Gongcheng Kexue Yu Jishu/Advanced Engineering Science

Thermal error prediction and compensation of CNC machine tools is an important technology to improve the machining accuracy and reliability of CNC machine tools. The thermal error of machine tool is time-varying and nonlinear. To improve the accuracy and robustness of thermal error prediction, a numerical control machine tool thermal error prediction model based on attention mechanism and deep learning network was proposed. Using the data conversion strategy, the original temperature data of CNC machine tool was transformed into temperature image, which could be directly used as the input of deep learning network. The complete information of the temperature field of the machine tool was retained by converting the temperature field data into the temperature image points. At the same time, the nonlinear and coupling problems between the temperature measuring points were avoided by using the deep learning modeling method. A recognition network of temperature sensitive points based on attention mechanism was proposed. According to the correlation degree between temperature measuring points and thermal error, different weights were given to each temperature measuring point to avoid the disadvantages of artificial selection of temperature measuring points. A 12–layer deep CNN learning prediction network was established to mine the nonlinear mapping relationship between temperature image and thermal error by using its powerful image feature learning ability. This method does not need to preselect the key temperature points, retained more relationship between thermal error and machine temperature characteristics, and can significantly improve the prediction accuracy of the model. In order to improve the accuracy and generalization ability of thermal error model, dropout regularization method and Adam optimization algorithm were introduced to optimize the structure and parameters of deep convolution neural network. The method shows high prediction accuracy in the thermal error verification of G460L CNC lathe. Compared with the thermal error models based on BP neural network, multiple regression and CNN network, the proposed method performs better in generalization performance.

Landslide dams are usually formed instantaneously by natural forces, the accumulation bodies have the characteristics of complex space structure, wide gradation of dam materials, poor dam stability, and they are easy to fail under the flow erosion. As a major natural disaster of flood and drought, safety evaluation and disaster prediction of landslide dams have been the focus of attention by the scholars around the world, but many questions remain unanswered, which are mainly manifested in: 1) Accumulation bodies are composed of natural wide-graded rockfill materials with significant state-dependent correlation, there is a lack of the state-dependent dilatancy theory and constitutive model of wide-graded rockfill materials. 2) After the formation of landslide dams, they would be affected by external loads, such as the rise of upstream water level of dammed lake, continuous unsteady seepage, landslide surge in the dammed lake, and earthquake, there is also a lack of standards and methods for stability evaluation. 3) Due to the lack of necessary flood relief facilities, the landslide dams are prone to fail; under the action of outburst flow, obvious nonlinear characteristics are manifested during the breach development, as well as strong unsteady flow characteristics of the hydraulic elements; there is a lack of numerical models for landslide dam breaching which can reflect the erosion mechanisms of wide graded materials. Therefore, it is necessary to conduct integrated scientific measures, such as field explorations, multi-scale physical model tests, and numerical simulation methods, so as to reveal the physical description, internal structure, macroscopic mechanical properties of the landslide dams and their spatial and temporal variations; and then, a state-dependent (i.e., gradation, pore ratio, and stress level) dilatancy equation for wide-graded landslide deposit will be presented, and a generalized elastic-plastic constitutive model that can adapt to complex stress paths and the limit equilibrium analysis method of landslide dam body will be established. Large-scale hydraulic model tests and centrifugal model tests of dam breaching will be conducted to reveal the dynamic erosion characteristics of landslide dam materials and the evolution law of breaches under the action of unsteady flow. Subsequently, the erosion equation of sand-laden flow under the dynamic boundary condition by the action of unsteady flow will be established, and the numerical model for dam breach process considering the fluid-solid interaction will be put forward to realize the numerical simulation of the characteristics of water flow movement, the law of dam material transport, the evolution process of breaches, and the structural instability of landslide dam in the whole process of overtopping and seepage failure. Integrating the reliability theory and numerical simulation method of dam breach process, an integrated numerical simulation platform for seepage, deformation, stability, and failure process of landslide dams considering fluid-solid will be developed; consequently, the theoretical system and method of safety evaluation and disaster prediction of the full-life cycle of the landslide dams will be established. The expected results achieved in the project will provide scientific theory and key technological support for improving the decision-making level of disaster prevention and reduction of landslide dams in China.

