The manual lifting of heavy loads by personnel is susceptible to the development of muscle fatigue, which, in severe cases, can result in the irreversible impairment of muscle function. This study proposes a novel method of signal fusion to analyse muscle fatigue during manual lifting. Furthermore, this study represents the inaugural application of the back-propagation neural network and bidirectional encoder representation from the transformer (BP + BERT) algorithm to the fusion of two sensor inputs for the analysis of muscle fatigue. Lifting action fatigue tests were carried out on 16 testers in this study, with both surface electromyography (sEMG) and mechanomyography (MMG) signals collected as part of the process. The mean power frequency (MPF) eigenvalues were extracted separately for the two signals, and the results of muscle fatigue labelling according to the trend of the MPF eigenpeak were merged to produce three datasets. Subsequently, the three datasets were employed to categorise muscle fatigue classes using the support vector machine and radial basis function (SVM + RBF), support vector machine and bidirectional encoder representation from transformer (SVM + BERT), back-propagation neural network (BP), and back-propagation neural network and bidirectional encoder representation from transformer (BP + BERT) algorithms, respectively. The results of the muscle fatigue classification model demonstrated that the sEMG and MMG fused dataset, imported into the BP + BERT algorithm, exhibited the highest average accuracy of 98.10% for the muscle fatigue classification model. This study indicates that the fusion of sEMG and MMG signals is an effective approach, and the performance of the BP + BERT muscle fatigue classification model is also enhanced.
In the present study, we used a transformer model and a fusion of biosignals to estimate rotational acceleration in elbow and shoulder joints. To achieve our study objectives, we proposed a mechanomyography (MMG) signal isolation technique based on a variational mode decomposition (VMD) algorithm. Our results show that the VMD algorithm delivered excellent performance in MMG signal extraction compared to the commonly used technique of empirical mode decomposition (EMD). In addition, we found that transformer models delivered estimates of joint acceleration that were more precise than those produced by mainstream time series forecasting models. The average R2 values of transformer are 0.967, 0.968, and 0.935, respectively. Finally, we found that using a fusion of signals resulted in more precise estimation performance compared to using MMG signals alone. The differences between the average R2 values are 0.041, 0.053, and 0.043, respectively. Taken together, the VMD isolation method, the transformer algorithm and the signal fusion technique described in this paper can be seen as supplying a robust framework for estimating rotational acceleration in upper-limb joints. Further study is warranted to examine the effectiveness of this framework in other musculoskeletal contexts.
In order to better design handling-assisted exoskeletons, it is necessary to analyze the biomechanics of human hand movements. In this study, Anybody Modeling System (AMS) simulation was used to analyze the movement state of muscles during human handling. Combined with surface electromyography (sEMG) experiments, specific analysis and verification were carried out to obtain the position of muscles that the human body needs to assist during handling. In this study, the simulation and experiment were carried out for the manual handling process. A treatment group and an experimental group were set up. This study found that the vastus medialis muscle, vastus lateralis muscle, latissimus dorsi muscle, trapezius muscle, deltoid muscle and triceps brachii muscle require more energy in the process of handling, and it is reasonable and effective to combine sEMG signals with the simulation of the musculoskeletal model to analyze the muscle condition of human movement.
1 January 2025 in International Journal of Intelligent Systems
The two‐degree‐of‐freedom rotation stage serves as a crucial component in ground unmanned platforms, and its servo performance is pivotal to the platform’s overall functionality. To enhance the servo performance of the two‐degree‐of‐freedom rotation stage, we proposed a novel adaptive control approach: SSA–GA optimization of RBFNN integration into PID control. This method leverages the SSA–GA algorithm to optimize the parameters within the RBFNN, which is then seamlessly integrated into the PID control framework. This integration enables precise control of the two‐degree‐of‐freedom rotation stage, overcoming the limitations of traditional PID controllers. By simulating various real‐world situations such as step, noise, sinusoidal, transient excitation, impulse and modelling errors, it is demonstrated that the proposed control method achieves significant improvement in terms of control accuracy, fast response and robustness. It offers a more effective and reliable method for controlling the two‐degree‐of‐freedom rotation stage, addressing the challenges posed by various interfering factors. Meanwhile, through comprehensive comparisons with various optimization algorithms, we have proven that SSA–GA has the shortest optimization time. Consequently, the proposed method exhibits excellent application potential and broad prospects for future applications.
In recent years, advances in exoskeleton technology have increased the demand for human motion classification in terms of both movement diversity and recognition accuracy, making the effective classification of more complex asymmetric movements increasingly important. This study integrated surface electromyography (sEMG) and inertial measurement unit (IMU) signals and employed mutual information (MI) and muscle symmetry features (MSF) to analyze the characteristic differences between symmetrical muscles in both legs during asymmetric movements, with the aim of improving the accuracy of asymmetric motion classification. sEMG and IMU signals were collected from six movements, including three asymmetric postures: asymmetrical stance, single-knee kneeled position, and crouching advance. The acquired signals were processed through energy envelope analysis, active segment extraction, empirical mode decomposition (EMD), feature extraction, MI extraction, and MSF extraction. The relevance based on weight feature selection (RWFS) combined with conditional mutual information (CMI) method was then applied to reduce feature dimensionality, prioritize features with significant fluctuations, and preserve key characteristics. Finally, the CNN-LSTM-Attention algorithm was used for classification. Experimental results showed that fusing sEMG and IMU signals achieved 96.30% accuracy in lower limb motion recognition. The proposed method improves asymmetric movement classification and may provide a potential basis for exoskeleton motion classification in special environments.