Qualitative modelling of a multi-step process: The case of French breadmaking

https://doi.org/10.1016/j.eswa.2007.11.006Get rights and content

Abstract

In this paper, we investigate a problem of qualitative modelling of a multi-step food process, French breadmaking. The French breadmaking process has been represented as a sequence of steps where each step is defined through its control variables, the state variables of its output, and the causal relations between the control and state variables. In addition, the output of a step is the input of the followed step, and then, the state variables of one step depend on its control variable and the state variables of its input. A qualitative model of mixing, certainly the most complex operation of breadmaking process, has been built up. Human experts reasoning has been represented through seven cognitive operations and a qualitative algebra (Q, ≈, ⊕, ⊗) has been defined to model the calculation of state variables of mixed dough from the mixing control variables and the ingredients (the mixing input) condition. The relations well known by the human experts and other relevant relations between state variables of the mixed dough have been found out through the qualitative equations established. The mixing model has been implemented using the QualiS© expert-system shell. The score established according to French standard of the dough after the mixing step was compared to the one computed. In most of the 81 cases simulated, satisfactory results were obtained since the only unfavourable cases had never been experimented by the experts before, which finally validated this approach, thus worth to be extended to the following steps of breadmaking process.

Introduction

The knowledge linked to the explanation and the achievement of manufacturing through industrial processes is mostly tacit and not stated explicitly. This is all the more significant in the case of agricultural products and food processing, an industry field where empiricism and recipe rules have been used for a long time. The expression of this knowledge requires a scientific analysis of the implied processes, by formalizing the result of this analysis, as well as a collection of the knowledge of the various actors involved. This collection focuses on the people (craftsmen, industrial) having a practical implication in the processes and also with the scientists of the research groups which study them. Scientific knowledge is distinct from the know-how. A scientific knowledge deals with phenomena that, starting from a cause, give an expected result; in this case, the cause, the result and the law that links them, are known. Conversely, in the case of know-how, only the repeatability of the result, given by a cause, is seeked, but the phenomena that relate them are not fully taken into account; only the cause and the result are ascertained. A common formalization of both concepts would allow to design a complete knowledge basis of the industrial process concerned. Such a knowledge basis would be the testimony of the actual practice and the starting point for future investigations.

In the present work, we deal with agricultural product processes, a field in which cognitive approaches of computer modelling are just emergent (Liao, 2005, Bimbenet et al, 2007). In these processes, the technology is based on material sources (ingredients, consumables and equipments), a context, the time and a technical pathway made of linked actions, to be executed. Contrary to many manufacturing industrial fields, raw materials are not constant, and products often change and the technical control of processes is still approximate. This great complexity has impaired the application of a qualitative physics approach, up to now (Shawn & Menon, 1990).

The knowledge used is generally granular and based on rules from know-how and profane knowledge. In this context, different kinds of variables are employed: measurements, observations, words… Representing and modelling of such complex systems require a systemic approach including a definition of an homogeneous calculus and operating space in order to simulate experts reasoning. Qualitative representing and reasoning, from qualitative physics (Hayes, 1979), makes it possible to study formally these systems. A methodology based on a space of quantities including seven symbolic elements (Guerrin, 1995) has been developed. In this frame, the cognitive operations have been defined to describe the reasoning of human experts in agricultural products processing (Ndiaye, Péron, & Fleurat-Lessard, 1998).

Here, we will focus on linear processes made of one of several operating steps. Cheese ripening is an example of a single step process (Perrot et al., 2004), whereas French breadmaking is representative of a multi-step process since it includes different operations of modifications of ingredients from dough mixing to baking. The aim of this work is to apply this methodology to this process (breadmaking) in order to extend the modelling of complex systems with knowledge basis. This process gathers all the challenges of linear processes of agricultural products: different kinds of modifications (biochemical, chemical and physical) take place at various structural levels in batch on the same product. Moreover, the features of the ingredients, and each step, may generate defects of the dough and bread. The notion of defect is a relative one, which means non-adequacy to sensory and/or economical target properties.

Section snippets

Quantities space (Q)

The French breadmaking process includes the following operations: mixing ingredients into a dough, dough rising, dough shaping, dough scoring and baking. The outputs of the processing operations are characterized through the dough and bread condition as state variables (dough stickiness, colour, etc.).

The state variables are influenced by the control parameters of the processing operations, the control variables. Every state and control variable takes a value qualified as being insufficient,

Definition of the French breadmaking problem

Bread results from a succession of operations that might look simple: mixing, fermentation, shaping and baking. These processing operations are unit processes linked up as steps of a breadmaking process.

From a physio-chemical point of view (Bloksma, 1990), the successive stages bring an initially dispersed granular medium (flour) that becomes hydrated once being mixed, to a viscoelastic macroscopically homogeneous mass (dough). Dough is structured by a network of gluten containing suspended

Results and discussion

With the transformation of the decision tables into qualitative functions, we note that in the current knowledge of the dough mixing domain in French breadmaking, none of the eight state variables is influenced by the four control variables (Fig. 7). We observe that (i) linear speed difference between the bowl and the mixing arm (LS) influences only the smoothing velocity (SV) although it is associated in the decision tables collected systematically with the dough overheatingT); (ii) the

Conclusion

This work aimed at building up a qualitative model of a multi-step food process, with special emphasis on breadmaking. Each output of one step was taken as the input of the next one, and was considered as state variable of bread dough, or baked bread, whereas every step was governed by control variables. So, the state variables of one step were influenced by input and control variables. Modelling the whole process could then be reduced to the modelling of the different steps in the same space

Acknowledgements

This work has been developed in the frame of the scientific project “AsCoPain” supported by department CEPIA (processing and characterization of agricultural products). The authors appreciate Hubert Chiron (INRA-BIA)’s most valuable contribution for his expertise in breadmaking.

References (26)

  • P. Hayes

    The naive physics manifesto

  • J. Lefebvre

    An outline of the non-linear viscoelastic behaviour of wheat flour dough in shear

    Rheologica Acta

    (2006)
  • S.-H. Liao

    Expert system methodologies and applications – a decade review from 1995 to 2004

    Expert System with Aplications

    (2005)
  • Cited by (0)

    View full text