Research on optical fiber sensing strain field prediction algorithm based on machine learning

To accurately predict the strain field distribution at the assembly position and achieve online calibration of the assembly process, built on an online monitoring system for assembly status based on FBGs network. The installation positions of various FBG sensors were designed through strain field simulation analysis. A strain field prediction model based on support vector machine (SVM) algorithm was designed. It uses FBGs data to predict the strain field distribution at the assembly position for correcting the assembly position and posture. In the calibration experiment, the strain response curves of six FBG sensors were tested, and their linearity was above 0.98. The maximum sensitivity was 2.67 pm/N and the minimum sensitivity was 1.12 pm/N. Three typical assembly anomalies were compared. For different types of assembly anomalies, the response wavelength values and distribution characteristics of each FBG sensor have significant differences. For the same type of assembly anomaly, there is a good linear relationship between the response wavelength values and strain values. In the prediction experiment, the maximum error of strain prediction for the validation set was 4.37 με, and the average error was 2.54 με. The predicted results are close to the training set results.