为实现钢球表面缺陷的自动检测，应用模式识别理论中的Fisher判别分析识别钢球表面缺陷。利用涡流方法检测钢球表面缺陷，提取了差分线圈输出信号的包络，以包络信号的绝对均值和幅值谱峰值为特征量，根据Fisher准则函数设计了识别算法。验证了Fisher算法识别钢球表面缺陷的有效性，探讨了阈值对识别结果的影响。试验结果表明，选定合适阈值后，该算法可有效识别钢球表面缺陷。

A Fisher discriminant analysis method was presented for the purpose of automatic detection of surface defects on steel ball. The eddy current testing method was utilized to detect the defects. Hilbert transform was employed for the envelopment demodulation analysis of the output signal. The features of the envelop signal was extracted and the classification algorithm based on Fisher criterion function was designed. The validity of the algorithm was verified and the threshold influence on the results was also discussed. The results showed that this method recognized the defects effectively after the threshold adjustment.