Deep learning's potential for intelligent fault diagnosis (IFD) is constrained by inadequate model generalization to unseen working conditions, which motivates research in domain generalization-based fault diagnosis (DGFD). However, most DGFD methods rely on multiple labeled source domains during the training phase, conflicting with the prevalent scarcity of labeled industrial data. To bridge this gap, this paper introduces a novel semi-supervised domain generalization-based fault diagnosis (SemiDGFD) method using adaptive pseudo-label selection and distributionally robust optimization (Ada-DRO). The Ada-DRO only requires one labeled source domain and multiple unlabeled source domains. Furthermore, this method combines the primary branch model with auxiliary branch models. Firstly, it employs multiple auxiliary branch models to assign pseudo-labels to unlabeled data through distribution alignment. Subsequently, an adaptive threshold selects high-confidence pseudo-labels to mitigate noise. Finally, an uncertainty set is constructed using mask-augmented labeled source data and selected pseudo-labeled data. Wasserstein distance constrains the set scope, and distributionally robust optimization (DRO) is performed over this set to enhance the primary branch model's cross-domain generalization performance. Extensive experiments demonstrate the proposed method's superior accuracy over state-of-the-art SemiDGFD methods.

Accurate acquisition of dispersion characteristics is essential for ultrasonic guided wave (GW) -based structural health monitoring. However, conventional approaches often rely on dense spatial sampling and are difficult to implement in practical engineering scenarios involving complex geometries and limited sensing regions. To address this challenge, this study proposes an ultrasonic GW dispersion analysis framework that integrates pencil lead break (PLB) actuation scanning with compressed sensing (CS). A dedicated PLB excitation device is developed to transform conventional passive PLB signals into a controllable active GW source, enabling repeatable scanning measurements. To ensure timing consistency, a spatially uniform sampling-based zero-point drift correction method is introduced. Furthermore, a CS-based sparse reconstruction framework is employed to recover the frequency-wavenumber spectrum from under-sampled data, thereby significantly relaxing the spatial sampling requirement. Experimental results on aluminum plates and composite laminates demonstrate that the proposed method achieves high-fidelity dispersion reconstruction with an average error below 4.2% while reducing the required sampling points by approximately 50%. The method is further validated on carbon fiber-reinforced aluminum laminates and filament-wound hydrogen storage cylinders, confirming its effectiveness for complex composite and curved structures. These results indicate that the proposed approach provides a practical and reliable solution for dispersion characteristic acquisition under spatially constrained conditions, offering strong potential for SHM applications in real engineering structures.

Ultrasonic guided waves are widely employed for defect detection in plate structures due to their long-range propagation and subsurface sensitivity. However, Lamb wave generation and detection exhibits inherent dependence on environmental temperature, adhesive layer thickness, and measurement uncertainties, which often introduce spurious artifacts in defect localization maps. To address this critical limitation, this study proposes a hybrid-domain imaging framework that synergistically integrates time-domain wave propagation analysis with frequency-spectral entropy quantification. A path-weighted elliptical algorithm is developed to mitigate environmental-induced noise while preserving spatial resolution. Finally, an experiment was conducted to verify the location accuracy of the proposed method, and the experimental results show that artifacts can be significantly suppressed.

Domain adaptation has been widely used in variable condition fault diagnosis of mechanical equipment, due to its ability to effectively address the degradation of model generalization performance caused by differences in data distribution. However, the success of domain adaptation methods typically depends on sufficient access to target domain data, which significantly limits their practical application scenarios. To tackle this problem, this article proposes a novel domain generalization method called integrating causal learning and distributionally robust optimization (ICLDRO). In this method, a causal learning-based encoding-decoding system is designed to generate augmented data that maintains consistent semantic information and constructs uncertainty sets by the augmented data. Distributionally robust optimization (DRO) is then executed on the uncertainty set to enhance the robust domain generalization performance of the model on unknown target domains. The effectiveness of ICLDRO is validated through experiments on one public dataset and two private datasets. The results demonstrate that ICLDRO outperforms several state-of-the-art methods across most generalization tasks.

