A vibration response identification neural network with resilience against missing data anomalies. (1st July 2022)

• Record Type: Journal Article
• Authors: Zhang, Ruiheng; Zhou, Quan; Tian, Lulu; Zhang, Jie; Bai, Libing
• Is Part Of: Measurement science & technology. Volume 33:Number 7(2022)
• Publication Date: 2022-07-01
• Subjects: vibration measurement -- convolutional recurrent neural network -- missing data anomaly; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/ac5c91
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: Vibration measurement provides valuable information pertaining to the mechanical properties of a vibrating structure. However, anomalies caused by sensor faults, system malfunctions, and environmental effects impede the accurate measurement of vibration responses that are essential in determining these properties. Usually, responses with anomalies need to be discarded to attain meaningful interpretation of vibration. In this study, a neural network (NN) model able to classify vibration responses with missing data anomalies is proposed, using convolutional recurrent neural network (CRNN) as the fundamental component for spatio-temporal feature extraction. The proposed model was employed using contact measurement in carbon fiber reinforced plastic plate to identify partially missing responses of different load locations. The integrity destruction algorithm is introduced to simulate responses with missing data anomalies. The proposed model utilized a multi-input structure to reduce the computation cost of the training process. Two CRNNs were evaluated and the results showed that the hybrid architecture of convolutional neural network and long short-term memory (LSTM) is a better choice for the proposed model compared to that of convolutional long short-term memory (ConvLSTM). The proposed model was evaluated by samples with missing data anomalies. The experimental results show that the proposed model has good per-class precision and recall (above 80%) for the classification of measurements with missing data anomalies.