Recent advances and applications of artificial intelligence and related technologies in the food industry

1. Introduction

Food business is currently more demanding subject area in order to meet the needs of a large number of people in Worldwide. In today’s hectic life style, some people prefer ready to eat foods to save time in the kitchen. However, inadequate man power necessitating the use of machines to meet the food demand. Artificial Intelligence (AI) is an emerging tool in the food sector for the processing of different types of foods through robots or machines. Hygiene plays an important role in the food processing industry. Two types of hygiene such as personal hygiene, equipment hygiene are important. Personal hygiene refers to the hygiene of people who are working in the industry or those who are handling the food products. The main sources are fingernails, hair, mouth infections, and wounds (Hoornstra et al., 2005). Sanitization of equipment should be carried frequently to avoid contamination. Cleaning-in-place (CIP) and cleaning-out-of-place (COP) systems are often used and small-scale industries prefer COP systems due to high cost of CIP systems. Large processing industries CIP systems are preferable to avoid maximum contamination (Less et al., 2003; Lelieveld et al., 2005 ). Best practices are intended to be focused on areas such as food innovation, preparation, and safety (Marriot, 1999). AI is the branch of science that deals with machines to discover answers to complex issues in a fashionable way (Huimin et al., 2018). With the help of human ideas, AI prepare its own algorithm and serve in a computer friendly way and it is a major innovation that increases public activity on a daily basis. It contributes inconceivably to the reasonable improvement of the economy and social issues. Recent times, AI has stood out as a key to have development in various countries, such as Europe, USA, China, India (Huimin et al., 2018). AI is utilized to tackle some true issues. On an extremely comprehensive record, the AI regions are classified into many types (Huimin et al., 2018). In recent years AI has become an important research field to improve the studies in areas like engineering, science, medicine, food and nutrition, marketing, stocks and many more fields. Recently AI is the key factor for the fourth industrial revolution. To advance manageability, new creation requires some viewpoints such as smart creation, escalated research endeavours in the field of AI, with various AI-based methods (Huimin et al., 2018). AGI is a system that portrays the machine to do anything as individuals can do consistently. A broad perspective of man-made consciousness is that it may be the mix of learning, discernment, critical thinking, and problem-solving for the system or structure including logic and reasoning. AI is divided into various ways and helps mankind in different sectors (Fig. 1) (Rupali & Amit, 2017). The classification of AI consists of two types: Weak AI and strong AI. In our daily life, there are weak AI’s for example SIRI, ALEXA, GOOGLE ASSISTANT because these weak AI’s do not have thinking capability, but they are already programmed before. Weak AI demonstrates that virtual capacities include thinking, talking, moving (Rupali & Amit, 2017).

1. Download : Download high-res image (246KB)
2. Download : Download full-size image

Fig. 1. Sectors in Artificial intelligence.

Strong AI machines will perform problem solving and think for themselves and also they predict the appropriate response or solution in the future. For example, IBM created the AI supercomputer named “WATSON” (Rupali & Amit, 2017). Strong AI machines have a mind of their own and can make decisions and be active at all times. These types of AI machines can experience consciousness, learning, thinking, self-awareness, problem-solving, and solve puzzles hence called true intelligence or AGI. Not only AI but the subfield of AI which is ML and robotics plays an important role in many industries.

2. Role of artificial intelligence in food

AI or ML is a moderately new computational tool that has discovered broad usage in real-world problems. AI has been used in different areas such as modeling, classification, and various analysis of data. Sample applications incorporate numerical handling, control plan, correspondence advancements, interpreting pyrolysis, gas Chromatography, mass spectrometry and HPLC information, design acknowledgment of structure RNA, protein, DNA, prediction of microbial growth, biomass, and life span of usability of food items and distinguishing of microbes (Sofu & Yesim, 2007). In food sector handling and engineering particularly neural systems, fuzzy rationale and hereditary calculations procedures have been utilized to improve execution. AI has been utilized in food science and processing for sorting, quality control of food samples, wine analysis. CIP and COP systems help the food industry to maintain its hygiene and keep their product at high standards. Even these systems work with the help of AI. These are called SOCIP (Self-Optimizing Clean In Place). As increasing world population food production is not sufficient due to decrease in agriculture lands, climate changes and increase in pollution (How et al., 2020). All these factors are affecting food production, environment, lack of nutrients, human health. Majority (billions) of people in the world are still suffering with lack of nutrients (How et al., 2020). All-encompassing techniques can be utilized to construct motivations to facilitate the progress and to accomplish better food security (How et al., 2020).

