ISA transactions最新文献

With the deep digital transformation of traditional manufacturing industry and the continuous automation level improvement of production lines, it is more important to predict the Key Performance Indicators (KPIs) of processes in a timely and accurate manner. The traditional laboratory destructive test method for obtaining KPIs consumes a large amount of time and incurs high costs, which not only fails to provide timely and effective guidance for production processes but also results in significant losses for manufacturing enterprises. To address these issues, an online prediction soft sensor model for KPIs based on a serial-parallel gated recurrent unit with self-attention mechanism (SPGRU-SA) soft sensor model is proposed. This model achieves accurate online prediction of KPIs by considering both the dynamic features of multi-unit processes and the static features of process setups. First, a serial-parallel gated recurrent unit model is designed to extract multi-unit dynamic features. Second, based on the self-attention mechanism, the attention weights of static features and dynamic features are calculated, which can reflect the correlation of the performance indicators. Then, the fully connected layers output the result. Finally, the comparative experimental results based on the hot rolling strip mill process and the Tennessee Eastman process show that SPGRU-SA can accurately predict the KPIs of complex multi-unit industrial processes.

High-power induction furnace (IF) is a highly complex thermoelectric system with strong nonlinear time-varying characteristics. The lack of direct online measurement methods complicates status awareness, leading to apparent "black-box" behavior and sensing difficulties. We propose a transferable layered physics-informed learning-based modeling approach to address the above challenges. At the underlying level, a series of linear physical models are established at multiple operating points to guide the data-driven models, thus comprehensively capturing the data characteristics and electrical dynamics of the IF. At the upper level, physical prior knowledge-based global nonlinear constraints are introduced to ensure the model accuracy and physical consistency. Each underlying model can be regarded as a single task, and the modeling problem is skillfully transformed into a multitask learning optimization with global nonlinear constraints. In addition, a transferable model training strategy with an architecture of cascaded shared layers and task layers is developed to facilitate knowledge transfer between adjacent melting batches and thereby optimize the training process. The feasibility and effectiveness of the scheme are validated by experiments using actual sampling data.

Soft sensors of quality indicators for industrial production processes compensate for the shortcomings of traditional measurement methods, which are essential for improving product quality. However, the complex mechanisms and time-varying delays in multi-unit, long-flow industrial processes pose significant challenges for multi-indicator soft sensing. Existing research on the fusion of mechanical and data-driven models is not sufficiently advanced. To address these challenges, a mechanism and data fusion driven multi-indicator soft sensor framework for industrial processes is proposed, using the hot strip rolling process (HSRP) as a case study. First, the mechanism of HSRP is analyzed and the unknown parameters in the mechanism model are identified. Second, the data derived from the mechanistic model are fused with the process data. Then Kolmogorov-Arnold Networks with an embedded time-series input layer (TS-KAN) are developed to address the challenge of time-varying delays caused by long production processes and production fluctuations. Finally, the proposed framework is validated using actual HSRP production data, achieving simultaneous accurate prediction of strip flatness and crown.

Data-driven soft sensing methods are widely used for product quality prediction in large-scale industrial processes. Traditional approaches face significant challenges, such as limited representation of multivariate couplings, difficulties in modeling nonlinear interactions, and slow adaptation to dynamic changes in complex industrial settings. To address these, we propose a manufacturing quality prediction model integrating Dual-layer Graph Supervised Embedding with Multi-Granularity Attention Enhancement Mechanisms (DGS-MA). The model constructs a dual-layer complementary graph structure: the first layer creates an original parameter relationship graph based on feature similarity to capture local static associations, while the second layer uses a label-aware supervised Node2vec algorithm to generate embedding vectors, reconstructing a global quality-driven topology. This forms a dual-view representation of 'original features - embedding vectors'. A multi-granular graph attention enhancement mechanism is introduced, which employs a dual-pathway attention network to aggregate neighborhood information from both original features and supervised embeddings. A cross-layer attention mechanism adaptively fuses the importance of these two feature types, enabling coordinated optimization of local details and global patterns. Additionally, an explicit supervision constraint is incorporated to enhance prediction accuracy and interpretability, embedding supervision signals in both the graph's embedding space and attention mechanism. The method is validated with a real-world production dataset, showing significant improvements in quality prediction under complex conditions.

