Page 18 - Food&Drink Magazine Jan-Feb 2019
P. 18

TECH FOCUS
✷ BY THE NUMBERS USING AI TO
INTERPRET DATA
AI systems can be based on any one of a range of different algorithms and mathematical models, which are the tools they use to interpret data. One prominent example is the use of artificial neural networks. These act as a framework for machine learning algorithms processing data inputs.
The operation of a neural network is best explained by example. Start with a data set which is manually tagged and can be used to train the machine learning algorithms. The data set might be 10,000 images of handwritten three-digit numbers that all differ from each other subtly or substantially. These are then fed into the neural network and checked against the manually tagged answers for accuracy. In this example network there are only two output nodes. The creator might be asking the network to identify whether, say, a number is even or odd. The machine-learning algorithms look for patterns in the input data nodes and push this through to create a hidden layer of nodes tracking the patterns.
Crucially, the network creator does not need to tell the algorithms what patterns to look for; they will automatically correct themselves by checking their performance against the manually tagged output data. This is particularly useful for complex systems, where it is not feasible to write a firm set of rules to govern the algorithms interpretation of the data. The neural network will also assign weightings to the various nodes at the input and hidden layers which will guide its determination of whether the number is odd or even. In our example the weightings should ultimately reflect the fact that the last digit in our three-digit numbers will determine whether the number is odd or even.
This is a relatively simple example. Anyone with access to a suitable data set and a YouTube tutorial could set it up. The more advanced forms of AI present a substantial opportunity for businesses, particularly those
in manufacturing, to improve their operations.
MOVUS into the future
A Brisbane company is helping companies such as beverage giant Asahi leverage data from legacy processing equipment with wireless sensors and AI tools. Michael Hughson reports.
THE world is increasingly awash with data, with no sign of it receding any time soon. One driver of the flood of data is the burgeoning ‘internet of things’ (IoT), which refers broadly to the range of machines and sensors that are connected to both each other and the internet. IoT devices enable us to gather data as never before, greatly expanding our ability to monitor and improve performance. However, the amount of data being thrown up can be somewhat overwhelming for the humans trying to process it, making it difficult to realise the supposed benefits. At this hurdle artificial intelligence (AI) steps into the picture.
The scope of AI is not strictly defined, but generally refers to capabilities of machines that would otherwise require human intelligence. At the low end of the complexity scale are tasks like recognising numbers or letters in an image (see box), while at the high end is the ability to manoeuvre a car through busy traffic. AI offers us a practical means of efficiently sifting through the data streaming out of IoT devices.
MOVING WITH THE TIMES
MOVUS, a Brisbane-based firm founded in 2015 is looking to help firms capitalise on the existing and growing potential ofAI.
“In the work I was doing in the corporate environment, it was clear there was a real
demand from a range of businesses for IoT and AI products,” the company's founder and CEO Brad Parsons says.
Given the growth of the company in just three years, it looks like he was right - the firm’s team of three staff at inception has expanded to around 30 today and looks set to continue to grow.
MOVUS straddles both IoT – through its wireless FitMachine sensors – and AI, which interprets the data generated by those same sensors (which are capable of monitoring equipment vibration, noise and temperature). As an added bonus they are easy to install, fixed onto the side of machines using magnets, which are
(at a regular interval regardless of condition) or ‘reactive’ (after machine failure) strategies. Maintenance shutdowns can thus be better planned to avoid costly production slowdowns or avoided outright by tweaking machine operation to prevent damage in the
first place.
FITMACHINE IN ACTION
Multinational beverage manufacturer Asahi recently deployed the company’s FitMachine sensors in one of its Queensland bottling plants. The objective was to improve asset efficiency, reduce production downtime (both planned and unplanned) and achieve maintenance efficiencies. Within two months FitMachine sensors detected
“ In the work I was doing in the corporate environment, it was clear there was a real demand from a range of businesses for IoT and AI products.”
carefully set up so as not to disturb the operation of the machine itself. MOVUS’ machine learning algorithms can then parse the sensor data to monitor the health of a machine and predict when it might require maintenance or be about to fail.
This enables firms to pursue a ‘conditional’ maintenance strategy (servicing machines only when monitoring indicates it is required) rather than traditional ‘preventative’
abnormal vibrations within one of the fillers on the production line and flagged the issue for staff at the plant to investigate. The staff found a fault in the drive motor bearings for the filler, which if left unchecked would have led to a breakdown in the filler and a significant interruption in production at one of the busiest times of the year for the plant.
This is a clear example of how MOVUS’ sensors and AI analytics engine can be used to
18 | Food&Drink business | January-February 2019 | www.foodanddrinkbusiness.com.au


































































































   16   17   18   19   20