基于神经网络的钢琴击弦机故障排除方法研究

Study on troubleshooting of Piano StrMachine based on Neural Network

针对传统钢琴击弦机故障诊断方法准确率低,导致机械故障排除效果不佳的问题,提出基于改进果蝇算法优化BP神经网络的故障诊断方法。基于果蝇算法FOA加入混沌映射、动态搜索半径策略和优化气味浓度判定公式, 得到改进的UFOA算法;然后利用UFOA算法优化BP神经网络,并构建基于UFOA-BP的击弦机故障诊断模型;最后获取钢琴击弦机械故障数据,并通过小波包分解法进行故障数据特征提取。将本模型应用到数据集中进行实验发现,相较于未优化的BP神经网络,提出的UFOA-BP模型的故障预测误差绝对值仅为1.01和0.61,通过UFOA算法提升了BP神经网络的预测精度。且在单弦和多弦故障诊断中,对比于其他诊断模型,本模型的故障诊断准确率分别提升了7.75%、10.08%和7.19%、9.05%。由此说明,通过本模型可提升钢琴击弦机故障诊断率和排除效果。

In view of the low accuracy of the fault diagnosis method of the traditional piano and string strike machine, the fault diagnosis method of optimizing the BP neural network is proposed.FOA algorithm based on F O A, dynamic search radius strategy and odor concentration determination formula, get improved UFOA algorithm, optimize B F O A neural network by UFOA algorithm and build UFOA-BP, obtain the mechanical fault data, and extract the fault data by the wavelet packet decomposition method.Application of this model to the dataset, we found that the absolute value of the UFOA-BP model is only 1.01 and 0.61, which improves the prediction accuracy of BP neural network through the UFOA algorithm.Moreover, in the single-string and multi-string fault diagnosis, the fault diagnosis accuracy of this model is improved by 7.75%, 10.08%, and 7.19% and 9.05%, respectively.This model shows that the troubleshooting rate and troubleshooting effect of the piano string strike machine can be improved.