The gear fault signal under different working conditions is non-linear and non-stationary, which makes it difficult to distinguish faulty signals from normal signals. Currently, gear fault diagnosis under different working conditions is mainly based on vibration signals. However, vibration signal acquisition is limited by its requirement for contact measurement, while vibration signal analysis methods relies heavily on diagnostic expertise and prior knowledge of signal processing technology. To solve this problem, a novel acoustic-based diagnosis (ABD) method for gear fault diagnosis under different working conditions based on a multi-scale convolutional learning structure and attention mechanism is proposed in this paper. The multi-scale convolutional learning structure was designed to automatically mine multiple scale features using different filter banks from raw acoustic signals. Subsequently, the novel attention mechanism, which was based on a multi-scale convolutional learning structure, was established to adaptively allow the multi-scale network to focus on relevant fault pattern information under different working conditions. Finally, a stacked convolutional neural network (CNN) model was proposed to detect the fault mode of gears. The experimental results show that our method achieved much better performance in acoustic based gear fault diagnosis under different working conditions compared with a standard CNN model (without an attention mechanism), an end-to-end CNN model based on time and frequency domain signals, and other traditional fault diagnosis methods involving feature engineering.

It is very important to predict the reaction of customers to future products. Sample pictures and line sketches are commonly used for discovering consumers’ demands. However, these two agents are difficult to express the real product form clearly and comprehensively, which may affect the accuracy of consumer affection prediction. In this paper, a method based on Kansei engineering via virtual reality (VR) has been proposed to construct a mapping relationship between form elements and consumer affection from four dimensions of the overall, unit, interrelation, and detail (OUID), so as to provide references for product form design. The correlation between product form elements and the perceptual image was studied with multiple linear regression analysis and partial correlation analysis. The different effects of VR, sketches, and pictures on user affection prediction were compared in this paper as well. A hairdryer has been employed as an example to illustrate the proposed method. The results show that compared with sketch and picture, VR based-prediction model is more accurate and reliable in predicting users’ emotional preferences. In addition to the unit form, the overall features, interrelation, and details are also the key factors affecting consumer preference. Highlight

Thermography has proven to be one of the most effective approaches to detect cracks in conductive specimens over a relatively large area. Pulsed eddy current stimulated thermography is an emerging integrative nondestructive approach for the detection and characterization of surface and subsurface cracks. In this paper, heating behaviors of edge cracks, excited by pulsed eddy currents, are examined using numerical simulations. The simulations are performed using COMSOL multiphysics finite element method simulation software using the AC/DC module. The simulation results show that in the early heating stage, the temperature increases more quickly at the crack tip compared with other points on the sample. The results indicate that to maximize sensitivity, the response should be analyzed in the early stages of the heating period, no more than 100 ms for samples in which we are interested. The eddy current density distribution is changed with a variation in inductor orientation, but the crack tips remain the “hottest” points during the excitation period, which can be used for robust quantitative defect evaluation. Signal feature selection, transient temperature profile of the sample, and influence of the inductor orientation on the detection sensitivity for edge cracks are investigated. The work shows that positioning of the inductor, perpendicular to the crack line, results in the highest sensitivity for defect detection and characterization. The crack orientation can be estimated through the rotation of the linear inductor near the sample edge and the crack tips.

Accurate lithological recognition and quantitative determination of multiple chemical elements have a wide market application prospect in geological and geochemical exploration. In recent years, with the development of machine learning, laser induced breakdown spectroscopy (LIBS) coupled with various chemometric methods has demonstrated great advantages to solve both classification and quantitative problems in rock analysis. However, these methods are usually applied for single-task analysis. To further increase the analysis efficiency, LIBS combined with a deep convolutional neural network with a two-dimensional (2D-CNN) algorithm is proposed here for simultaneous determination of lithology and seven major chemical elements in rock samples from six different types. The structure of our CNN model was designed with two different outputs, which could complete classification and regression tasks at the same time. Based on the prediction results of the test set, the experimental results show that the classification accuracy of the CNN model could reach 99.26% the coefficient of determination (R²) of target elements could reach a good level (R_Si ² = 0.9917, R_Al ² = 0.9935, R_Ca ² = 0.9906, R_Mg ² = 0.9926, R_Fe ² = 0.9716, R_Na ² = 0.9693, R_K ² = 0.8781), and both the mean absolute error (MAE) of each test sample and prediction relative error of each element are improved. To verify the superior performance of this CNN model, the CNN algorithm was compared with other models, including k-nearest neighbor (kNN), support vector machine (SVM), partial least squares discrimination analysis (PLS-DA), back propagation artificial neural network (BP-ANN) and partial least squares regression (PLSR). The results indicate that the CNN has great potential in the field of rock quantitative and qualitative analyses and provides a reliable data processing method for LIBS to complete multi-task analysis with high efficiency and good accuracy.