A deep learning-based framework for impact load identification using strain data on composite plates

• Published in: Measurement : journal of the International Measurement Confederation Vol. 256; p. 118497
• Main Authors: Wu, Yang, Zhou, Jiangbei, Wu, Shaoqing, Yu, Hualing
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2025
• Subjects: Deep learning neural network; Experimental verification; Impact load; Online identification; Partition collection strategy of training sample; Online identification; Deep learning neural network; Partition collection strategy of training sample; Experimental verification; Impact load
• ISSN: 0263-2241
• DOI: 10.1016/j.measurement.2025.118497

•A novel deep learning-based framework is proposed to identify the impact load on composite plates from strain data.•The error propagation from the load localization to impact load time history reconstruction is considered.•A partition collection strategy of training samples is proposed to enhance the ability of ISMA-SE-CNN-BiLSTM network. Accurate localization and quantification of impact loads are critical for predicting the impact-induced damage and assessing the remaining life of composite structures. To enhance the reliability of impact damage predictions for composite panels, a deep learning-based online impact load identification method is proposed. Firstly, impact load localization is conducted by using a Bayesian-optimized deep neural network combined with a triangulation strategy. Secondly, a new ISMA-SE-CNN-BiLSTM model is proposed to reconstruct the time history of the impact load, in which an improved multi-strategy slime mold algorithm (ISMA) and the Squeeze-Excitation (SE) attention mechanism are adopted to optimize the model’s performance and to enhance the dynamic feature weighting. Numerical simulation studies validate the effectiveness of the proposed method and demonstrate the advantages of the proposed network models. Additionally, a partition collection strategy of training sample is proposed to further improve the accuracy of the reconstructed impact load time history. An experimental study is subsequently conducted on a composite laminated plate subjects to impact load, where the impact loads are reconstructed solely from measured structural strain data. The influence of different sensor arrangements and data collection strategies on load localization and time history reconstruction is discussed. Results show that the proposed deep learning-based approach for impact load localization and time history reconstruction offers robust and accurate prediction results.