Method Combining Machine Vision and Machine Learning for Reed Positioning in Automatic Aerophone Manufacturing

The free reed aerophone, such as accordion, harmonica and melodica, is one of the most popular categories of music equipment in the world. The key operation of the free reed aerophone manufacturing process is to weld multiple reeds onto the reed frame precisely and quickly. In this paper, we propose a method combining machine vision and machine learning algorithms to assist the mechanical device to estimate adjusting displacement and to determine the correctness of the reed positioning operation. Images of reeds on frames are captured and processed, and then some novel features are defined and extracted. Classification models and regression models such as artificial neural network (ANN), support vector machine (SVM), decision tree (DT), k-nearest neighbor (KNN) and linear regression (LR) are applied and trained to estimate if the reed position is correct and to measure the adjusting displacement if necessary. It is found that the Back propagation neural network (BPNN) presents 100% accuracy for the correctness estimation and $\pm 0.025\mathrm{mm}$ measuring precision.