太原城市职业技术学院

Cast aluminum alloys are widely used in aerospace，automotive and other industries due to their excellent mechanical properties. However，traditional alloy design faces challenges such as vast composition space，high costs of trial-and-error experiments and difficulty in predicting the nonlinear relationship between composition and properties. This paper proposes a machine learning model that combines backpropagation neural networks，principal component analysis，and genetic algorithms for multi-objective property prediction of cast aluminum alloys. The model establishes the relationship between alloy composition and properties through the nonlinear mapping of backpropagation neural networks，reduces dimensionality via principal component analysis，and optimizes network parameters using genetic algorithms-thereby improving prediction accuracy and training efficiency. The results show that the optimized model has mean squared error of 36.28，correlation coefficient of 0.91，and mean absolute error of 2.44. In the experimental verification of ultimate strength，yield strength，and elongation after fracture，the error between experimental values and predicted values is controlled within the range of ±5%. This high prediction accuracy demonstrates the efficiency and reliability of the proposed model.

Background: Light volleyball (LVB) is a popular recreational sport with increasing participation. Hamstring strain (HS) is common among LVB players and leads to significant time loss and reinjury risk. Clinical flexibility tests such as passive knee extension (KE), straight leg raise (SLR), and forward flexion distance (FFD) are widely used for HS screening, but prospective evidence of their predictive value is limited and conflicting. Shear-wave elastography (SWE) can quantify muscle mechanical properties, yet its role in predicting future HS remains unclear. This study aimed to compare the predictive value of a dynamic SWE parameter—the change in shear-wave velocity (ΔSWV) of the distal biceps femoris long head (BFlh) from relaxed to maximally stretched position—against KE, SLR, and FFD for 3‑month HS risk in LVB enthusiasts. Methods: In this prospective cohort study, 75 healthy LVB players (42 males, 33 females, age 25.7±6.2 years) underwent baseline measurements of SWV at distal/proximal BFlh and semitendinosus in relaxed and stretched positions (ΔSWV calculated), along with KE, SLR, and FFD. Participants were followed for 3 months to record HS occurrence. Predictive performance was assessed using receiver operating characteristic (ROC) curves, and comparisons between tests were made with the DeLong test. Results: Nine participants (12.0%) sustained HS. Only distal BFlh ΔSWV was significantly higher in the HS group (median 8.17 vs. 4.40 m/s, P<0.001). Its area under the ROC curve (AUC) was 0.985 (95% CI: 0.954–1.000), with an optimal cutoff of 7.37 m/s, sensitivity 0.889, and specificity 1.000. AUCs for KE, SLR, and FFD were 0.427, 0.536, and 0.603, respectively; the combined AUC was 0.641, all significantly lower than ΔSWV (P<0.001). Firth regression yielded an odds ratio of 28.14 per m/s (P<0.001). Conclusions: Distal BFlh ΔSWV accurately predicts 3‑month HS risk in LVB players, far outperforming KE, SLR, and FFD alone or combined. These findings suggest that dynamic SWE measurement could serve as a valuable screening tool for HS risk, although further validation in larger populations is warranted. Clinical trial number: Not applicable.

Deep excavation support structures exhibit nonlinear and spatiotemporally coupled deformation under complex geological conditions and construction disturbances, making single-sensor monitoring insufficient for real-time safety warning. This study proposes a multi-source sensing and real-time data fusion method for safety early warning of deep foundation pit support structures. A heterogeneous sensor network composed of strain gauges, inclinometers, axial force gauges, and hydrostatic levels is deployed to monitor pile bending moment, horizontal displacement, support axial force, and surface settlement. Sensor streams with different sampling rates are temporally aligned by cubic spline interpolation and spatially mapped onto a unified support-pile profile. Outliers are removed using the Pauta criterion, and adaptive inverse-variance weighting is combined with Kalman filtering to obtain robust fused feature sequences. An improved grey relational analysis model with exponential time weighting and dynamic thresholds is then used to identify typical failure modes and issue three-level warnings. Field experiments using 30 days of monitoring data show that the method achieves 98.3% effective data retention, 20.3 dB signal-to-noise ratio, RMSE of 0.34 mm, warning response time of 4.2 s, and false alarm rate of 8.6%. The method improves structural sensing reliability and real-time safety warning capability.

Rectangular concrete-filled steel tube (RCFST) columns are widely adopted as primary load-bearing components in engineering structures, and making reliable estimation of their axial compressive capacity crucial to structural design and safety assessment. However, existing theoretical and design equations generally rely on a series of simplifying assumptions and empirical modification coefficients, which limit their prediction accuracy and applicability. In this context, a data-driven strategy is proposed in which capacity prediction and model interpretation are integrated within a unified procedure. The proposed approach establishes the nonlinear mapping between high-dimensional input features, including geometric and material properties, and axial compressive capacity, while employing the SHapley Additive exPlanations (SHAP) method to quantify the contributions of critical input parameters to the model predictions. An experimental database comprising 745 axial compression tests on RCFST columns is established through an extensive literature survey. Prior to training, correlations within the candidate inputs are examined using Pearson coefficients. An SMA-LSSVM–ANN hybrid prediction model is then developed, in which SMA iteratively searches for the optimal combination of the LSSVM-ANN hyperparameters using mean squared error as the fitness function. Comparisons with several benchmark machine learning (ML) models and existing design-code formulations validate the predictive performance of the proposed model, which achieves an R2 of 0.9571 and an RMSE of 302.84 kN on the testing dataset. Furthermore, SHAP analysis is employed to interpret the proposed model from global and feature-dependency perspectives. Global SHAP analysis identifies concrete compressive strength and cross-sectional dimensions as the dominant features, while feature-dependence analysis reveals generally positive effects of cross-sectional dimensions and material strengths and a negative effect of column length on the predicted capacity.

Multi-factor information multimedia big data mining analysis is a thorough method for generating valuable insights from large, diversified datasets containing multimedia material. Exploration and interpretation of multi-dimensional data, including text, photos, videos, and audio, are part of this sophisticated analytical process. Conventional multi-dimensional information multimedia big data mining analysis suffers from increased processing complexity and has trouble managing high-dimensional data. With parallel processing and a novel pruning strategy, our proposed modified Apriori algorithm effectively addresses these issues, greatly decreasing computational overhead and enhancing scalability for high-dimensional datasets. The data was collected using modern light and electron microscopy techniques. To enhance both the signal's quality and the network's general efficiency, a Wiener filter was used to pre-process the acquired data for noise reduction. Principal component analysis was used to extract pre-processed data (PCA). Multimedia big data mining uses PCA to minimize high-dimensional data while maintaining important information and reducing redundancy. This allows for more effective analysis and feature extraction for improved insights and resource optimization. The proposed approach was tested in a simulated environment, yielding the following performance metrics: accuracy performance (89