Harnessing the potential power of AI across industry
pic: Shutterstock
With AI making global headlines in the past 18 months, industry consultant Graham Godfrey offers his view on its notable potential right across industry, including within the confectionery sector
AI is a hot topic at present as the media (and authorities) become aware of its potential and dangers. There is a huge amount of ill informed (even hysterical) comment as to its dangers and some positive speculation, largely around its application to “driverless” cars and facial recognition.
Facial recognition technology, with the need to identify and analyse patterns and specific characteristics is potentially the most useful aspect of the technology which might be adapted for the confectionery industry
While AI is probably – at least at this stage – a technology for adoption by larger companies with scale manufacturing operations, in the longer-term specialist technology suppliers may be able to offer AI supported solutions to smaller businesses through “embedded” AI in specific equipment
The greatest potential benefits for the confectionery industry would appear to be in the technology’s ability to handle, process and continuously learn from large amounts of diffuse data. While neural networks have had the potential to achieve some of this, historically they have been severely limited by available computing power. It is to a large extent this increase in computing power which has led to the growth of AI.
Potential usage
The confectionery industry has a number of difficult areas where it may have real potential benefits, which might include: Collation of data from different sensors and rapid corrective reaction to maintain product quality or plant operation.
AI Based process and operations control has great potential in quality control and reducing yield loss and optimising cleaning schedules. Other areas include effective pattern recognition which could evaluate “in process” quality such as physical appearance in panned goods, degree of and bubble size in aeration.
Moreover, it could also evaluate operating conditions by alternative means where accurate measurement is difficult or conditions are changing rapidly. It could also predict response to changes in ingredient characteristics which may affect process operations, cost or quality.
Other evaluation tasks include monitoring of rejection and scrap levels, collation of market response to product changes or developments, and evaluating new products against consumer preferences.
It should not be forgotten, however, that AI can only be beneficial when properly trained using large scale inputs of relevant data and relationships. This learning process – which can continue in practical operation – will be expensive and time consuming and will require expertise outside that which may be available internally.
Also, always remember “garbage in, garbage out” – ensure that the training data you use is actually valid and unbiased. Defining and setting up a beneficial AI system is likely require investment in external skills and complex data processing systems.
Sensor Collation and Innovation
In confectionery manufacture many key flavour and colour characteristics arise from the conditions in a process which may range from a few fractions of a second to many hours.
Those critical conditions may have major issues in terms of measurement, access, etc. or may be related to the complex interactions of several conditions at different phases of a process.
AI has the potential to move beyond conventional measurement techniques and use inductive approaches – similar perhaps but more reliable to those used historically. The way in which a high solids paste moves in a mixer under vacuum during evaporation is an example which springs to my mind of a situation where conventional measurement is extremely difficult but the ability to monitor and optimise the process would be invaluable.
It may also be possible to significantly optimise chocolate refining and conching operations by using visual data combined with temperature, power, vapour analysis and other parameters which would be difficult to combine conventionally. The characteristics of many of our ingredients can change subtly depending on many factors such as source, growing conditions or harvesting practice.
Though these variations are often (but not always) reasonably well understood their effect on product quality, manufacturing efficiency and consumer acceptability may be less easy to correlate.
An AI system with sufficient training and data input system could potentially predict and correct for these variations and may even be able to facilitate the use of a wider range of ingredient sources or characteristic – potentially invaluable as climate change affects the conditions in key ingredient source areas.
Furthermore, the physical appearance of a finished product is important. Historically this has been largely achieved through the human eye and intervention, but large-scale manufacturing with wide, high speed lines has made this less effective or affordable
Modern high speed imaging systems can identify and reject unsatisfactory product but AI can potentially relate incidences back to causes and make rapid corrections to process operations. An example might be analysing the surface of a moulded or enrobed product to evaluate and correct tempering conditions upstream.
Confectionery products are generally the result of multi-stage processes and one or more of these can fault to cause product or quality problems. AI, because it can handle multiple data streams and interpret them is potentially a strong tool for quality control and even consumer complaints response and resolution
Changes in manufacturing processes, ingredients, packaging etc. can all trigger a response in the market place. Product and manufacturing development can produce changes in products (particularly flavour and texture) for which the exact cause are often difficult to identify because of the complexity of flavour chemistry and textural development.
While analytical techniques can identify the many flavour identities and precursors which may be present in products (frequently hundreds), the importance of each individual identity, its relationship to others and the key parameters involved in its production are often extremely difficult to understand. Analytical techniques and multi-dimensional “spider” charts can present data but are difficult to interpret directly.
Similarly, product textures (particularly that of pastes) are frequently central to a product’s acceptability or to its manufacturing process. These are some of the most difficult things to characterise, they are very difficult to measure analytically – for example the texture of a paste as it is squeezed by the hand of an operator will often allow him to make small instinctive changes to a process, but reproducing that “measurement” by an instrument may be close to impossible.
The sector is very much at the beginning of the process of recognising the potential of AI development, not least because while many of our processes are fairly well understood, the effect of small variations in process conditions and changes in the exact characteristics of natural materials are much more difficult to evaluate and respond to.