Based on the public demand farmers were adapted new harvesting techniques to produce high yield by using artificial intelligence (Kleineidam, 2020). The development of advanced horticulture and its related innovation has opened various new investigations. AI in agribusiness requires data analysis, decision making, and good activity by applying machine power to the early recognizable proof of yield sicknesses, to give domesticated animals with required nourishment and to increase farming sources of inputs and profits based on demand and supply (How et al., 2020). Some technologies are pest control management, pesticide information which helps farmers to get more yields. Bayesian Network (BN) is a solution for the above problem (How et al., 2020). For this type of networks there is no need for any software skills even beginners can also use it. The Bayesian methodology makes simulations conceivable by incorporating earlier information into data analysis, prior to producing predictive inferences. It is not dependant on traditional frequent approach and does not require any null hypothesis tests. This approach helps to calculate mutual information (MI), which is a probability between data distributors (How et al., 2020). Utilizing these procedures in BN and the collaborations between theoretical factors in the GFSI can be inspected. These Bayesian models are more related to food safety areas in worldwide. BN models are used to display the user in an easy way (How et al., 2020). Hence, AI plays a vital role in food industry not only in quality and food security but also in different areas such as manufacturing, packaging, and cleaning.

3. Applications of AI in dairy products

3.4. Lactose removal in milk by ANN

Lactose is the major carbohydrate along with other nutrients such as fats and proteins in the milk. To avoid the problems caused by lactose such as indigestion or lactose intolerance some of manufactures are removing excess lactose from milk so that consumers can easily digest or consume little amount of lactose (Pawlowska et al., 2016). The removal of lactose through adsorption process which in turn use empirical models by ANN performed through fixed bed column (Balieiro, et al., 2016).With these methods the adsorption rates are also increases because the adsorbent is continuous contact with the solution (Leite et al., 2019). Molecular Imprinting (MIP) can increase the adsorbent efficiency by using ordinary materials in automatic extraction of solid stage engraved silica in a fixed-bed segment (Oliveira et al., 2015). ANN consists of three layers such as input, internal layers, and output layers where input layers consist of independent variable such as time, temperature, bed height, granulometry, flow rate. Output layer consists of dependent variables such as initial concentration, final concentration that exit from bed, the weights were taken from hidden layers to output layers, both layers are connected through intermediate layer with multiple neurons (Balieiro, et al., 2016).

HPLC (High Performance Liquid Chromatography) is used for the detection of lactose content after absorption process. Along with MIP model (RBF) Radial Basis Function model gives high efficiency, speed and simplicity. For different tasks RBF requires a greater number of neurons in each layer but it needs more neurons in hidden layers when compared to MIP. These two models were developed using MATLAB by researchers which in detail are explained by Leite et al. (2019). Both MIP and RBF constructed neural structures are mentioned in detail; these structures’ performance is further compared with some models to check the accuracy (Fig. 2) (Leite et al., 2019).

1. Download : Download high-res image (564KB)
2. Download : Download full-size image

Fig. 2. Schematic representation of inputs by ANN structure.

4. Artificial intelligence in beverages/soft drinks

Drinks are grouped into three categories (a) Alcoholic drinks (b) Non-alcoholic drinks (c) hot drinks. Alcoholic beverages/ drinks contain beer, wine, spirits, and Non-alcoholic beverages/drinks such as juices, carbonated water, milk, and soft drinks such as coca cola, thumbs up, Miranda etc., hot beverages such as tea, coffee, hot chocolate (Viejo et al., 2019). The fermentation process of beer starts by adding yeast to the aerated wort. Yeast start consuming the nutrients present in the wort and grows. In the meantime, yeast helps to produce alcohols and metabolites. This fermentation ends when the sugars present in the wort decrease up to predefined concentration level, as wort gets lighter more alcohol produces from sugars. The factors that need to regulate in the fermentation process are oxygen content, temperature, pitching rate. Some other factors that causes affect in the fermentation they are wort composition and yeast condition. The lack of consistency of the yeast and wort might be counterbalanced. In this way, when the reasonability of yeast is limited, distilleries could build up the pitching rate or somewhat hoist the oxygen levels and temperature. In traditional process continuous monitoring is not possible but we can monitor and control the brewery process by using technology with the help of AI and its programmed tools (Vassileva & Mileva, 2014).

