Prediction Pack Application - Comparison of noodle flour quality results between RVA and Viscograph

Jennifer M.C. Dang (1), Yejia Wei (2) and Mark L. Bason (3)

(1) Perten Instruments of Australia, 13/2 Eden Park Drive, Macquarie Park NSW 2113, Australia
(2) Perten Instruments (Beijing) Co., Ltd, Room 818, Huaheng Office Building, South Binhe Rd, Xicheng District, Beijing, China
(3) Perten Instruments AB, P.O. Box 9006, SE-126 09 Hägersten, Sweden

Introduction

The pasting properties of flour provide a useful indication of its suitability for making noodles, bread, cakes, batters and other products. Flour pasting quality is now commonly measured using the Rapid Visco Analyser (RVA), nonetheless many users still require the results to be converted to historical Viscograph units for trade. Most commonly this is done by simple correlations, however multivariate analysis (MVA) offers a potentially superior method.

MVA techniques use multiple variables to describe the variation in a sample attribute, such as the wavelength versus intensity spectrum from near-infrared spectroscopy (NIRS) used to predict sample moisture content in grain. For rheological instruments like the RVA, doughLAB or TVT, the corresponding viscosity/torque/force versus time 'spectrum' (curve) can likewise be employed. In this application MVA can potentially predict a desired functional quality more accurately than a single rheological measurement (Juhasz, et al., 2005, Meadows & Barton, 2002, Mijland, et al., 1999, Ohm, et al., 2006).

The Prediction Pack is an optional feature available to users of the Thermocline for Windows (TCW) software for the RVA, the doughLAB for Windows (DLW) software for the doughLAB and micro-doughLAB, and the TexCalc software for the TVT. The Prediction Pack allows the use of MVA techniques on the available data to predict a desired parameter. The Prediction Pack is used to load prediction models first created using The Unscrambler X software, which can then be used to create analysis results.

This paper summarises the comparison of wheat noodle flour quality results between the RVA and Viscograph using univariate and multivariate analyses.

Materials and Methods

Forty-seven noodle wheat flours with varying proximate (ash, gluten) and rheological qualities (Viscograph, Farinograph, Extensograph and Falling Number values) were obtained from a commercial source in Tianjin, China. Viscosity analyses were performed on a Perten Instruments RVA 4500 using the noodle profile (Table 1), at the same sample concentration used in Viscograph tests (15.1%). Viscograph analyses of interest were peak, hold, final, breakdown and setback viscosities, peak time and pasting temperature.

Table 1: RVA noodle profile. 

Time

Type

Value

00:00:00

Temp

60°C

00:00:00

Speed

960 rpm

00:00:10

Speed

160 rpm

00:02:00

Temp

60°C

00:08:00

Temp

95°C

00:12:00

Temp

95°C

00:16:00

Temp

50°C

00:20:00

End of test

-

Idle temperature: 60 ± 1°C, Time between readings: 4 s

Univariate comparisons of the Viscograph and the corresponding RVA results were performed by regression analyses (Minitab ver. 13). Method precision was evaluated from one-way analysis of variance of data as the coefficient of determination (R2) and root mean square of residuals (RMS). RVA viscosity spectra at four-second intervals were exported from the TCW software (v. 3.16) in ASCII format into The Unscrambler software (v. 10.3, CAMO ASA, Oslo, Norway) for multivariate analysis. Partial least squares (PLS) regression models were created between the reference Viscograph values and RVA viscosity spectra, and validated using random cross-validation. The models were created for Viscograph analyses combined (all analyses were predicted with one model) or individually (a separate model was created to predict each individual parameter). The R2 and the root mean squared error (RMSE) were determined for all statistical analyses. A suitable (optimum) model is one with large R2 and small RMSE of calibration (RMSEC) and cross-validation (RMSECV). RMSECV was used in preference to RMSE of prediction (RMSEP) in this study, since it indicates model robustness in the absence of a separate validation data set.

Results and Discussion

Typical RVA and Viscograph curves for each of the test profiles are shown in Figure 1. 


Figure 1. Example viscosity curves of wheat noodle flours tested on the RVA using the noodle profile (left) and Viscograph using the standard profile (right). Points on the Viscograph graph: A - point of pasting temperature, B - peak viscosity, C - end of heating phase, D - end of maximum temperature holding phase, E - end of cooling phase, F - end of test.

