The eXTP (enhanced X-ray Timing and Polarization) satellite is a prominent X-ray astronomy satellite designed primarily for conducting deep space X-ray astronomical observations. The satellite’s scientific payload consists of X-ray focusing mirrors. In order to fulfill the requirements of weight reduction and enhanced effective area, the thickness of mirrors is reduced to the sub-millimeter range and a multi-layer nested structure is employed. Manufacturing mirrors poses a significant challenge to both their quality and efficiency. The present research investigates the optimal replication process for mandrel ultraprecision machining, polishing, coating, electroforming nickel, and demolding. It analyzes the factors contributing to the challenging separation and the inability to release the mirror shells. Additionally, an automatic demolding device is developed, and the X-ray performance of the replication mirrors is verified. The fabrication process flow of the mirrors was initially introduced. To ensure the easy release of the mirror shells from the mandrels, a layer of diamond-like carbon (DLC) was applied as a release layer between the Au and NiP alloy. The adhesion strength of Au-C was found to be significantly lower than that of Au-NiP, as demonstrated by both molecular dynamic simulation and tensile testing. The development of an automatic demolding device with force feedback has been successfully completed. The reduction in the half-power diameter (HPD) of the mirror from 48 inches to 25 inches is an improvement that surpasses the production target.

Tool condition is crucial for guarantee processing efficiency and surface quality. However, the complex industrial environment, limited computational conditions, and less training data present a challenge to the accuracy and efficiency of tool condition monitoring (TCM). This article introduces a lightweight deep learning method that achieves highly accurate end-to-end image segmentation of tool wear, thereby reducing computational complexity and memory consumption. To reduce the model parameters, axial depthwise (DW) convolution is used, which combines cross-shaped and DW convolutions to expand the reception field and extract effective contextual semantic information. In addition, the squeeze-and-excitation module (SE) is introduced into the multilevel jump path to provide attention information on the tool wear region. The tool wear is then obtained through traditional image processing including morphological filtering, Canny edge detection, and Hough line detection. In situ machine vision system with industrial edge deployment is designed to capture and process the images of milling cutters throughout the tool life-cycle. The experimental results validated the proposed lightweight model that could provide precise prediction of tool wear.