4.3. Artificial neural networks in hot beverages

An electronic nose is an innovation for the examination of smell, it works on the concept of mammalian olfactory by utilizing gas sensors that can be used in wine and coffee industries for smelling (Thazin et al., 2018; Roy et al., 2019). Electronic nose is utilized in food industries and beverages to control the quality of the products. The efficiency of roasted coffee is identified by coffee cupping method, in accordance with the standard set by the Technical Standards Committee of the Speciality Coffee Association of America (SCAA) (Thazin et al., 2018). Initially 5 g of coffee beans were grinded into powder, further they steep in boiling water for 3–5 min at temperature of 93–97°C (Lingle & Menon, 2017). The data is collected by using cupping method which is forecasted by radial basis function ANN. Signals from the gas sensors present in e-nose are analysed by computer using LabVIEW program. Data from e-nose is analysed by multi variable data analysis (Thazin et al., 2018). This experiment procedure is divided into three parts one is the effect of temperature (10°C to 90°C) of liquid coffees (smell characteristics) and second is, categorize the acidity levels of various degrees of coffee roasting. At last, the bitterness levels are tested by humans and instructed with radial- based function ANN (Thazin et al., 2018). Similar to the e-nose, e-tongue is also utilized to check the quality of different types of beverages such as milk, coffee, tea, wine and beer etc. Parameters that are detected using this e-tongue are saltiness, sourness, and bitterness (Tan & Xu, 2020).

Quality of tea can be estimated by using e-tongue. Theaflavin (TF) and Thearubigin (TR) are flavouring components. Pu-erh tea tests of different assessment types and backgrounds have always been broken utilizing e-tongue assessment and different instrumental methods, including synthetic examination and electronic tongue (Goa et al., 2016). A pulse voltammetric e- tongue helps in the detection of type of tea according to its age. This can be acquired only by UV-VIS spectrophotometer-based equipment analysis (Roy et al., 2019). Coffee beans classification process is carried out in accordance with the standards, category, fault, quality characteristic and nature of the beverage produced (Pizzaia et al., 2018). The types of Arabic espresso are numbered from the grouping by type or imperfection, from two to eight. Investigating the quality, one should look at the factors that have an impact on the recognition of the market and increase the value of last item. The order cycle is assessed in the business appraisal, hence assessing the shape, size, shading, grain types and sort of refreshment, attributes which happen in the marketing cycle (Fig. 3). With the help of image analysis, we can predict the defects present in grains according to their size and shapes. It is easier to remove defected beans and maintain the quality of the product (Pizzaia et al., 2018).

1. Download : Download high-res image (500KB)
2. Download : Download full-size image

Fig. 3. Image processing algorithm for detection of defects.

5. Artificial intelligence in cutting and sorting of fruits and vegetables

Fruits and vegetables are major contributors to the nation, but its production is slowly decreased due to improper cultivation, lack of maintenance, losses during harvesting and post harvesting and increase in labour cost. The automation method is used to increase efficiency and to decrease losses. This automation requires CV, which is a part of AI, uses different softwares to classify vegetables and fruits based on quality and also identification of damages in them (Jasmeen et al., 2014; Makkar et al., 2018). X-ray imaging technology is one of the best non-destructive methods which help to detect diseases. For example mangoes are detected for quality by using X-ray imaging techniques and Artificial Immune Systems (AIS) which was developed by Ahmad et al. (2005). Another technique named Image processing technique was developed by Yimyam et al. (2005) which analyzes, detects and segments the mangoes physical properties such as colour, shape, size and surface area from images. Some other techniques such as Principal Component Analysis (PCA) and Linear Discriminate Analysis (LDA) states that mango yield at week 7 or 8 to get the best efficiency (Razak et al., 2012). Competitive Learning Neural Network classifier formulated methods using fuzzy image processing to analyze and calculate the rating of mangoes. The most important step is to detect and sort the mangoes based on an accuracy rate of >80%. These techniques are not only used for mangoes but also for other fruits (Jasmeen et al., 2014).