To address the challenges of multimode interference and complex signal interpretation in guided wave-based delamination monitoring of composite structures, while overcoming the drawbacks of conventional single-mode piezoelectric transducers—including high cost, limited bandwidth and coupling dependence—this study proposes a novel upright waveguide transducer (UWT) based on the mode conversion mechanism. Using a dual-source excitation strategy, a pure S0 mode Lamb wave is generated and guided along the upright waveguide. The dominant in-plane motion of S0 mode in the upright waveguide is efficiently converted into out-of-plane motion of the single A0 mode within the horizontal test specimen. This mechanism suppresses interference from S0 mode and quasi-mode, denoted here as S0′, enhancing both the excitation purity and energy coupling efficiency of the A0 mode. Furthermore, a mode-conversion sensing network based on the UWT is developed, along with a path-grouping discrimination and localization algorithm. This integrated framework effectively addresses challenges such as baseline dependency and environmental variation, while also providing a practical solution for real-time implementation with localization accuracy and simplified measurement. Finite element simulations and experimental validations confirm that the proposed transducer significantly reduces signal interference and improves interpretability, offering a robust technical solution for composite structures’ online structural health monitoring when coupled with the proposed algorithm.

Ultrasonic Lamb wave detection technology constitutes a non-destructive evaluation approach extensively employed for the identification of flaws within plate-like structures. The conventional method for detecting and localizing defects in isotropic plate-like structures using ultrasonic Lamb waves relies on baseline signal data. However, the reliability of baseline data as a reference value is diminished due to varying working conditions of the structure. Therefore, to overcome the influence of mismatched baseline data, this paper proposes a novel non-baseline Lamb wave defect detection and localization method. Through simulation and experiment studies, it is discovered that defects at different positions have varied impacts on the amplitude of direct wave-packets under the same propagation path. By eliminating differences in the piezoelectric excitation characteristics of the sensing array (normalized through boundary reflect wave), the direct wave amplitude of multiple sensor pairs in the circular array can be compared and ranked. The paths closest to the location of the damage can be identified, enabling to obtain the defect location information. In this paper, the feasibility and effectiveness of this method has been verified by simulation and practical experiments. The experimental data and imaging results obtained over a four-month period demonstrate that, compared to the traditional baseline localization method, the baseline-free method proposed in this study exhibits a greater ability to resist interference caused by changes in environmental temperature. By increasing the number of sensors from 16 to 32, the positioning accuracy can be significantly improved, reducing the positioning deviation from 13 mm to 0.42 mm. This new non-baseline method based on path amplitude matching demonstrates enhanced practicality within the realm of engineering. Notably, this method holds the potential to be synergistically incorporated and applied in conjunction with various other measurement techniques.

针对家用高压清洗机风冷串激电机定子绕组温升过高的问题,以一台HC8840F单相风冷串激电机为研究对象,建立电机全域三维流固耦合传热模型.在此基础上,使用流固耦合传热计算方法模拟获得串激电机内部流场及温度场,并搭建试验平台,通过不同工况下样机定子绕组的温升试验验证串激电机三维流固耦合传热模型的合理性.在模型验证的基础上,分析了风冷系统流场和温度场的分布规律,为结构优化设计指导方向.此外,从降耗设计的角度出发,研究了风冷系统中后端离心风轮对电机内流场与温度场的影响.仿真结果表明,内部导风轮的设置改善了气流流动,能使定子绕组平均温升降低了 3.3℃.

As an important mechanical transmission component, the healthy operation of planetary gear box is related to the safe operation of the entire engineering unit. Convolution neural network（CNN） is often used to solve the problem of planetary gearbox fault classification. However, due to the existence of various noise sources in the actual monitoring, the vibration signal components are complex, and the signal-to-noise ratio is reduced. Only using convolution neural network for fault diagnosis is not effective.Therefore, in this paper, a Singular Value Decomposition（SVD） noise reduction scheme optimized by the Sparrow Search Algorithm（SSA） was proposed. which is used to denoise the monitored vibration signal, highlight the low-frequency fault characteristics, and combine with convolution neural network to realize fault diagnosis of noisy vibration signal. The experimental results showed that the combination of the two methods could make the convolution neural network model converge faster and improve the diagnostic accuracy to 97.43%.

