Zibo Vocational Institute

Sealing defects in pharmaceutical plastic bags pose significant risks to drug safety, as micro-leakages may remain undetected until transportation, causing economic losses and hazards. Traditional manual inspection and existing automated methods suffer from low efficiency, poor sensitivity to subtle defects, and difficulties in addressing class imbalance due to scarce defective samples. To address these issues, this study proposes a comprehensive detection framework that integrates thermal imaging analysis, physics-guided data augmentation, and a novel Temporal Multi-Feature Fusion Network (TMFFNet). Thermal imaging reveals defective areas with distinct localized temperature elevations, providing a reliable basis for defect identification. A physics-guided augmentation method is developed to synthesize realistic defects: it models defect contours via hybrid polynomials, simulates thermal diffusion using dual-Gaussian operators, and fuses synthetic defects into normal samples under geometric constraints. This method effectively mitigates class imbalance, expanding the number of defective samples from 28 real ones to 2104 synthetic ones, with a total of 4385 samples in the dataset. The proposed TMFFNet, a dual-branch temporal network, processes three consecutive thermal frames to capture temporal dynamics. Its global-local fusion module enhances sensitivity to small defects, while a channel-aware SE-Dense module suppresses background noise, reducing false alarms. Experimental results show that TMFFNet outperforms traditional networks with a test set accuracy of 0.9809, and other evaluation metrics also demonstrate favorable performance. This framework provides an efficient, non-destructive solution for full pharmaceutical packaging inspection, improving drug safety and production efficiency.

To address the long development cycles, reliance on experience, and poor fabric compatibility of traditional textile patterns, this paper proposes a multimodal textile pattern generation and optimization algorithm (MPGO-Textile) and builds a complete AI-assisted design process. The MPGO-Textile algorithm integrates multimodal features from textual style descriptions, designer sketches, and fabric textures. It generates initial patterns using an improved Diffusion Model, introduces a periodic loss function to ensure seamless pattern splicing, and optimizes patterns using a fabric compatibility assessment model. The algorithm also supports real-time interactive adjustments by designers. To validate the algorithm's performance, a dataset of 3,000 textile patterns and 20 fabric textures was constructed. MPGO-Textile was compared with StyleGAN2 and Stable Diffusion on an NVIDIA RTX 4090 GPU. Results show that MPGO-Textile achieves a seamless stitching error (SSE) of 0.82, a 65.1% reduction compared to StyleGAN2 (2.35). Its peak signal-to-noise ratio (PSNR) on silk fabrics reaches 32.6dB, a 10.5% improvement over Stable Diffusion (29.5dB). The time required to generate a single 1024×1024 pixel pattern is 8.3s, a 33.6% reduction compared to StyleGAN2 (12.5s). In a subjective evaluation, 20 professional designers gave it an average Meaning of Score (MOSS) of 4.23, an improvement of over 13.7% over the comparison algorithm. An AI-assisted design process based on this algorithm reduced the pattern development cycle for a summer silk dress from 15 days to 3 days. A survey of 15 textile companies revealed that 86.7% of designers recognized the efficiency gains. Research demonstrates that the MPGO-Textile algorithm can effectively improve the quality and efficiency of textile pattern design, providing technical support for digital design in the textile industry.

