Real-time non-contact optical detection system for roller deformation based on artificial neural network

A non-contact real-time optical fiber measurement system for roller deformation is designed based on the principle of reflective displacement fiber-optic sensor. The theory and mathematical model of the sensor are deduced in detail. By using a novel probe of three optical fibers with equal transverse space, the effects of fluctuations in light source, reflective changing of target surface and the light intensity losses in fiber lines are automatically compensated. Meantime, an optical fiber sensor model of linearizing output signal and correcting static error based on artificial neural network (ANN) is set up. By using interpolation method and value filtering to process the signals based on ANN, the influence of random noise and the vibration of roller bearing can be reduced effectively. Therefore the linearity, accuracy and resolution are enhanced remarkably. Experimental results proved the accuracy of the system could reach to the demand of practical production process. It provides a new method for high speed, accurate and real-time detection of roller deformation.