A three-degree-of-freedom (3-DOF) piezoelectric ultrasonic transducer is a critical component in elliptical and longitudinal ultrasonic vibration-assisted cutting processes, with its geometric structure directly influencing its performance. This paper proposes a structural optimization method based on a convolutional neural network (CNN) and non-dominated sorting genetic algorithm II (NSGA2). This method establishes a transducer lumped model to obtain the electromechanical coupling coefficients (X-$k_e$ and Z-$k_e$) and thermal power (X-P) indicators, evaluating the bending and longitudinal vibration performance of the transducer. By creating a finite element model of the transducer with mechanical losses, a dataset of different transducer performance parameters, including the tail mass, piezoelectric stack, and dimensions of the horn, is obtained. Training a CNN model with this dataset yields objective functions for the relationship between different transducer geometric structures and performance parameters. The NSGA2 algorithm solves the X-$k_e$ and Z-$k_e$ objective functions, obtaining the Pareto set of the transducer geometric dimensions and determining the optimal transducer geometry in conjunction with X-P. This method achieves simultaneous improvements in X-$k_e$ and Z-$k_e$ of the transducer by 22.33% and 25.89% post-optimization and reduces X-P to 18.97 W. Furthermore, the finite element simulation experiments of the transducer validate the effectiveness of this method.

Quartz flexible accelerometers (QFAs) are a type of temperature-sensitive sensor, whereby a change in temperature will cause the key parameters of the accelerometer to drift and cause stability errors. Due to the absence of effective methods for sensing the temperature of internal accelerometer components, existing temperature error correction approaches primarily rely on shell temperature measurements to establish correction models. Consequently, most correction methods achieve higher accuracy during the steady-state heat conduction phase of the accelerometer, whereas the correction error markedly increases during the transient heat conduction phase. To elucidate the temperature discrepancy between the QFA shell and its internal components and to support the development of a temperature error correction method for QFAs based on the internal temperature as a reference, this paper investigated the heat exchange dynamics between the interior and exterior of a QFA. A thermal conduction simulation model of the QFA was established, from which the spatiotemporal distribution patterns of the internal temperature field were derived. The results indicate that the temperature of the QFA shell changes significantly faster than that of the internal meter head in the early stage of the temperature change. The temperature gradient between the shell and the meter head first increases and then decreases, and the rate of temperature change in the upper part of the accelerometer is faster than that in the lower part. Before thermal equilibrium is reached, the temperature distribution inside the accelerometer is uneven in terms of time and space. Inside the accelerometer, the yoke iron, swing plate assembly, servo circuit, and magnetic steel assembly are the main components that exhibit differences in the internal temperature change in the QFA. When developing the temperature error correction method, it was crucial to address and mitigate the impact of temperature variations among these components. The average RMSE between the predicted temperature from the heat transfer model established in this paper and the experimental results was 0.4 °C. This indicates that the model can accurately predict the temperature variation within the QFA, thereby providing robust support for investigating the temperature behavior inside the QFA and offering essential technical foundations for enhancing the accuracy of the temperature error correction method.

Aiming at the problems that the semantic representation of information extracted by the shallow layer of the current remote sensing image scene classification network is insufficient, and that the utilization rate of primary visual features decreases with the deepening of the network layers, this paper designs a multi-scale reverse master–slave encoder network (RMSENet). It proposes a reverse cross-scale supplementation strategy for the slave encoder and a reverse cross-scale fusion strategy for the master encoder. This not only reversely supplements the high-level semantic information extracted by the slave encoder to the shallow layer of the master encoder network in a cross-scale manner but also realizes the cross-scale fusion of features at all stages of the master encoder. A multi-frequency coordinate channel attention mechanism is proposed, which captures the inter-channel interactions of input feature maps while embedding spatial position information and rich frequency information. A multi-scale wavelet self-attention mechanism is proposed, which completes lossless downsampling of input feature maps before self-attention operations. Experiments on open-source datasets RSSCN7, SIRI-WHU, and AID show that the classification accuracies of RMSENet reach 97.41%, 97.61%, and 95.9%, respectively. Compared with current mainstream deep learning models, RMSENet has lower network complexity and excellent classification accuracy.