Combining Ultrasonic Measurement Methods and Machine Learning Techniques to Assess Baked Product Quality

E. Gulsen, D. E. Morris, S. Grebby, A. Ibrahim, N. J. WatsonFaculty of Engineering

University of Nottingham

References:1.Adeola, AA, Ohizua, ER. Physical, chemical, and sensory properties of biscuits prepared from flour blends of unripe cooking banana, pigeon pea, sweet potato. Food Sci Nutr. 2018; 6: 532-540. 2. Ibisworld.co.uk. (2019). Bread & Bakery Goods Production (UK) - Industry Report | IBISWorld. [online] Available at: https://www.ibisworld.co.uk/industry-trends/market-research-reports/manufacturing/manufacture-of-food-products/bread-bakery-goods-production.html [Accessed 14 May 2019]. 3.Salazar, J., Chávez, J., Turó, A., García‐Hernández, M. (2017). Ultrasonic Applications in Bakery Products. In Ultrasound in Food Processing (eds M. Villamiel, A. Montilla, J. V. García‐Pérez, J. A. Cárcel and J. Benedito).

Why is it important?• Baked products are one of the most popular ready-to-eat snacks

in the world[1]. • The market share of the bakery sector has been steadily

growing. The current year expected growth rate is reported around %2.4, reaching £8.3 billion[2].

• In the food and drink industry, products must meet quality assurance standards, so they meet consumer’s expectation and are fit for sale.

• Most products are still assessed qualitatively in factories by human operators, and there is a need for online non-destructive sensor technologies to improve these.

• The aim is creating a non-invasive, online, smart sensor system using machine learning techniques to meet the expectation of consumers and food industry.

Quality Attributes of Baked ProductsThe baked product industry is one sector that would benefit from a new online quality assessment. Within this sector, the quality of products is determined by the parameters below.

MethodologyThree packets of biscuits were opened on different days to produce changes in the moisture content and texture of the biscuits. One packet of biscuits was placed on a tray and left opened in the laboratory for six days. Another packet of biscuits was left opened in the laboratory for four days. The final packet of biscuits was opened on the day of the measurements.

The speed of sound (SOS) through the biscuits was calculated by measuring the distance and time-of-flight (TOF) of an ultrasonic wave propagating through the biscuit sample. The amplitude of the signal, and the energy were calculated from the received of waveform . The break force of the biscuits biscuit was recorded using a texture analyser.

MATLAB was used to create a correlation graph of the different ultrasonic and TA data as a software.

How can Ultrasonic Measurements be Used? Low-intensity ultrasound is used as an analytical technique either to control a process or to obtain information about different properties of foods such as bubbles in aerated foods, the texture of biscuit and bread dough characterization[3].Ultrasonic sensors are low cost, non-invasive and can operate on opaque systems and therefore attractive to the food industry to be employed in quality assurance, on-line process control, and non-destructive inspection, as they are also hygienic and easy to maintain[3]. The aim of this work was to determine if ultrasonic measurements are capable of detecting changes in the texture of baked products such as biscuits.

ResultsThere is a visible difference between the recorded waveform and data of the six days old and fresh biscuits. In some points, four days old biscuits correlate with fresh biscuits but mostly correlate with six days old biscuits in terms of their amplitude, break force and energy values. The results are based on a small number of samples yet still show differences in the ultrasonic results.

Future Work and DevelopmentsIn the next steps of the research include: increasing sample size and investigating different supervised machine learning methods based on texture and quality parameters. A range of methods will be studied including Support Vector Machines, Artificial Neural Networks and Random Forests. Experimental work will focus on manufacturing biscuits with known textural differences and performing measurements with a non-contact ultrasound system.

Acknowledgments:This PhD project is financially supported by the Ministry of National Education of the Turkish Republic.

Texture Colour

Shape Moisture

Biscuit

The typical ultrasound set-up for experimental measurement is similar to the diagram above.