5.1. Potato sorting by AQS 602

Sorting potatoes by using AQS 602 machine sort’s potatoes 3-5 tons/hr (Nachev et al., 2012; Titova et al., 2015). Potatoes are fed through forming of stream through belt convener’s which consists of individually taking care and direction (Fig. 4). These sorting devices sort the potatoes based on tubers formation to assess the quality with the help of a PMC Photometric camera (D₁, D₂, D₃) (Nachev et al.,2012; Titova et al., 2015).

1. Download : Download high-res image (395KB)
2. Download : Download full-size image

Fig. 4. Steps involved in automatic potato sorting process.

6. Artificial intelligence in bakery

Automation in every industry is increasing at a rapid rate, and the baking industry is no exception (Ervin, 2017). Bakery products are in many shades, forms, and dimensions. In bakery production grading plays a very important role during handling and packaging many irregularities. Olden days the QA method where human visual inspection is usually used (Nashat & Abdullah, 2016). There are various processes that a bread should go through in the bakery industry. Usually, bread making requires introduction of living microorganisms, i.e., yeast is inoculated into dough. Any of the steps in making bread is not performed perfectly, then the bread quality is hampered (Utepbergenov et al., 2017). Proper arrangement of production of bakery products, economical use of resources in bread making and implementation of automatic control systems (ACS) are the priorities of the industry, which influences the quality of products and cost reduction, and profit growth. Introduction of an ACS enterprises bread making industry provides by increasing the growth of productivity and efficiency. Saving electricity and fuel by main technological lines in bread production is important. Improving the efficiency of resources, increase in the authentic information, in making bread products is another approach (Khorolskyi et al., 2018).

Robotic technology complex includes an automated control system. Rheological properties of sourdough and dough, lifting force on the dough, the active acidity of the sourdough, the acidity of the dough and the smell, forming the ability of a dough preparation, the duration of dough proofing, proofing temperature, humidity in the proofing case and the weight of dough is controlled by the sensors (Khorolskyi et al., 2018). The porosity of the bread, the acidity, dimensional stability, moisture, temperature of the soft part centre, the duration of baking the dough are indirectly controlled by the system of sensors and the visualization system (VS). Intellectual system to support managerial decision-making based on the information of the sensor blocks such as database, knowledge base, training base, output block, expert systems, changing the operating modes of ultrasonic systems. This is accomplished through the performing mechanisms by doing optimal management of impacts on heterogeneous technological environments. The raw material parameters are evaluated by the expert system of product quality control. This method also improves the properties of flour, sourdough, and dough for fortification properties of bread (Khorolskyi et al., 2018).

Overall, for the complete baking process starting from mixing and till packaging requires about 3 h. It usually takes one hour for the product quality, which means the quality data of bread cannot be directly put back to an oven, a mixer, or a device. A food quality sensor converts the food property responses into an electrical signals (Sharuda & Kyshenko, 2010). Based on the mode of operation, the sensors are online or offline. To check the quality of bread it is usually done by checking inner portion of the bread by using a camera after cutting. The size and dough porosity are determined to decide loaf quality. These devices assist to monitor the bakery products, depending on the bread range and quantity helps to analyse data in real time (Utepbergenov et al., 2017).

