In recent years, distributed fiber optic sensing (DFOS) has been widely used for structural deformation monitoring. Conventional deformation monitoring methods using DFOS have a limitation in that they are based on the conventional assumptions in continuum mechanics such as uniformity and continuity. These methods are unsuitable for cracked concrete structures that exhibit discontinuity (discrete body). This paper proposes an advanced neural network method that can accurately measure the deformation of cracked concrete structures. The basic principle of the method is introduced, and its performance is validated by conducting a concrete beam loading test. Subsequently, the deformation calculation results obtained using the proposed method are compared with those obtained using the conventional conjugate beam method. The two methods could accurately estimate the deformation of the concrete beam before crack formation. The proposed neural network method could accurately determine the deformation with a maximum error of only 4.3% at all loading levels even after crack formation, whereas the conventional method exhibited a maximum error of 33.6% in this stage. The test results confirm the advantages of the proposed method in making accurate estimations of the deformation in concrete structures with cracks.

近年来，分布式光纤传感（DFOS）已被广泛用于结构变形监测。使用DFOS的常规变形监测方法具有局限性，因为它们基于连续性力学中的常规假设，例如均匀性和连续性。这些方法不适用于具有不连续性（离散体）的开裂混凝土结构。本文提出了一种先进的神经网络方法，可以准确地测量开裂的混凝土结构的变形。介绍了该方法的基本原理，并通过进行混凝土梁荷载试验验证了其性能。随后，将使用建议方法获得的变形计算结果与使用常规共轭梁方法获得的变形计算结果进行比较。两种方法可以准确地估计裂缝形成前混凝土梁的变形。所提出的神经网络方法即使在裂纹形成后也能在所有载荷水平下准确地确定变形，最大误差仅为4.3％，而传统方法在此阶段的最大误差为33.6％。测试结果证实了该方法在准确估计带有裂缝的混凝土结构中的变形方面的优势。