Efficiency Evaluation of a Combining Compact 1D CNN and Signal Filtering Prototype Algorithm for Vibration Signal Classification

This paper presents an experiment and results of the modified CNN algorithm, it was developed by combining a compact 1D convolution neural network with a tuned signal filter (low-pass filter in this experiment). The algorithm was specifically designed for signal shape detection, prediction, and classification. An experiment aimed to evaluate the created algorithm's structure and signal filter to ensure its suitability for vibration signal data analysis. A testbed was built using an inverter motor equipped with various unbalanced conditions by increasing screws. Four-class unbalanced conditions were studied compared with the normal condition. Each conditions produced the distinct signal shape, with slight differences, making them suitable for evaluating the proposed algorithm's performance. The results demonstrate the satisfied performance of the algorithm, which effectively classifies vibration signals unbalance-3 (precision 54%), unbalance-4 (precision 77<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) and a group of normal, unbalance-1, and unbalance-2. The group of normal, unbalance-1, and unbalance-2 can't be easily classified due to the tiny differences in the signal shapes, caused by small mass variations of the unbalanced screw unbalance-1, and unbalance-2. In conclusion, the algorithm proved satisfied effectiveness for its intended purpose of vibration signal classification.