Measurement of curvilinearly varying strain using a Brillouin-gain-spectrum-based method with a multiline strain model

Brillouin gain spectrum (BGS)-based distributed measurements are useful for monitoring the states of structures. A neural network (NN)-assisted BGS-based method is proposed to measure a locally varying strain. We incorporated a multiline strain model (in which the strain is represented by several straight lines) into the NN-assisted BGS-based method. This method trained the NN using the <i>G</i>_MOD pattern, which was a set of BGS values calculated from the model. The measurement error was investigated for a curvilinearly varying strain (specifically, sinusoidal strain produced in a curved beam) using numerical simulations. We observed that as the number of straight lines in the <i>G</i>_MOD pattern increased, the error initially decreased and then increased. The results were discussed based on the relationship between the number of straight lines in the model and the error, which was examined as a basic error characteristic of the method. Furthermore, curvilinearly varying strains were measured using the NN trained using BGS patterns mixed with <i>G</i>_MOD patterns calculated from the multiline strain model with various numbers of straight lines. The number of mixed patterns was determined based on the relationship between the number of <i>G</i>_MOD patterns and the error, which was another basic error characteristic. The maximum error obtained using the mixed patterns was approximately equal to or less than the minimum maximum error of the measurements with the individual multiline strain models. The errors obtained when the BGS observation section (<i>s</i>_z) was 2 m were greater than those when <i>s</i>_z was 1 m.