摘要
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。
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 MATLAB 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.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2016年第17期2341-2347,共7页
China Mechanical Engineering
基金
国家自然科学基金资助项目(61007003)
河北省自然科学基金-钢铁联合研究基金资助项目(E2015203354)
河北省教育厅科学研究计划
河北省高等学校自然科学研究重点项目(ZD2016100)
关键词
T-S云推理网络
板形识别
数字信号处理器
硬件实现
遗传算法
T-S cloud inference network
flatness recognition
digital signal processor(DSP)
hardware implementation
genetic algorithm(GA)