GA-T-S 云推理网络板形模式识别的 DSP 实现

摘要

The existing neural networks we are mostly software simulation and the training time was long,thus that would not conducive to engineering applications.In view of the above problems, flatness pattern recognition via GA-T-S cloud inference network implemented by DSP was presented herein.Firstly,the DSP’s design of T-S cloud inference network was implemented by using TI TMS320F2812 on the basis of flatness pattern recognition via GA-T-S cloud inference network.Then T-S cloud inference network parameters were optimized through genetic algorithm toolbox of MAT-LAB in off-line manner and these parameters were transmitted to DSP later.The network was run on MATLAB and DSP separately.Finally,the two results of T-S cloud inference network,which was run on MATLAB and DSP respectively,were compared and analyzed.Experimental results confirm that GA-T-S cloud inference network have high accuracy in terms of flatness pattern recognition,it can identify the defect types of flatness correctly.At the same time,the experimental results verify that the T-S cloud inference network can run on the hardware TI TMS320F2812 in a fast speed and it provides a basis for neural networks applied to practical engineering.%针对现有神经网络大多是在软件的基础上进行仿真，训练时间长，不利于工程实际应用的问题，提出了 GA-T-S 云推理网络板形模式识别的 DSP 实现方法。首先以设计的板形模式识别 GA-T-S云推理网络模型为基础，利用 TI TMS320F2812完成 T-S 云推理网络的 DSP 设计；然后利用MATLAB 遗传算法工具箱离线优化 T-S 云推理网络参数，将优化后的网络参数存入 DSP 中，进而分别在 MATLAB 与 DSP 上运行该网络；最后将运行结果分别进行显示与对比分析。实验结果证实了基于 GA-T-S 云推理网络的板形模式识别模型有较高的板形识别精度，能够正确识别出板形缺陷的类型，同时验证了 GA-T-S 云推理网络在硬件 TI TMS320F2812上实现的可行性与快速性，从而为神经网络推广应用到实际工程中提供了依据。