基于振动及EEMD-CMAC算法的鸭蛋散黄在线检测

针对鸭蛋长期存储以及运输过程中造成的散黄问题,构建一种基于振动信息的鸭蛋散黄在线检测流水线,可实现鸭蛋的自动触压和随动检测。通过磁致伸缩振子对鸭蛋扫频振动进行音频信息增强,对音频振动信号进行集合经验模态分解,并通过主成分分析进行降维提取主要特征,基于此,构建基于小脑神经网络的鸭蛋散黄检测模型。试验中,对320枚鸭蛋进行检测(训练集200枚,测试集120枚),结果表明,基于累积贡献率达98.14%的前5个主成分的鸭蛋散黄检测模型训练集和测试集识别率分别达98.66%和97.03%,每枚鸭蛋在线检测时间约1 s。研究表明,所研制的检测流水线基于磁致伸缩振子扫频激励未知品质鸭蛋,再结合EEMD-CMAC进行鸭蛋散黄检测是可行的,可满足流水线在线检测的要求。

Online detection of scattered duck egg based on vibration and EEMD-CMAC

Abstract: Aiming at solving duck egg scattered-yolk problem during transport or the long-term storage, a method is proposed which is a duck egg online non-destructive testing based on vibration information. At the moment, for lack of online non-destruction detection researches for scattered-yolk problem in duck egg'' pipeline in domestic and international sections. Eggs during packaging, processing and transportation link prone to yolk scattered, and result in the decrease of freshness, which makes necessary to pick the scattered egg out timely to avoid flowing into the market. Aiming at the difficult problem of the detection for scattered egg, in this study, we conducted a nondestructive experiment to detect scattered egg effectively by use of combinations of two ways. On the one hand, information of scattered egg was strengthened by putting the magnetostrictive vibrator frequency sweep vibration enhanced audio information; and on the other, multi-dimensional information was analyzed by using the acoustic method .The research was based on the egg detection pipeline system whose central unit is a tailor-made press structure. The pipeline system is controlled by PC software Programmed based on LabWindows\CVI platform. By the connection between upper machine and lower machine, controlling the press structure cooperate with production line to realize detection. Egg can be detected under a constant pressure automatically when it is coming. Experiments were carried out on 100 fresh eggs and 100 scattered eggs. Since the vibration was stationary random signal when eggs in contact with the magnetostriction. Ensemble Empirical Mode Decomposition (EEMD) analysis is suitable for nonlinear and non-stationary signal. Then, in this range, we collected voice band signals of fresh egg or scattered one. The voice band signals were analyzed by combing EEMD analysis with PCA (Principle Component Analysis) primary element analysis. Under the different pressure, the energy spectrum entropies of fresh egg and scattered egg were extracted respectively which were obtained from different frequency band between 21 - 8000 Hz and processed by PCA. Accordingly, the most suitable frequency sweep range was selected when the difference on EEMD was most remarkable between the fresh egg and scattered egg. On the basis of these, there were four kinds of Neural Networks models to build, Partial Least Squares Regression (PLSR), Back Propagation Neural Network (BPNN), Radical Basis Function Neural Network(RBFNN) and Cerebellar Model Articulation Controller(CMAC) in the research, the detection model of scattered egg based on Cerebellar Model Articulation Controller CMAC was the best among the four models. To verify this, we conducted another experiment, using 320 eggs (200 training set, including 100 fresh eggs and 100 scattered eggs. 120 testing set, including 60 fresh eggs and 60 scattered eggs). Two hundred of them were used for CMAC model development and the rest were used to test the model. The results showed that the best frequency sweep range was 201-6000 Hz. In the frequency sweep range, discriminant rate for testing the last 120 eggs including fresh and scattered eggs hybrid modeling training by the system was 98.66%, and the test set recognition rate reaches 97.03%. The detection time was 1s which can meet the need of online detection of pipeline. This study showed that it was feasible to test the scattered-yolk duck egg with the magnetostrictive vibrator frequency sweep to motivate the unknown quality duck egg combined with the EEMD-CMAC detection model which can meet the requirements of agricultural products pipeline detection. It meant this new method effectively solves the difficulty of scattered egg detection with excellent performance which can be applied to industrial production line.