针对垃圾目标分类检测中物体重叠检测效果差的问题,文中设计一种改进YOLACT图像分割模型,并应用于垃圾实时检测中.根据COCO数据集制作适用于垃圾分类的数据集,通过YOLACT图像分割模型进行训练和评估监测,改进YOLACT的主干网络模块;使用Swish激活函数调整Resnet进入层模块和下采样模块以提升图像特征,同时改进YOLACT结构的检测模块;再使用Pointrend方法对检测出的分割结果与特征图像进行多层感知机(MLP)迭代融合,渲染深化边缘点特征,以融合得到的新掩码层取代原输出的掩码层.最后,进行改进YOLACT算法、SOLO算法、Mask-RCNN算法比较和消融实验.结果表明,改进YOLACT算法可提升精度、速度及垃圾图像分割的边缘效果,能够解决一部分垃圾重叠检测问题,在垃圾实时检测方面有较好的应用价值.

餐桌清理机器人能够对餐具进行自动分类与定位,并通过机械臂抓取回收,实现自动清理餐桌的功能.针对餐桌清理机器人硬件性能的局限性和餐具目标的多尺度特点,文中提出一种基于多尺度分离的改进YOLOv4-Tiny实时性餐具检测模型.设计HS-CSP(Hierarchical-Split Cross Stage Partial)模块增强模型的多尺度特征提取能力;应用Mish激活函数改进Leaky ReLU激活函数;应用AdaBelief优化器改进Adam优化器,在自建的餐具数据集上进行训练.通过对比实验得出3个改进点可有效提升模型的检测精度.最后,对比YOLOv3算法、CenterNet算法和YOLOv4-Tiny算法的检测准确性与检测速度.实验结果表明:文中的改进模型有较好的综合性能,准确率达到86.13％,检测帧数达到176 f/s;与另外两种算法相比,参数量相近时,YOLOv4-Tiny算法的检测精度有所提升,且检测速度可以满足实时性要求.该模型在餐桌清理机器人的餐具检测方面具有较好的应用价值.

文章在回顾过程装备与控制工程专业发展历程的基础上，辨析了新经济形势下该专业存在的问题，并提出“四位一体”的专业升级模式及具体举措。该模式的实施，有助于提高学生的工程实践能力、创新能力及跨界整合能力，同时有助于提升本专业的影响力。

针对气动调节阀停机检修损失大、运行状态不易评估、故障模式复杂且故障诊断极为依赖工程经验等问题,结合残差法实现了故障的在线检测,并对传统的遗传算法进行改进,提出一种基于改进遗传算法优化的支持向量机(Improved Genetic Algorithm optimized Support vector machine,IGA-SVM)的故障诊断算法,作为实现气动调节阀在线故障诊断的算法基础.实验选取四类典型故障进行模拟、测试及数据分析,对阀门故障进行检测与诊断,结果表明:基于残差法的故障检测方法可以有效检测气动调节阀的故障发生,IGA-SVM算法测试诊断率达到92.67％,相较于传统方法具有提升.

Pressure vessels are prone to defects due to environmental conditions, which may cause serious safety hazards to industrial production. The probabilistic ellipse imaging method, based on ultrasonic guided wave, is a common method for locating defects on plate-like structures. In this paper, the research showed that the accuracy of the traditional probabilistic ellipse imaging method was severely affected by the truncation length of the signal. In order to improve the defect location accuracy of the probabilistic elliptic imaging algorithm, an adaptive signal truncation method based on signal difference analysis was proposed, and a novel probabilistic elliptic imaging method was developed. Firstly, the relationship model between the signal difference coefficient (SDC) and the distance coefficient was constructed. Through this model, the distance coefficient of each group signal can be calculated, so that the adaptive truncation length for each group of signals can be determined and the truncated signals used for defect imaging. Secondly, in order to improve the robustness of the new imaging method, the relationship between the defect location accuracy and SDC thresholds were investigated and the optimal threshold was determined. The experimental results showed that the probabilistic ellipse imaging algorithm, based on the new adaptive signal truncation method, can effectively locate a single defect on a pressure vessel.

电动调节阀在石化、核电等领域中有着广泛应用,但由于其结构复杂、工况恶劣、故障众多且呈现非线性关系等特点,开展电动调节阀在线监测与故障检测的研究显得尤为重要.为此,基于多年的经验提出一种基于阈值的故障检测方法,并搭建在线监测实验平台,通过模拟卡涩故障实验来验证故障检测方法的可行性.针对实验信号的信噪比低和信号微弱等问题,使用了一种基于均值滤波和S-G滤波的阀门数据处理方法,研究结果表明:基于阈值的故障检测的方法,能有效地识别电动调节阀的运行状态.

