Indirect Detection of Embedded Tiny Torque Screw Preloading Using Suspended Fiber Bragg Grating Sensor and MFCC-Based Probabilistic Neural Network

The assessment of fastener tightness is essential for ensuring the safety and dependability of fastener assemblies. Monitoring the preloading of an embedded screw in a confined space poses challenges, as most existing sensors are unsuitable for this purpose. This study introduces a new sensor system based on a suspended-Fiber Bragg Grating (FBG). A slender tail of an optical FBG sensor was affixed to the monitored structure, which was subjected to periodic impulse excitation. This excitation generated a multi-harmonic vibration response containing information about the tightness, resulting in a multi-frequency acoustic-elastic wave traveling along the suspended-optical fiber and being captured by the FBG sensor. Following signal extraction, a combined approach utilizing adaptive Mel Frequency Cepstral Coefficient (MFCC) feature extraction and a Probabilistic Neural Networks (PNN) classifier was proposed to detect screw loosening. By optimizing the parameters, an average accuracy of exceeding 99.7% was achieved. This study presents a valuable technique for monitoring loosening, particularly in fastener structures with embedded tiny torque screws in confined spaces.