the actual processing to reduce the influence of the external disturbance and the system load change. The proposed control scheme is verified experimentally, and the results indicate that when the proposed method is used, the PMLM system is consistently bounded by using the Lyapunov theorem. Finally, the simulation and experimental results verify the effectiveness of the proposed control strategy. The proposed method could be applied to laser-processing equipment.

Health prognostics for the machinery is a key objective of condition-based maintenance, with its primary goal being the remaining useful life (RUL) prediction. To efficiently structure prognostic data and address the challenges associated with uncertain operating conditions (OCs), this article introduces the digital twin (DT)-enabled RUL prediction system in smart manufacturing. The system mainly includes three layers, i.e., physical infrastructure layer (PIL), information interaction layer (IIL), and DT service layer (DT-SL). In the PIL, the multisource data are generated from different equipment and then transmitted to the IIL. In the IIL, the DT health prognostics information model is designed to organize the prognostic data in a structured manner. In the DT-SL, the virtual model, prognostic model, and sample generation model are constructed for the visualization of prognostic data, RUL prediction under uncertain OCs, and model validation. Finally, the effectiveness of the proposed system is experimentally demonstrated through two industrial cases, highlighting efficient prognostic data organization and accurate degradation tracking under uncertain OC scenarios.

Battery performance is significantly reduced at low temperatures, posing a challenge. To overcome this issue, the reconfigurable battery system (RBS) based hybrid self-heating (HSH) method is proposed in this article. This innovative approach leverages the flexible mode-switching characteristics of the RBS, achieving HSH with a high temperature rise rate and minimal energy loss. Additionally, employing square ac heating current with high frequency and low amplitude further mitigates damage to the battery. The physical configuration of the RBS- based HSH method is designed, and the modularized three-switch reconfigurable topology is proposed. Furthermore, the heating strategy is developed to further reduce battery fading. The experimental results demonstrate the efficient heating capability of this approach: the battery can be rapidly heated from –20 °C to 10 °C in just 239 s, consuming only 6.29% of the nominal capacity.

Accurate and reliable state-of-health (SOH) estimation is an important topic in battery management. Single data-driven model based SOH estimation suffers significant discrepancy problems over different cases. Moreover, existing ensemble based SOH estimation methods suffer serious problems, such as insufficient diversity of base models, complicated weight calculation, and severe overfitting. To address these problems, a stacking-based ensemble learning method for SOH estimation is proposed in this article. A second-level learner is used to integrate three heterogeneous base models without any weight calculation step. Fused datasets are generated by cross validation, maximizing the model generalization. Comprehensive validations are performed on batteries with two different cathode materials using two training strategies. The results show that the proposed ensemble method outperforms not only all base models (29% better than the optimal base model), but also the average method (more than 32%) and the state-of-the-art ensemble method (more than 44%).

Real-time monitoring and control of breakthrough stage are essential issues in ultrafast laser drilling (ULD) process. This article proposed a novel intelligent methodology to address the decision bias problem in monitoring and control of breakthrough stage. Eight time-domain statistical features were extracted after analyzing the correlation between the optical emission signal and the drilling process and data processing. To build the identification model of breakthrough stage, support vector machine method was adopted and the identification accuracy can up to 98.42%. Even so, there is a large deviation between the prediction breakthrough time by the identification model and the actual breakthrough time, which can up to 3.81 s. Considering the decision bias, the comprehensive Gaussian weight sequential three-way decision (CGW-S3WD) method was proposed for the first time. Compared to the actual breakthrough time, the mean decision time is only delayed by 0.375 s based on the combination of the identification model and CGW-S3WD method under simulated conditions. Finally, the method was applied to monitoring and control of breakthrough stage in the actual ULD process. Experimental results demonstrate that the control delay time is only 0.354 s and the high-quality holes and protection of the back wall can be achieved simultaneously. The research confirms that the method can reduce the decision bias and provides a new perspective for accurate monitoring of complex industrial scenarios.

Material removal rate (MRR) modeling is crucial for process and quality control in abrasive finishing processes. In this article, a novel broad echo state learning system (BESLS) is developed to predict MRR in the magnetically driven internal finishing process based on the force signal. This BESLS model employs reservoirs with echo state properties to replace the enhancement layer nodes of the conventional broad learning system. Three representative BESLS architectures are designed and experiments are conducted to validate the developed models. The effect of feature-fusion approaches on prediction accuracy is investigated. Results demonstrate that the prediction error of MRR is only 6.86% for the proposed BESLS model. The model has one reservoir with variable nodes (BESLS-III) and fuses extracted force features and process parameters as input. It also shows remarkable computational efficiency with a testing time of 0.134$\rm s$, making it a potential candidate for online process control.

Condition monitoring (CM) is the escort of smart manufacturing, which guarantees safety, precision, and efficiency of production process. However, traditional CM capacities with limited storage and computing resources cannot follow the explosive growth of manufacturing data. Empowered by the emerging Internet of Things, cyber-physical system, and cloud computing, cloud manufacturing (CMfg) characterizes a new service-oriented technology, which is reshaping the CM paradigm into an agile, scalable, and interoperable mode. In this article, a CMfg-based CM platform for smart manufacturing is proposed. A 5G wireless communication network-based edge computing system is presented to achieve data acquisition, feature extraction, and emergency response with ultrareliability and low-latency communication. In order to deal with the issue of data management and analysis, a generic open platform communications unified architecture information model oriented to the workshop is proposed in the fog layer. The cloud monitoring platform is constructed encompassing message delivery, data storage, status visualization, and training of artificial intelligence models. Finally, a CMfg-based CM system and a CMfg-based tool wear estimation system are outlined as two industrial cases. The experiments verify the feasibility and effectiveness of the proposed platform.

Battery performance will be dramatically reduced at low temperatures. To solve this problem, a hybrid self-heating method (HSHM) for batteries used at low temperature is proposed in this article. The HSHM owns features of low cost, high temperature rise rate, low energy loss, etc., which has the potential to be widely used to heat batteries. The physical and electric configuration of the HSHM is designed, and the working principles are analyzed. The heating strategy of the HSHM is then introduced to illustrate the heating performance. To validate the proposed method, the experimental workbench is established. Experimental results show that even with smaller heating current, the heating speed of the HSHM is faster, and less energy is needed to heat the battery. Compared with the traditional self-heating method, the performance of the HSHM is improved by 1.3 times for the temperature rise rate and improved by 55.6% for the energy loss rate, respectively.

This paper proposes an improved disturbance observer to realize accurate contact force estimation using the joint torque sensor. The joint torque sensor separates the dynamics of the link side from the motor side of the robot manipulator. Therefore, only computing the partial dynamics on the link side can realize external force estimation. This can considerably reduce the modeling workload and error terms that may affect the estimation results. Furthermore, this paper presents that the observed residual value during free motion can be considered as the friction dynamics, which is approximated by the neural network (NN) due to its inherent capacity in approximating nonlinear functions. After that, the estimation accuracy of the observer is considerably improved. Compared to other local NN approximation method, we analyzes the properties of the friction force in detail to select appropriate excitation trajectory for accurate global approximation results using only limited training data. We have presented that the suitable excitation trajectory and the use of global basis function are the sufficient conditions for global friction approximation. The proof of this theorem is also given. The Kalman filter is also utilized to reduce the noise of the estimation results in real time. The experimental results also demonstrate the efficacy of the proposed method, which accurately estimates the contact force for flexible joint robots.