The measurement of dynamic forces within spacecraft sources is critical for safeguarding mission accuracy and stable operation. Two major challenges impede accurate measurement of dynamic forces from large sources: the severe nonlinearity of platforms; leading to inaccuracies in traditional methods; and the explosion of redundant sensor combinations, lowering optimization efficiency. To address these issues, this article proposes an intelligent framework that synergistically integrates the genetic algorithm with adaptive-learning-rate-based back propagation (GAARBP) and Bayesian optimization (BO). The GAARBP module addresses the nonlinear input-output mapping problem of the platform, and enhances the efficiency and accuracy of model training through adaptive learning rate adjustment. The BO module employs a blockwise Matérn kernel function tailored for sensors on horizontal and vertical planes to construct a Gaussian process surrogate model, alongside an expected improvement-based acquisition function for efficient optimization. These two modules operate in a closed loop, wherein GAARBP provides accurate fitting errors to guide BO’s selection, and BO feeds back optimal sensors to enhance GAARBP’s model training. Experimental results show that the framework achieves higher measurement accuracy (error: 2.28%) and superior decoupling performance (error: 1.13%), sufficient to satisfy the measurement requirements of large vibration sources in spacecraft. Moreover, compared with conventional methods (exhaustive search, D-optimization, particle swarm optimization, etc.), the proposed framework selects the best sensor combination more effectively for the optimization problem of six sensors selected from 24, with outcomes that are virtually globally optimal. This article offers a novel, effective approach to synergistically solving nonlinear dynamic force measurement and sensor optimization in complex environments. It thereby promotes the intelligent advancement of testing technologies in these fields.

This paper proposes an innovative virtual chain-based kinematic calibration for the 4PPa-2PaR parallel manipulators with subchain architectures. Conventional calibration methods for such architectures suffer from inherent limitations due to coupled parameter constraints and restricted solution spaces caused by joint displacement and structural parameter dependencies. The presented methodology introduces three fundamental advancements: (1) a novel parameter assignment strategy enabling independent joint/link parameter definition across different kinematic chains, (2) systematic transformation of constrained optimization into an unconstrained one, and (3) significant expansion of error parameter solution space through virtual chain modeling. Comparative experiment on the physical prototype demonstrate improvements in both orientation and position accuracy compared to existing methods.

Most bioinspired cable-driven continuum robots (CDCRs) usually employ a flexible backbone to realize the continuous deflection. For the CDCR to merely produce bending motions, its flexible backbone has to be designed with low bending stiffness but high tensile and torsion stiffness. In this article, a pattern-based design approach is employed for the flexible backbone, which adopts rectangle-shaped patterns inspired by elastic couplings. As it is rather difficult to derive accurate analytical stiffness models for such a pattern-based backbone structure with large nonlinear deflections, a novel data-driven stiffness modeling approach is proposed. The Gaussian process regression method is employed to train the stiffness model with respect to structure parameters of the backbone, while the dataset is generated through a commercial finite element analysis software package. To narrow the distribution of the training data and make the predicated stiffness values always positive, the natural logarithm transformation is utilized for data preprocessing, which significantly increases the accuracy of prediction results. The average errors of the bending, tensile, and torsion stiffness between simulation results and predicted results converge to 1.88%, 2.33%, and 2.11%, respectively. The particle swarm optimization algorithm is employed for the structure parameter optimization based on the data-driven stiffness model. The stiffness errors of the optimized flexible backbone between simulation results and experimental results are 5.19%, 19.09%, and 5.38%, respectively. Experimental results show that the average position repeatability and orientation repeatability of a CDCR are 0.8822 mm and 0.0046 rad and the CDCR can carry the 500 g payload.

Feature selection is challenging in high-dimensional and small-sample data, particularly in industrial informatics with diverse noise sources. The information entropy of feature noise is included in mutual information of a label and noise-corrupted features, which can be removed to increase classification accuracy. In this article, we propose a robust feature selection method by eliminating feature noise in the relevance measure. Feature noise is modeled as a zero-mean censored normal distribution, so its entropy is determined by solving the variance equation based on the maximum entropy principle. Then, a noisy channel for feature transmission is proposed to extract class-relevant noise component. Furthermore, a noise-free mutual information metric is developed by removing noise entropy within mutual information. Eventually, a novel criterion is proposed by maximizing relevance based on noise-free mutual information while minimizing redundancy. Experimental results confirm the effectiveness of our approach on datasets from various industrial sectors.

The unit dual quaternion (UDQ)-based product-of-exponential (POE) formula has achieved efficient kinematic calibration for serial manipulators. However, due to the presence of unknown passive joint displacements, it is difficult to directly establish explicit forward kinematic models for parallel manipulators (PMs). This forms a barrier to subsequent error modeling and compensation. This work establishes a novel UDQ-based forward kinematic model for a PM by utilizing constraints on the identical pose of the moving platform across all its chains. Furthermore, the adjoint transformation of UDQ's twist is derived for PM's error modeling. Notably, an index matrix is introduced to achieve a unified representation of active or passive joint displacement. Thereby, this UDQ-based kinematic error modeling method is applicable to general PMs. In addition, an error compensation method is proposed for a PM using the UDQ-based local POE formula, which incorporates the developed forward kinematic model to adjust active joint displacements. The proposed method offers significant runtime savings compared to the traditional homogeneous-transformation-matrix-based POE formula due to the compact data structure and reduced arithmetic operations.

In a multiwire slicer, it is very crucial to keep wire sending smooth and tension constant to guarantee the quality of silicon wafers. Since the wire sending status has impact on the wire tension, it needs to be controlled to be smooth enough first. The measurement of the wire sending state is based on the two tensometers near the wire sending point and the tension adjusting unit, respectively. The measuring principle is given in detail. The smooth wire sending control is realized through a position adjusting unit. To achieve good control performance, the wire sending state is divided into several sections, and a novel feedforward multiconditioned P controller is designed. After the smooth wire sending controller is designed, the factors related to the wire tension vibration are analyzed deeply, and then, a self-tuning fuzzy controller is proposed. The parameters of the controller can be tuned according to the wire moving velocity. To show the validity of the proposed method, experiments on real platform are well conducted.