Accurate temperature forecasting relies on traditional meteorological parameters that are essential for monitoring weather informatics and guiding forecasting efforts. This study introduces a deep learning architecture for high-precision climate temperature forecasting via an improved Kolmogorov-Arnold Networks, named Tem²-KAN. Grounded in the Kolmogorov-Arnold representation theorem, Tem²-KAN explores replacing conventional linear weights in neural networks with spline-parameterized univariate functions, enabling dynamic learning of nonlinear climate patterns while maintaining intrinsic interpretability. The proposed framework uniquely integrates the universal approximation capabilities of Multi-Layer Perceptrons (MLPs) with physically meaningful feature visualization through its adaptive activation functions, addressing critical limitations of black-box climate models. A temperature prediction pipeline is established that first preprocesses raw meteorological data from UK monitoring stations, then trains Tem²-KAN to map historical trends to multi-horizon forecasts. Rigorous evaluations on real-world climate datasets demonstrate Tem²-KAN's dual advantage achieving state-of-the-art prediction accuracy while utilizing fewer trainable parameters. In addition, a systematic ablation study quantifies the sensitivity of key Tem²-KAN-specific hyperparameters (spline order k, grid resolution grid) on forecasting performance. Finally, we theoretically prove Tem²-KAN's universal approximation capacity through function space analysis, and practically, we demonstrate its interpretability and prediction performance. These innovations position Tem²-KAN as a paradigm-shifting tool for climate informatics, offering meteorologists both high predictive performance and mechanistic insight into temperature dynamics. The framework's reduced hyperparameter complexity further enhances its viability for operational forecasting systems.

Data-driven industrial soft sensor modeling techniques have been widely applied in predicting key variables in complex industrial processes. However, with industrial processes becoming increasingly intricate, the data they produce exhibit characteristics such as strong temporal dependencies, high dimensionality, and local structures, posing significant challenges for soft sensing. To address these issues, this paper proposes novel Semi-Supervised Sparse Stacked Autoencoder integrated with the Local Linear Embedding algorithm (SS-SAE-LLE). Unlike traditional autoencoders (AE), which capture hierarchical data features by minimizing global fitting error, SS-SAE-LLE simultaneously accounts for the spatio-temporal characteristics of the data through the local linear embedding algorithm. Moreover, it incorporates supervised tuning by leveraging labeled data and training with a semi-supervised learning framework, further improving prediction accuracy. To evaluate the feasibility of the proposed method, experiments are conducted on PTA solvent and SRU system datasets. The simulation results demonstrate that SS-SAE-LLE achieves higher prediction accuracy than other models, highlighting its applicability in the field of industrial soft sensor modeling.

Aiming to address the problem of robot path planning in environments containing narrow passages, this paper proposes a novel global path planning method: the DSR (Dual-source Light Continuous Reflection Exploration) algorithm. This algorithm, inspired by the natural reflection of light, employs the concept of continuous reflection for path planning. It can efficiently generate an asymptotically optimal path on the map containing narrow passages. The DSR algorithm has been evaluated on different maps with narrow passages and compared with other algorithms. In comparison with the bidirectional Rapidly-exploring Random Tree algorithm, the DSR algorithm achieves a significant reduction in both path length (by 27.08% and 34.35%) and time consumption (by 98.47% and 91.03%). Numerical simulations and experimental analysis have demonstrated the excellent performance of the DSR algorithm.