Univariate Analysis of RVA and Viscograph Results

Univariate analyses of the corresponding primary results (peak, hold and final viscosities) compared well between RVA and Viscograph (R2 > 89.5%) (Table 2, Figure 2). Derived viscosity parameters (breakdown and setback) had weaker correlations (R2 50 – 70%); this was expected since the derived values were dependent on variation between the multiple parameters. Weak or no correlation (R2 < 50%) was observed between the instruments for peak time and pasting temperature using univariate analyses. This was probably due to the different heating rates used.


Figure 2. Univariate regressions of RVA with Viscograph primary parameters (peak, hold and final viscosities; left) and derived parameters (breakdown and setback; right). n=47.

Table 2. Goodness of fit of univariate and PLS regression models between RVA and Viscograph data. n=47.a

Response

Univariate

 

PLS calibration

 

PLS validation

RMS

R2

 

RMSEC

R2

 

RMSECV

R2

VG Peak

22.2

91.9

 

14.1

96.6

 

17.2

95.2

VG Hold

17.1

93.2

 

10.5

97.4

 

12.7

96.3

VG Breakdown

17.4

56.8

 

9.82

85.7

 

12.2

79.0

VG Final

32.8

89.5

 

17.2

97.0

 

23.8

94.5

VG Setback

22.2

68.3

 

9.90

93.4

 

15.5

84.6

VG Peak time

0.33

46.6

 

0.14

90.2

 

0.17

85.8

VG Pasting temperature

2.23

4.93

 

0.23

88.0

 

0.31

80.2

aPLS = partial least squares, VG = Viscograph, RVA = Rapid Visco Analyser, R2 = coefficient of determination (%), n = number of samples, RMS = root mean square error fit of univariate regression, RMSEC = root mean squared error of calibration and RMSECV = root mean squared error of cross-validation. Error values for viscosity (peak, hold, breakdown, final, setback) are in BU, time in min and temperature in °C.

Figure 3. RVA Prediction Pack software analysis definitions (A) and corresponding models (B) using RVA viscosity spectra (C) to predict Viscograph results (D).

Multivariate Analysis of RVA and Viscograph Results

Multivariate analyses (PLS) of RVA viscosity spectra gave better predictions of Viscograph values than univariate analyses for every parameter (smaller residual errors and larger R2 values, Table 2). Generally, better predictions of Viscograph results were also achieved using the entire RVA curve compared to selected RVA analysis points, and with separate models rather than a combined model (results not shown).

The model using RVA noodle viscosity spectra gave good predictions of Viscograph peak (R2 = 95.2%, RMSECV = 17.2 BU), hold (R2 = 96.3%, RMSECV = 12.7 BU) and final (R2 = 94.5%, RMSECV = 23.8 BU) viscosities, and reasonable predictions of Viscograph breakdown (R2 = 79.0%, RMSECV = 12.2 BU), setback (R2 = 84.6%, RMSECV = 15.5 BU), peak time (R2 = 85.8%, RMSECV = 0.2 min.) and pasting temperature (R2 = 80.2%, RMSE = 0.3 °C).

Predictive models created in The Unscrambler can be exported to the RVA software using the Prediction Pack (Figure 3). With the Prediction Pack, a test configuration combining predictive models and other analysis functions (e.g. for peak, hold and final viscosities) can be created. Additional covariates (e.g. amylose content) could also be added to the model, in which case the value for these parameters will be requested by TCW each time a test is run and then used in the prediction. When the given test configuration is used to test future samples, the software will output the predicted Viscograph values using the RVA viscosity curve.

Conclusions

The RVA 4500, using the noodle test profile and multivariate analysis (PLS) of the viscosity spectra, gave good predictions of Viscograph results. The Prediction Pack can be used to set up the RVA software to perform tests on future samples and output the predicted Viscograph results based on the models created.

References

Juhász, R., Gergely, S., Gelencsér, T. and Salgó, A. 2005. Relationship between NIR spectra and RVA parameters during wheat germination. Cereal Chem. 82(5):488-493.

Meadows, F. and Barton, F.E. 2002. Determination of Rapid Visco Analyser parameters in rice by Near-Infrared Spectroscopy. Cereal Chem. 79(4): 563-566.

Mijland, P.J.H.C., Janssen, A.M. and de Vres, P.A.M. 1999. Multivariate comparison of the Brabender Viscograph and the Rapid Visco Analyser using cross-linked starch. Starch 51(1):33-39.

Ohm, J.-B., Ross, A.S., Ong, Y.-L. and Peterson, C.J. 2006. Using multivariate techniques to predict wheat flour dough and noodle characteristics from size-exclusion HPLC and RVA data. Cereal Chem. 83(1):1-9.