7. Artificial intelligence in restaurant

The use of AI robots has become well known worldwide and the implementation of this technology has spread from industrial manufacturing to service industry. Varieties of different types of robots have replaced. Humans are easily replaced by robots in guiding, reception, and delivery services. The restaurant’s robots can be used in many aspects, starting from washing, cutting of vegetables, sorting of the dishes, even for serving and cleaning of dishes (Yuqi & Young-Hwan, 2020). The robots used in restaurants have different functions, and characteristics. Most of the robots deployed at the back of the house are industrial robots. The robots at the front of the restaurant are service robots. The type of robot that is specified as waiter robot consists of a base and an automatic dumb waiter. This has three stages and has been replaced by human waiters. They also help by keeping the food within time which also keeps it clean and warm. At the arrival of the robot to the table, a drawbridge door opens and the tray that was supposed to be delivered is raised to the door opening and later the food is rolled out of the door and reaches the customer. Energy source that is used here consists of one pair of 24V lithium polymer (LiPo) batteries (Yuqi & Young-Hwan, 2020). For the access of the robots at the restaurants which consist of tables at a height, a specific kind of motor is connected between the base and dumbwaiter and is used for lifting it to that height. In ROS (Robot Operating System), a specified mathematical model and programming code is created and applied on the robot model as well as the original one. Control of the robotics is carried by visualized maps which show the complete restaurant table layout. This visualized map can be brought into action from a ground plan, or it can also be built by using SLAM (Simultaneous Localization and Mapping) algorithms. Using the Adaptive Monte-Carlo (AMCL) the robot model is located within the map (Omara & Sahari, 2015).

Locating a mobile robot is common environmental issue (Fox et al., 1998). So, AMCL uses likelihood techniques and particle filters which helps to track the location of robot from a map server. Depending on the size of environment and particle size the robot is in or the localization precision varies. Usually, particles with bigger size give a better analysis of the position of the robot (Zhang et al., 2009; Cheong et al., 2016). In another situation where the task is to be foremost customer to the specified table, the human waiter just clicks the “lead” option in the list of main menu and also enters the required table number, then that number continuously shows on the screen of the robot while moving also. So that, on the arrival of the customer, they just need to select OK on the screen and later the robot will return for standby at front desk (Malik et al., 2016) . Robot can function for recycling of the garbage collected from the plate and also gives self-delivery. Once the customer completes eating, they just press the bell placed beside the dining table for calling the robot. At the starting waiters come to the table for the cleaning function, once robot reaches its destination table, the customer takes the charge of placing the plates, containers, tableware etc. Then the customer needs to select “return to origin” (Yuqi & Young-Hwan, 2020).

Vast quantities of data are being generated in restaurants through packages that control everything from scheduling food delivery and shift staffing form taking reservations and inventory for paying bills. AI helps restaurants in cutting down the waste material, improving the supply chain and home delivery (Berezina et al., 2019; Antony & Sivraj, 2018). Chatbots provides automated and personalized client service to assist restaurants and collects payment and orders. Requiring no app transfer and restricted setup prices (Hoy, 2018). Digital menu boards are a lot easier than manually ever-changing costs and things. A chatbot is an online program capable of carrying human-like chat conversations. Chatbots work on ML and natural language processing (NLP) to understand the conversational data given and react to that in same manner as humans would be responding (Lasek & Jessa, 2013; Shawar & Atwell, 2005). Chatbots can answer frequently asked questions, helping customers with placing or tracking an order, processing payments. Chatbot learns with every interaction with customers and is exposed to more data. The main advantages of using chatbots at a restaurant is their ability to work with no downtime, sick, or vacation days and it is providing immediate assistance to the customers at any time (Berezina et al., 2019).

AI based biometric technologies are used to build detailed user profiles, enhance user recognition and security, grant or deny access, increase personalization, approve or deny payment transactions to minimize the trickery situations. Biometric systems are designed with four key components. The first is scanning and reading devices that usually record biometric identifiers like fingerprint, face, iris (Unar et al., 2014). Second component is software that converts the scanned biometric data into a digital code using a secure algorithm. Third is the component for comparing and finding matches. Lastly, the fourth component will secure database that stores the coded biometric data for further comparison (Berezina et al., 2019). Restaurants have started using some of the biometric technologies (e.g., facial and fingerprint identification). Fingerprint is a more convenient and reliable option. The fingerprint technology prevents employees from duplicity also for proper maintenance of the records of customers as well as employees. Facial recognition system is great for customers in restaurants, as people are usually identified through an image of their face (Berezina et al., 2019).