As the penetration of renewable energy increases to a large scale and power electronic devices become widespread, power systems are becoming prone to synchronous oscillations (SO). This event has a major impact on the stability of the power grid. The recent research has been mainly concentrated on identifying the parameters of sub-synchronous oscillation. Sub/Super synchronous oscillations (Sub/Sup-SO) simultaneously occur, increasing the difficulty in accurately identify the parameters of SO. This work presents a novel method for parameter identification that effectively handles the Sub/Sup-SO components by utilizing the Rife-Vincent window and discrete Fourier transform (DFT) simultaneously. To mitigate the impact of spectral leakage and the fence effect of DFT, we integrate the tri-spectral interpolation algorithm with the Rife-Vincent window. We use the instantaneous data of the phasor measurement unit (PMU) to identify Sub/Sup-SO-related parameters (Sub/Sup-SO damping ratio, frequency, amplitude and phase). First, the spectrum of the Sub/Sup-SO signals is analyzed after incorporating the Rife-Vincent window, and the characteristics of the Sub/Sup-SO signal are determined. Then, the signal spectrum is identified using a three-point interpolation algorithm, and the damping ratio, amplitude, frequency, and phase of the Sub/Sup-SO signals are obtained. In addition, we consider the identification accuracy of the algorithm under various complex conditions, such as the effect of Sub/Sup-SO parameter variations on parameter identification in the presence of a non-nominal frequency and noise. The proposed algorithm accurately identifies the parameters of multiple Sub/Sup-SO components and two Sub-SO components that are in close proximity. Testing with synthetic and real data demonstrates that the proposed algorithm outperforms existing methods in terms of identification accuracy, identification bandwidth, and adaptability.

This study investigates pigeon-like flexible flapping wings, which are known for their low energy consumption, high flexibility, and lightweight design. However, such flexible flapping wing systems are prone to deformation and vibration during flight, leading to performance degradation. It is thus necessary to design a control method to effectively manage the vibration of flexible wings. This paper proposes an improved rigid finite element method (IRFE) to develop a dynamic visualization model of flexible flapping wings. Subsequently, an adaptive vibration controller was designed based on non-singular terminal sliding mode (NTSM) control and fuzzy neural network (FNN) in order to effectively solve the problems of system uncertainty and actuator failure. With the proposed control, stability of the closed loop system is achieved in the context of Lyapunov's stability theory. At last, a joint simulation using MapleSim and MATLAB/Simulink was conducted to verify the effectiveness and robustness of the proposed controller in terms of trajectory tracking and vibration suppression.The obtained results have demonstrated great practical value of the proposed method in both military (low-altitude reconnaissance, urban operations, and accurate delivery, etc.) and civil (field research, monitoring, and relief for disasters, etc.) applications.

A multi-sensor information fusion algorithm based on fault-tolerant Kalman filter is proposed in phase-sensitive optical time-domain reflectometer (Φ-OTDR) system, for achieving fading-free distributed vibration sensing. Firstly, a fault-tolerant dual-core complementary array model is designed. The Rayleigh scattering signal denoising, and vibration existence judgment of localization points are carried out to obtain the differentiated frequency demodulation results of the sensing points of the dual-core fiber array. Then a fault-tolerant control strategy is used to determine the sensor weight coefficients and vibration judgment coefficients during data fusion processing, and array data fusion is carried out based on time series data using Kalman filter to realize error value identification and filling. The advantage of this method is the combination of redundant data in a complementary way to improve the system stability. The frequency response ranges from 10 Hz to 2400 Hz and the localization accuracy is 98.33%. The influence of key parameters on the frequency demodulation performance of fault-tolerant Kalman filter is discussed, and a standard deviation of 14.6 Hz and an average error of 7.6 Hz are obtained. The demodulation frequency data matrix obtained by the classical demodulation method has a demodulation error probability of 89.18%, which proves the widespread existence of demodulation errors in vibration signals. The fusion error of demodulation frequency is reduced to 0.25 Hz, the frequency demodulation accuracy reaches 100%, and the demodulation error caused by interference attenuation can be completely eliminated. This system based on fault-tolerant Kalman filter has the characteristics of simple multiplexing structure, interference fading resistance and stable demodulation performance.