鸡蛋蛋壳裂纹敲击振动功率谱信号特征参数筛选和分析

Detection of Cracked Shell Eggs Based on Vibration Signal by Using Variable Selection and Analysis

运用自行研制的禽蛋裂纹检测装置,可以采集并分析敲击鸡蛋产生的响应信号,检测裂纹蛋。提取敲击响应信号功率谱的10个特征参数,并采用逐步回归法和遗传算法进行优化和筛选,以期选取更有效的特征参数,提高模型检测精度。结果表明,遗传算法筛选结果明显优于逐步回归法。当采用遗传算法筛选的4个特征参数(功率谱信号的第1共振峰对应的频率点、第1共振峰的功率谱与其前4个频率功率谱的方差、前3个共振峰功率谱方差、中低频段功率谱能量比)作为判别模型的输入向量,模型能取得最优结果,预测集判别率可达到97.2%。

A system based on vibration signal was developed for eggshell crack detection.It was achieved by analysis of the dynamically measured frequency response of eggshell excited with a light mechanical.Genetic algorithms and stepwise regression algorithms were employed to select variables from response frequency characteristic of eggs.The performance of genetic algorithms was better than stepwise regression.Optimal discrimination model was obtained when four characters were selected by genetic algorithms.The identification rates of linear discrimination analyses(LDA) the prediction set is 97.2%.The four characters were the frequency numbers of max amplitude value,variances of the value of max amplitude to last four frequency numbers,variances of top three frequency amplitude values,ratio of frequency amplitude values of middle frequency numbers to low frequency numbers.