铸件多模态音频检测模型与在线学习优化

Multimodal Audio Detection Model for Casting and Online Learning Optimization

铸件敲击产生的共振频率与铸件尺寸变化有关,通过共振频率来预测铸件的球化率或蠕化率必然产生较大的偏差。为解决这一问题,提出了由多个共振频率组成的多模态音频检测模型,模型系数由音频检测数据通过在线学习获得。结果表明,多模态音频检测模型能够有效辨识出铸件的弹性模量变化及变化规律,而与铸件尺寸变化无关。模型系数可以通过铸件产品质量的提升及在线学习得到进一步优化。以某种型号的铸件为例,实测了20件样品,通过多模态音频检测模型,提取出了反映铸件产品质量变化的模型值及波动范围,为铸件产品的在线检测及质量划分提供了依据。

The resonance frequency generated by casting percussion is related to the the change of casting size.It is necessary to predict the spheroidization rate or verminization rate of the casting by resonance frequency.To solve this problem,this paper proposes a multi-mode audio detection model composed of multiple resonant frequencies,and the model coefficients were obtained from the audio detection data through online learning.The results show that the multi-modal audio detection model can effectively identify the change of the elastic modulus of the casting and the change law,and has nothing to do with the the change of casting size.The model coefficient can be further optimized through the improvement of casting quality and online learning.In some types of casting as an example,the measured 20 samples,by multimodal audio detection model,extract the model values to reflect the actual changes in quality of casting products and range,for casting products provide a basis for the online detection and quality classification.