Evaluating cognitive load in mixed reality (MR) has become a significant challenge in human-computer interaction (HCI). To address this, we established an MR multimodal experimental platform with three distinct environments to induce varying levels of cognitive load. Participants engaged in MR-based CNC machine tool interaction tasks within these environments. Using the built-in sensors of the HoloLens 2 mixed reality head-mounted display (MR-HMD) and wearable heart rate sensors, we collected device and physiological data from participants wearing the MR-HMD while performing these tasks. The cognitive load of participants was assessed by using the NASA-TLX questionnaire. Experimental results indicated that the operation time required in the MR environment increased by 49% under high cognitive load compared to low-load conditions. High-load environments also led to increased anxiety, frustration, and decreased performance among participants. Through comparative experiments, we identified suitable sensor data streams and algorithms for cognitive load classification and designed an MR digital twin factory cognitive load warning prototype system. This system utilizes an improved Transformer-CL algorithm, achieving a cognitive load classification accuracy of 95.83%. The system provides high cognitive load warnings, reducing the risks associated with high cognitive load tasks in MR work environments.

Bio-inspired design (BID) is an abstract process, if we can visualize the process of fusing abstract biological inspiration with figurative product shapes, and combine it with artificial intelligence technology to express the designer’s creativity, it will greatly improve the efficiency and accuracy of product shape bionic design. To address this problem, we combine BID with deep generative (DG) model to build a co-creative deep generative bio-inspired design (DGBID) model. Firstly, the designers used perceptual engineering and eye-movement experiments to select the bionic creature that best fits the bionic product and the suitable bionic product and bionic image, respectively. Then, the images are embedded into the potential space of StyleGAN, and the potential relationship between the two is visualized using StyleGAN’s image morphing technique, which generates a new bionic fusion scheme. Finally, the contour lines of the solution are extracted as a reference, the designer is involved in the optimization of the scheme as a sketch, and the hand-drawn sketch is transformed into a real product solution using style migration techniques. The entire bionic design experiment process is a co-creative approach with artificial intelligence technology as the lead and designer participation. The feasibility of the method is verified using the side view of a car as a bionic product. The results show that the integration of bionic technology with deep generative model technology can accelerate the innovation and development of bionic products and provide designers with design references and rapid-generation tools.

To create alternative complex patterns, a novel design method is introduced in this study based on the error back propagation (BP) neural network user cognitive surrogate model of an interactive genetic algorithm with individual fuzzy interval fitness (IGA-BPFIF). First, the quantitative rules of aesthetic evaluation and the user’s hesitation are used to construct the Gaussian blur tool to form the individual’s fuzzy interval fitness. Then, the user’s cognitive surrogate model based on the BP neural network is constructed, and a new fitness estimation strategy is presented. By measuring the mean squared error, the surrogate model is well managed during the evolution of the population. According to the users’ demands and preferences, the features are extracted for the interactive evolutionary computation. The experiments show that IGA-BPFIF can effectively design innovative patterns matching users’ preferences and can contribute to the heritage of traditional national patterns.

Manual material handling (MMH) is commonly demanded in the manufacturing industry. Occupational muscle fatigue of the arm, shoulder, and back, which arise from MMH tasks, can cause work absences and low efficiency. The available literature presents the lack of the fatigue comparison between targeted muscles, on the same part or on different parts. The main aim of the present study was to evaluate and compare the fatigue of upper-body muscles during repetitive bending tasks, an experiment involving 12 male subjects has been conducted to simulate material handling during furniture board drilling. The vertical lifting distance was chosen to be the single independent variable, and the three levels were 0, 250, and 500 mm. Surface electromyography (sEMG) was used to measure the muscle fatigue of the biceps brachii, upper trapezius, and multifidus, while the sEMG parameters, including the normalized electromyographic amplitude (Normalized EA) and mean power frequency (MPF), of the target muscles were analyzed. The experimental results reveal that during the manual handling tasks, the biceps brachii was the most relaxed muscle, contributing the least muscle tension, while the multifidus was the most easily fatigued muscle. Furthermore, the EMG MPF fatigue threshold (MPFFT) of multifidus muscle tension was tested to estimate its maximum workload in the long-term muscle contraction. In conclusion, bending angle should be maintained to a small range or bending should even be avoided during material-handling tasks.

To effectively organize design elements in virtual reality (VR) scene design and provide evaluation methods for the design process, we built a user image space cognitive model. This involved perceptual engineering methods and optimization of the VR interface. First, we studied the coupling of user cognition and design features in the VR system via the Kansei Engineering (KE) method. The quantitative theory I and KE model regression analysis were used to analyze the design elements of the VR system’s human–computer interaction interface. Combined with the complex network method, we summarized the relationship between design features and analyzed the important design features that affect users’ perceptual imagery. Then, based on the characteristics of machine learning, we used a convolutional neural network (CNN) to predict and analyze the user’s perceptual imagery in the VR system, to provide assistance for the design optimization of the VR system design. Finally, we verified the validity and feasibility of the solution by combining it with the human–machine interface design of the VR system. We conducted a feasibility analysis of the KE model, in which the similarity between the multivariate regression analysis of the VR intention space and the experimental test was approximately 97% and the error was very small; thus, the VR intention space model was well correlated. The Mean Square Error (MSE) of the convolutional neural network (CNN) prediction model was calculated with a measured value of 0.0074, and the MSE value was less than 0.01. The results show that this method can improve the effectiveness and feasibility of the design scheme. Designers use important design feature elements to assist in VR system optimization design and use CNN machine learning methods to predict user image values in VR systems and improve the design efficiency. Facing the same design task requirements in VR system interfaces, the traditional design scheme was compared with the scheme optimized by this method. The results showed that the design scheme optimized by this method better fits the user’s perceptual imagery index, and thus the user’s task operation experience was better.