The mechanical properties of the interface between soil and structure have always been a hot topic in geotechnical engineering. In order to explore the interfacial shear characteristics of non-water reaction polymer and concrete, the effects of vertical stress and shear rate on shear strength and shear modulus of polymer concrete interface were studied based on monotonic direct shear test. The experimental results showed that: under the given vertical stress and shear rate, with the increase of shear displacement, the polymer concrete interface presented shear softening phenomenon. The shear rate had little effect on the interfacial shear strength, cohesion and friction angle, but had a significant impact on the interfacial shear modulus, and the shear modulus value decreased with the increase of shear rate, and the decrease amplitude was obvious; the vertical stress had a significant impact on the shear strength and shear modulus of polymer concrete interface, and the shear strength and shear modulus of polymer concrete interface changed with the increase of shear rate. The vertical stress increased continuously. At the same time, the hyperbolic constitutive model formula of polymer concrete interface was systematically deduced, and the validity of the model was preliminarily verified according to the relevant experimental results.

With the rise of development and utilization of marine resources, inclined pile foundation with good horizontal bearing capacity has been widely used as marine structures such as cross-sea bridge, offshore platform and high pile wharf. Under the action of strong nonlinear superposition of wave and current, severe local scour will occur around the inclined pile, which will reduce the safety of the structure and even lead to structural damage. In order to understand the influence of inclination angle on the scour characteristics of downstream inclined piles under different wave heights and flow velocities, and the difference from that of vertical piles, local scour tests under combined waves and current and pure current were carried out in the wave-current flume. By conducting tests on α=0°, 10°, 20° and 30° downstream inclined piles, the influence of inclination angle on scour duration, relative time scale and scour hole morphology was studied. The influence of dimensionless parameters Fr, KC and Ucw on maximum scour depth of downstream inclined pile was analyzed, and detailed comparison was made with previous vertical pile data. The results showed that when the pile was inclined to downstream, inclination angle had an important influence on the scouring characteristics of pile under combined waves and current. Compared with the vertical pile, with the increase of inclination angle, development rate of scour depth, scale and depth of scour hole gradually decreased, while ralative time scale increased gradually. For the 20° downstream inclined pile, when d50=0.403 mm，Uc=0.206 m/s，H=7 cm, the time for maximum scour depth point to be transferred from lateral front to front was about 60 min, which was much longer than that of a vertical pile. The dune of downstream inclined pile was distributed symmetrically along the central axis and had a bimodal structure, and the scour hole shape was obviously different from that of a vertical pile. Under the condition of moveable bed, regular sand waves were formed on the bed surface, the scour hole scale was further reduced, and the double peaks were more obvious. The relative scour depth of downstream inclined pile was the function of dimensionless parameters Fr, KC and Ucw, which was consistent with the trend for a vertical pile’s fitting curve.

Deflection is the basic data of bridge health monitoring. It can provide quantitative information for both structural safety evaluations and maintenance purposes. By analyzing the relationship between the displacement and its first and second derivatives (inclination and curvature), a method to detect the local stiffness damage of the structure by using the area difference of curvature of the deflection curve before and after the damage was proposed. The area enclosed by the curvature curves before and after damage was divided into several elements, and the ratio of the square of each element area to the sum of the square of all elements area was used as the damage location parameter ΔAκ. The peak value of ΔAκ can locate the damage. As a theoretical example, a simply supported T–beam bridge model with different degrees (5%～50%) and different numbers of local stiffness damage was simulated. By calculating ΔAκ value of each element, the damage in the structure was accurately detected. The method was also applied on a PMMA model bridge in the laboratory. The results showed that ΔAκ value at the damage location was much larger than that at the undamaged location, which could accurately locate the minor local stiffness damage in the structure, and ΔAκ was independent of the damage degree. The damage could be clearly located even in the case of measurement noise. This method is helpful to improve the efficiency of daily safety inspection of bridge, quantify the inspection data, objectify the identification results, and promote the application in actual bridge engineering.