| The crispness and firmness of fruits that have a high water content,are the key quality attributes,which are associated with maturity,freshness,and taste.However,the sensory evaluation method of crispness and firmness has a number of weaknesses such as easy to be interfered by the ambient noise and unable to quantitative analysis.The mechanical evaluation method usually by a texture analyser is a precise method but the device is expensive and is a destructive method.In this work,we use acceleration sensors to pick up the vibration signals that the sample deliveries.Analysis the vibration characteristics to establish the relation between texture properties and modal parameters and feature parameters in the time/frequency domain.Thus laid a foundation for using the vibration characteristics to forecast food texture.The main research contents and conclusions are as follows.1.vibration signal acquisition and selection of texture characteristics.Designed the vibration test system.Through preliminary experiments and data analysis,selected the appropriate devices and sampling parameters including the cap material of the force hammer,the signal excited location,the acceleration sensors,the response signal measuring position,and the sampling frequency,gain,trigger mode,frequency band et al.Collected the excited and response signals effectively.Through the puncture test by a texture analyser,get the mechanical characteristics related to the sample firmness and crispness according to the F-D curve:quantity of fractures,fracturability,maximum force in the first cycle(MF1)、maximum force in the second cycle(MF2),hardness work done in the first cycle(CW1)、hardness work done in the second cycle(CW2).Analysised the change of the texture characteristics in the process of storage using the ANOVA method.And through the sensory evaluation of the samples,analyzed the correlation of sensory score and texture index.The results showed that the sensory score of sample crispness has correlation with the quantity of fractures and fracturability,the R2 were 0.92 and 0.956 respectively.The sensory score of firmness has correlation with MF1,MF2,CW1 and CW2,and the R2 were 0.971,0.655,0.921 and 0.514,respectively.Thus selected the characteristics:quantity of fractures,fracturability,MF1 and CW1 which has a better relation with crispness and firmness as the study objects.2.Vibration signal preprocessing and modal analysis.(1)Preprocessed the acceleration signals collected including the zero adjustment,the trend relieving,and Intercept fragment signal effectively et al.and then get the exact frequency response functions.Identified the natural frequency and damping ratio of the sample through the orthogonal polynomial method.(2)Analysised the pearson correlation of quantity of fractures,fracturability,MF1,CW1 and the 12 order natural frequencies of the samples of different storage time.The result showed that the fist,second,forth,fifth,and the sixth order frequency(RFl,RF2,RF4,RF5,RF6)all has a significant correlation with the four texture characteristics.Respectively,do a scatter diagram of quantity of fractures,fracturability,MF1 and CW1 with the frequencies of the corresponding significant correlation.Results showed that the best correlation between quantity of fractures,fracturability,MF1,CW1 and the response frequencies was 0.933,0.936,0.839,0.842,respectively.(3)Analysised the pearson correlation of quantity of fractures,fracturability,MF1,CW1 and the 12 order damping ratios of the samples of different storage time.The result showed that the forth order damping ratio(D4)has a significant correlation with all the four texture characteristics,the eighth order damping ratio(D8)has a significant correlation with fracturability,CW1,the third order damping ratio(D3)has a significant correlation with fracturability,and the fifth order damping ratio(D5)has a significant correlation with CW1.Respectively,do a scatter diagram of quantity of fractures,fracturability,MF1,CW1 with the damping ratios of the corresponding significant correlation.Results showed that the best correlation between quantity of fractures,fracturability,MF1,CW1 and the damping ratios was 0.599,0.914,0.623,0.633,respectively.(4)Analysised the first,second,forth,fifth,and the sixth vibration mode.Results showed that the first mode was a pure torsional vibration mode,the second mode was a breathing mode,the forth mode and the fifth mode were similar bending-torsional mode,and the sixth mode was a bending mode.No significant difference was observed between the corresponding order vibration mode of samples of different storage time.3.feature parameters in the time/frequency domain extraction.(1)Extracted 7 feature parameters from the input and output signal in the time domain,including the ratio of the amplitudes,the difference of the amplitudes,the ratio of the mean values,the ratio of the mean square,the ratio of the effective values,the ratio of the variances,and the ratio of the biggest short-term frame energy.Analysised the Pearson correlation of the parameters and the four texture characteristics.Results showed that the ratio of the mean values,the ratio of the mean square,the ratio of the effective values,and the ratio of the variances,all have a significant correlation with the four texture characteristics.Respectively,do a scatter diagram of quantity of fractures,fracturability,MF1,CW1 with these features in time domain.Results showed that the best correlation between quantity of fractures,fracturability,MF1,CW1 and the features in time domain was 0.904,0.813,0.878,0.864.respectively.Follwed to study the effect of multiple linear regression.(2)Totally extracted 24 feature parameters from the amplitude-frequency curve and phase-frequency curve,including the amplitudes at the 12 order frequencies(Al,A2,A3...A12)and the phases at the 12 order frequencies(Pl,P2,P3...P12).Analysised the Pearson correlation of the parameters and the four texture characteristics.Results showed that the quantity of fractures has a significant relation with A4,A7,P12,respectively.Fracturability has a significant relation with A3,A4,P9.respectively.MF1 has a significant relation with A4,A8,All,P12,respectively.CWl has a significant relation with A4,A7,P12,respectively.Respectively,do a scatter diagram of quantity of fractures,fracturability,MF1,CW1 with these features.Results showed that the best correlation between quantity of fractures,fracturability,MF1,CW1 and the features in time domain was 0.715,0.576,0.614,0.567,respectively.Follwed to study the effect of multiple linear regression.4.food texture regression model of vibration characteristics.Build the regression model of the four texture characteristics by the stepwise multiple linear regression(SMLR)method and princinpal component regression(PCA-MLR).Compared the effect of the two kinds of model,and select the optimal model.For the quantity of fractures,the SMLR model is better and select RF6 as the independent variable.The effect is equal to one-variable linear regression.The adjustment coefficient(R2)is 0.933.and the average relative error is 3.51%.For the sample fracturability,the SMLR model is better and select RF2 and D8 as the independent variables.The adjustment coefficient(R2)of the model is 0.984,higher than that of one-variable linear regression,and the average relative error is 1.21%.For the sample MF1,the SMLR model is better and select the ratio of the variances and A8 as the independent variables.The adjustment coefficient(R2)of the model is 0.949,a little lower than that of one-variable linear regression,and the average relative error is 2.04%.For the sample CW1,the SMLR model is better and select the ratio of the variances,D3 and A7 as the independent variables.The adjustment coefficient(R2)of the model is 0.994,higher than that of one-variable linear regression,and the average relative error is 0.72%. |