8. Artificial intelligence in 3 D printing

3D printing represents to a group of additive manufacturing process and has evolved in late 1980s from prototyping tools into an ecosystem of troublesome technologies. This technology varies a lot, but the two have common applications. Mainly used for building layers of the materials like metal powder which is melted using laser or with the help of ultraviolet light to liquid polymer. The complex shapes can be created by using this technology. It is possible to make digital image which helps to decentralise production and allows customisation of parts (Oscar et al.,2020). 3D printers are like robots. The main reason behind the manufacturing of more robotics is of easy development of electronics, easily available cloud computing superpower and also the usability of high-quality sensors. In earlier times, robots were very costly and were mainly used in heavy manufacturing industries like the automotive industry. Usually companies invest a minimum of $1,000,000 for adding that into a production field (Brian, 2015). 3D printers and robotics are mainly making a great change in the confectionery industry. These new developments have brought attention to companies for easy working and faster working (Brian , 2015; Godoi et al., 2016) .

A 3D printed object starts its process from a file present in CAD software environment. Every feature of the model in CAD is well described, changed, clarified and optimized in the 3D designing software (Matthew et al., 2019). The file is saved in the form of a .stl file once the digital model is ready and later it is transferred to the 3D printing software interface. 3D printing programs are open source and free to be used by others. While working it basically slices the object into layers and individually commands are generated. Many parameters like speed, temperature, height and thickness using robotics arms are to be set using 3D printing software. Later the final design and the features are exported that are suitable for the upload on the 3D printer (Lanaro et al., 2019).

An advanced production of 3D system with a 3D printing of culinary innovation centre has been opened to allow food industry chefs to discover ChefJetTM Pro professional food printer. Connection with Hershey found a prototype of a chocolate 3D printer CocoaJet which came in 2015 (Matthew et al., 2019). In Australia, a company named TM Retail Food Group uses chocolate 3D printing in their store for personalised cake messages in Michel’s Patisserie franchise with the intention to roll it out to the national market in 2018 (Ziegler & Hogg, 2019; Dankar et al., 2018). With some modification in commercial printers and successfully obtained a “food grade” by the Federal Agency for the Safety of the Food Chain (FASFC) (Matthew et al., 2019).

In designing the model, slicing and optimization the usage of software is very critical factor. The software used for slicing is an intermediary for planning and evaluation of the sections between 3D model and 3 D printer. Slicing software tool can help to change from digital into hypostatic model, and also transforms .stl files into g code file. For setting up the software it is quite easy, only a few features needed to be altered. For most of them features like nozzle temperature, speed of printing, thickness, temperature of platform are important factors to optimize the printing process with additional support designing and repairing (Guo et al., 2019).

9. Future challenges

Enormous changes in the food science makes the food based products as a supplement of many nutrients to protect from many diseases. AI is used to detect changes in water, role of fertilizers in crop yield with the help of cameras and drones. AI algorithms help to find potential threats and alert alarms. Elimination of food waste in production sector in any industry can be carried with the help of AI. It is used in food restaurants to scan the food to know nutritional value of food items. Development of methods for quantification of nano materials in food is one of an important aspect (Barlow, 2009). AI helps intelligent packaging to find food contact nanoscale form substances (Barlow, 2009). AI can clear the gap between manufacturers and transfer the prior information to the cloud to create a huge set of data (Xu et al., 2020). Achieving automatic orchard harvesting where fruits will grow in a usually unsuited environment, helps in labour saving and also optimizing the total yield (Misra et al., 2020). Fluctuating supply demand variations and narrow food hygiene and suitable targets can be achieved in future by using AI which is a major tool in supply chain management (Misra et al., 2020). Prediction of expiry date of food present inside the package with the help of sensors is another great challenge and humans can get benefited to know about the food spoilage and avoid food borne diseases (Misra et al., 2020). Further the development in the vendor applications is costly and almost exclusively being developed for larger forms. Expansion of more such applications under AI makes it easier to work with restaurant robots in the near future.

10. Conclusion

As the world population increases day by day there is a shift of people from rural areas to urban life which increases the demand for food and food-based products. Advances in science and technology find the solutions for supplementation of foods. To reach consumer expectations many innovations are developed. To fulfil the desires of consumer AI will help to optimize the factors to develop a healthy hygienic diet. Hence this review focused on the role of AI in various food sectors. Relevant literature on scope of robotics in food and beverages have been reviewed and discussed critically. Further intense research in advancing 3D printing helps to improve food business from manufacture to servicing has been discussed with future vision.