Industrial furnace temperature control is critical to product quality, energy efficiency, and equipment longevity, yet remains challenging due to strong system nonlinearity and model uncertainty.Conventional Proportional-Integral-Derivative (PID) control and single offline optimization methods are therefore inadequate for achieving optimal performance over full operating regimes.To address these limitations, a thermal-aware genetic algorithm-optimized PID (GA-PID) control framework with dynamic strategy switching was proposed.A closed-loop architecture integrating thermal perception, mode decisionmaking, GA optimization, and knowledge accumulation was developed to enable precise, efficient, and robust temperature regulation in industrial furnaces.A nonlinear time-delay model was first established based on furnace heat transfer mechanisms to quantify key thermal parameters.A multi-source thermal perception module was then designed to extract related feature indicators.A fuzzy inference mechanism was then employed to achieve adaptive decision-making among three modes, with mode-specific GA strategies tailored to distinct thermal optimization objectives.Finally, an online self-evolving industrial furnace knowledge base was constructed to accumulate optimal GA-PID parameters and control experience under diverse thermal operating conditions.Simulation and industrial experiments demonstrated that the proposed dynamic GA-PID control strategy consistently outperformed conventional offline GA-PID, classical PID, and Particle Swarm Optimization (PSO)-PID methods across all operating modes.Specifically, setpoint tracking overshoot was reduced to 2.3%-2.8%with rise times of 48-55 s; steady-state temperature fluctuations were constrained within 0.18-0.20,achieving thermal efficiencies of 85.3%-86.7%;and disturbance recovery times were shortened to 9.5-11.5 s.The proposed framework provides a novel and systematic solution for high-precision, low-energy-consumption control of complex thermal systems and offers substantial theoretical significance and engineering application potential.

At present, animation generation and multi-modal interaction in virtual reality environments still face problems such as low generation quality, poor real-time performance and insufficient fusion between modes, which seriously restrict the authenticity and interaction efficiency of immersive experiences. A real-time synthesis and multi-modal interactive optimization method of generative adversarial network animation for a VR environment is proposed. In animation synthesis, a generation architecture with a spatiotemporal consistency adversarial training mechanism is constructed, and a multi-scale feature fusion strategy is combined to achieve high-quality and low-latency animation generation. Experiments were conducted on the VR (Virtual Reality)-Gesture-Voice dataset (60,000 training samples, 15,000 testing samples) and benchmarked against state-of-the-art (SOTA) models including VideoGAN and StyleGAN3. Key results: For the isolated GAN synthesis module: average rendering frame rate = 23.7 FPS (58% higher than VideoGAN), synthesis delay ≤ 4.5 ms; For the end-to-end VR system: average rendering frame rate = 85–92 FPS (meeting VR’s ≥72 FPS standard), end-to-end latency ≤17 ms (51% lower than StyleGAN3). In the real-time synthesis test of generative adversarial network (GAN) animation in the VR environment, two sets of key metrics are reported to clarify different system scopes: For the isolated GAN animation synthesis module (excluding end-to-end transmission and rendering), the improved algorithm achieved an average rendering frame rate of 23.7 FPS (58% higher than the traditional method) and controlled the synthesis delay within 4.5 Ms. Regarding system resource usage, GPU (Graphics Processing Unit） memory consumption is reduced by 0.6 GB, model reasoning time is reduced by 49.5%, and 85% real-time rendering efficiency can still be maintained at 8K resolution.

Medical environment comfort directly affects patient treatment outcomes and recovery processes. This study constructs a machine learning prediction model for patient excessive discomfort based on environmental monitoring data from medical infusion rooms. The research collected 1,000 samples with 11 environmental feature data, including temperature, humidity, noise level, air quality index, wind speed, lighting intensity, oxygen concentration, carbon dioxide concentration, air pressure, air circulation speed, and air pollutant concentration. Through comparative analysis of 10 machine learning algorithms, XGBoost model demonstrated the best performance with accuracy of 85.2%, precision of 86.5%, recall of 92.3%, F1-score of 0.893, and ROC-AUC of 0.889. Using SHAP and LIME interpretability methods, analysis revealed that air quality index (importance score 1.117) and temperature (importance score 1.065) are the most critical factors affecting patient comfort, followed by noise level (0.676) and humidity (0.454). SHAP partial dependence analysis revealed specific impact patterns of environmental factors: humidity shows positive correlation with discomfort, noise level exhibits strong linear positive correlation, temperature demonstrates nonlinear relationships, and air quality deterioration significantly increases patient discomfort. LIME local explanations validated the consistency of analysis results, providing scientific basis for personalized environmental control. The research results indicate that machine learning methods based on multi-sensor environmental monitoring can effectively predict patient discomfort. Interpretability analysis reveals the influence mechanisms of environmental factors, providing important support for intelligent management of medical environments and formulation of scientific control strategies.