转子图像空时域增强与振动测量方法研究

Rotor space-time domain image enhancement and vibration measurement method

提高视觉测振精度的主要因素取决于单位周期内采集图像的样本数量以及图像中振动目标边缘的清晰度,但工业相机有限的采集频率和固有的分辨率会导致最终获得的时序位移信号因空时域信息丢失而出现明显的偏移误差。因此提出一种转子图像空时域增强与振动测量方法。通过视频插帧算法在相邻的两个样本图像之间插入一个拟合样本,从而增强单位周期内采集的样本数量。为了缓解采集的低分辨率图像中存在模糊、失真、噪点等现象,利用超分辨率重建算法恢复图像内的高频信息。在权衡算法耗时和计算资源损耗后,将视频插帧和视频重建进行任务集成,实现两个模型任务的特征信息共享。在自制的高速振动转子图像数据集中进行测试的试验结果表明:与当前的先进算法相比,本文构建的模型提高了0.3 dB,相应的轻量级模型在降低模型参数17.9%的同时比先进的模型任有更好的重建精度。两种采样频率下测得的时域与频域信号显示,所提出的算法获得的振动信号具有更好的周期性和稳定性。

The main factors for improving the accuracy of visual vibration measurement depend on the number of samples in the captured image per unit period and the sharpness of the edge of the vibrating target in the image.However,the limited acquisition frequency and inherent resolution of industrial cameras will lead to obvious offset errors in the final obtained timing displacement signals due to the loss of spatial-temporal information.Therefore,this paper proposes a method for spatial-temporal enhancement and vibration measurement of rotor images.A fitting sample is inserted between two adjacent sample images through the video frame interpolation algorithm,thereby enhancing the number of samples collected in a unit period.In order to alleviate the phenomenon of blur,distortion and noise in the collected low-resolution images,the super-resolution reconstruction algorithm is used to restore the high-frequency information in the images.After weighing the time consumption of the algorithm and the consumption of computing resources,this paper integrates the video frame insertion and video reconstruction tasks to realize the feature information sharing of the two model tasks.The experimental results tested on the self-made high-speed vibration rotor image dataset show that compared with the current advanced algorithm,the model constructed in this paper is improved by 0.3dB and the lightweight model has better reconstruction accuracy than the advanced model while reducing the model parameters by 17.9%.The time-domain and frequency-domain signals measured at two sampling frequencies show that the vibration signals obtained by the proposed algorithm have better periodicity and stability.