Neural network impact vibration characterization for mobile additive manufacturing

Impact vibration characterization in mobile additive manufacturing

Abstract

Mobile additive manufacturing (AM) offers a way to ameliorate the long build time problem and benefits greatly if the print job can be abandoned and restarted anew as soon as a significant defect is anticipated. This study proposes a monitoring system using accelerometer data to analyze impact vibrations experienced when vehicle-mounted printers encounter speed bumps. A neural network (NN) Gaussian model method was developed to process these vibration signals, and demonstrated superior performance when compared to the traditional damped least squares approach. The NN method achieved more favorable loss distributions, which were characterized by higher density of loss values near the origin as well as loss distributions that were narrower and skewed towards lower values. The NN approach also achieved good performance when applied on continuous experimental data, achieving throughput of up to 10,133 samples per second. These findings portend the feasible use of the NN method to support the strategy of expeditiously deciding print job cessation based on impact vibration characteristics in mobile AM.