为研究弹簧刚度对往复泵排液阀阀芯运动特性的影响,采用FLUENT软件对往复泵排液过程中柱塞的运动与阀芯启闭动作进行动态耦合数值求解,得到了阀芯运动规律及各时刻流场分布,在模型验证的基础上,分析了定刚度弹簧及变刚度弹簧对阀芯运动特性的影响.结果表明:使用定刚度弹簧时,随着刚度系数增大,阀芯的关闭滞后高度减小,有利于提高往复泵的容积效率.但是刚度系数增大会导致阀芯的升程减小,排液性能降低;减函数形式的两组变刚度弹簧使阀芯的升程分别提高26.1％和11.4％,滞后高度分别降低25.9％和11.1％.增函数形式的两组变刚度弹簧对阀芯升程分别降低2.4％和提高0.1％,滞后高度分别升高63％和20.4％;减函数形式的变刚度弹簧随刚度系数变化范围增大,往复泵泵阀的运动性能提升越显著.

为增加垃圾拾取机器人的自主感知能力,提出了一种用于垃圾跟踪视觉系统的基于YOLOV4改进的轻量级目标检测算法YOLO-TrashNet.针对视觉跟踪系统速度与精度权衡问题,在YOLOV4的基础上将主干网络替换为MobileNetV3,分析了SE (Squeeze-and-Excitation)注意力机制、CBAM(Convolutional Block Attention Mod-ule)注意力机制以及CSP跨级局部网络结构对算法性能带来的影响.搭建了垃圾回收机器人视觉系统,使用了能提高目标定位能力Realsense深度相机,采集了公共场所最常见的15类垃圾,完成了室内垃圾跟踪实验.实验结果表明,提出的以CSPMobileNetV3-CBAM为主干网络的模型能大幅提升检测速度,与YOLO-V4相比计算量降低了93.3％,权重大小仅为19.5 MB,内存消耗低于YOLOV4-tiny;在Jetson Nano运行环境上相比YOLO-V4的垃圾检测牺牲了4％的精度,但是速度提升了6倍,mAP为86.3％.

针对过山车轨道寿命预估难题,完成了基于轮轨耦合关系的过山车轨道寿命预估.首先在ADAMS中基于轮轨耦合进行仿真计算得到轮轨接触力,随后在ANSYS中建立过山车全轨道有限元模型并进行仿真计算,结果显示轨道螺旋段应力明显大于其他轨道单元段,轨道结构中枕轨应力集中现象突出,故以螺旋段枕轨应力最大处作为疲劳校核点.在获得疲劳校核点的应力-时间历程后运用雨流计数法提取循环载荷,最后采用基于疲劳损伤累积理论对轨道进行寿命预估.

Splitter blades are often used to improve the performance parameters of super-low specific speed centrifugal pumps, while the number of splitter blades is the most important influencing factor of head and efficiency. In order to study the effect of different number of splitter blades between long blades on the external characteristics and internal flow field of centrifugal pumps, the numerical simulation of four impeller models has been carried out by Fluent. The results show that splitter blades can raise head and efficiency of the centrifugal pump, so that the performance curve moves toward large flow rate area. The low speed area on the pressure surface of the long blade reduces with the slip and separation inhibited. When the number of splitter blades increase gradually, the head of the centrifugal pump has not much increase while the efficiency decreases. Most of the medium flow out through the runner on the pressure surface leaving the others blockage due to the backflow and vortexes, resulting in large hydraulic loss.

针对有源噪声控制中非线性因素影响建模精度和控制效果的问题,采用神经网络代替传统模型,推导对应的控制算法.用训练结果验证了神经网络对次级通道辨识模型精度的提高.以管道为实验对象,搭建有源噪声控制实验平台,进行噪声控制实验,将传统次级通道模型与优化次级通道模型的实验结果进行对比.结果表明:在低频条件下,针对单一频率和两种频率混合的噪声源,相比传统模型和算法,神经网络优化模型和算法取得了较好的效果.

过程流体机械是过程装备与控制工程专业的主干课程,也是与工程实践紧密结合的一门课程,是培养学生工程意识和工程创新能力的重要课程.文章以华东理工大学过程流体机械课程为例,在回顾该课程发展历程的基础上,结合新工科对创新人才培养的具体要求,分析课程面临的挑战,从新理念、新方法、新内容等方面提出课程改革的思路,以供全国同类课程的改革与实践借鉴.