Elsevier

Meat Science

Volume 55, Issue 1, May 2000, Pages 13-24
Meat Science

Prediction of temperature and moisture content of frankfurters during thermal processing using neural network

https://doi.org/10.1016/S0309-1740(99)00120-5Get rights and content

Abstract

An artificial neural network (ANN) was developed to predict temperature and moisture content of frankfurters during smokehouse cooking. Fat protein ratio (FP), initial moisture content, initial temperature, radius of frankfurter, ambient temperature, relative humidity and process time were input variables. Temperature at the frankfurter centre, average temperature of the frankfurter and average moisture content (d.b) of the frankfurter were outputs. Network training data were obtained from validated heat and mass transfer models simulating temperature and moisture profiles of a frankfurter. Backpropagation method was used for ANN training. Selection of hidden nodes, learning rate, momentum and range of input variables were important to ANN prediction. The FP was not an important factor in predictions.

Introduction

Analysis of the fundamental physical processes and the determination of those physical properties important for the water removal processes, provide valuable information as how to prevent undesirable processes leading to impairment of quality. Carefully controlled thermal processes are necessary to develop initial colour, characteristic skin and internal texture, and eliminate microbial population without emulsion destabilization, excessive moisture loss or texture changes (Mittal & Blaisdell, 1982). Little work has been conducted on the prediction of temperature and moisture content using artificial neural network (ANN) during thermal processing of meat emulsion products.

Mittal et al., 1983, Mittal & Blaisdell, 1982 reported prediction of moisture and temperature distribution during thermal processing of meat emulsion products of slab and cylindrical geometries as a function of composition, relative humidity and process time. Prediction of moisture and temperature profiles using computer simulation, based on mathematical models, is a time consuming work. Developing ANN based on simulation data from validated models will be a useful tool to use results in practice and for process control in smokehouses. An ANN is a data processing system based on the structure of the biological neural system. Prediction with ANN is not like modelling and simulation, but by learning from the data generated experimentally or using validated models. ANN has the ability for relearning according to new data (Hertz, Krough & Palmer, 1991).

Unlike other modelling techniques such as simultaneous heat and mass transfer, kinetic models, and regression analysis, an ANN can accommodate more than two variables to predict two or more parameters. ANN differ from conventional programs in their ability to learn about the system to be modelled without prior knowledge of the process variables relationships (Ramesh, Kumar & Rao, 1996). ANN is now used for food process related problems. Park, Chen, Whittaker and Miller (1993) developed a feed-forward back-propagation neural network model for predicting and classifying beef palatability attributes using ultrasonic spectral features as descriptive parameters. Arteaga and Nakai (1993) developed ANN for predicting foam capacity and stability, and emulsion activity index using physicochemical properties of 11 food-related proteins. Greeraerd, Cenens, Herremans and Van Impe (1997) used a low-complexity, black box ANN model to predict microbial growth considering the effects of temperature, pH and water activity. Ruan, Almaer, Zou and Chen (1997) used two spectrum analysis techniques, fast Fourier transformation and power spectrum density for data preprocessing as part of ANN development for predicting rheological properties of a cookie dough.

Therefore, the objective of this study was to develop an ANN for predicting temperature and moisture content of frankfurters during thermal processing using simulation data from validated mathematical models.

Section snippets

Mathematical modelling and data generation

For data generation, mathematical models developed, simulated and validated against experimental data for frankfurters cooking by Mittal and Blaisdell (1982), were used. Brief descriptions on modelling and simulation are given here.

Mathematical models in non-dimensional form characterizing simultaneous heat and mass transfer are represented as:∂θ∂t=αR2(2θ∂ψ2+1ψ∂θ∂ψ)∂C∂t=1R2∂Dm∂ψ∂C∂ψ+Dmψ∂C∂ψ+Dm2C∂ψ2With the following initial and boundary conditions:∂θ∂ψψ=0=0,∂C∂ψψ=0=0,θ(0,ψ)=0,C(1,θ)=0∂θ∂ψψ=1

Network training and testing

From the generated data (13500 sets), 1543 and 1416 data sets were randomly selected as testing and production sets, respectively (approximately 10% each of total). The rest, 10541 data sets were used for ANN training. After every data set training, ANN weights were adjusted. Testing data were fed to test trained ANN after training 400 epochs. Testing errors were recorded. In the beginning of training, testing error decreased with the training process. Training was continued until testing error

Conclusion

  • 1.

    Predicting temperature and moisture content of frankfurters using ANN with simulated data is a simple, convenient and accurate method. A few high relative errors in prediction were caused by boundary patterns in production sets.

  • 2.

    Selection of hidden nodes was important to ANN predictions.

  • 3.

    Appropriate combination of learning rate and momentum improved the predictions.

  • 4.

    The prediction accuracy was improved by shrinking range of input variables.

  • 5.

    The fat protein ratio (FP) was not an important factor in

Acknowledgements

The research was supported by the Natural Sciences and Engineering Research Council of Canada.

References (13)

  • G.S. Mittal et al.

    Moisture mobility in meat emulsion during thermal processing: analysis of slab moisture profile

    Meat Science

    (1983)
  • G.E. Arteaga et al.

    Predicting protein functionality with artificial neural networks: foaming and emulsifying properties

    Journal of Food Science

    (1993)
  • I.J. Dunn et al.

    Biological reaction engineering

    (1992)
  • A.H. Greeraerd et al.

    A prototype grey box model using neural networks for prediction of microbial growth

  • M.Y. Hamdy et al.

    Evaluating film coefficient in single kernel drying

    Trans. of Am. Soc. Agric. Eng.

    (1969)
  • J. Hertz et al.

    Introduction to the theory of neural computation

    (1991)
There are more references available in the full text version of this article.

Cited by (37)

  • Artificial neural network based process modeling

    2019, Handbook of Farm, Dairy and Food Machinery Engineering
  • Artificial Neural Network (ANN) Based Process Modeling

    2013, Handbook of Farm, Dairy and Food Machinery Engineering: Second Edition
  • Modeling sterilization process of canned foods using artificial neural networks

    2005, Chemical Engineering and Processing: Process Intensification
View all citing articles on Scopus
View full text