Development and fuzzy modeling of viscosity of new Halawa (Sesami) products

Halawa industry is very well developed and constantly trying to obtain some competitive products through improving their product quality and to produce new products. Halawa quality products is mainly determined by features such as texture, shape, colour, flavour, and overall acceptance. Halawa are the sweet products of tahina which is typically made by the addition of 50 % tahina and 50 % sucrose, acid and Halawa roots. Halawa is produced through mixing all these ingredients together. In this research, different types of spreadable Halawa were produced through using new additives and evaluated for its textural properties and consumer acceptance. Six different treatments were designed in addition to the control (non spreadable Halawa), produced and evaluated for sensory and rheological properties. It was found that all of these treatments are Non-Newtonian fluids (Viscosity changes with shear rate) and with pseudoplastic behavior which means the viscosity is decreased with increasing the shear stress and shear rate. Moreover, the proposed approach adopted a neuro-fuzzy model (i.e., adaptive neuro-fuzzy inference system, ANFIS) to predict new produced product (i.e., new Halawa) rheological properties, with the shear rate modeled using the inputs (i.e., the speed and torque variables) and the viscosity as the main output. It was found that treatment 5 had the highest viscosity (200–1150 Pa.S) at different shear rate compare to all other treatments and this could be due to the combination effect of 58 % Tahina and 2.4 % Dimodan (emulsifier). However, after the sensory evaluation, it was found that the highest acceptability was found to be for treatment 2. This treatment was less viscous than treatment 5 with less browning. Adding some other types of flavours and colorants can be used to enhance the quality of the final products. The proposed prediction neuro-fuzzy-based prediction model was validated by comparing the actual values of viscosity with the neuro-fuzzy modeled ones. The validation results indicated that the proposed ANFIS model resulted in a 97 % accuracy which is considered very satisfactory when compared with other prediction techniques.