Machine Learning In The Warehouse

Machine Learning – Artificial Intelligence & Algorithms

One piece of technology with the potential to have an enormous impact on logistics and the warehouse is machine learning. Machine Learning is a branch of artificial intelligence that utilises computer algorithms to look for patterns in big datasets and use the insights to improve its performance or understanding of a particular problem over time. It’s a vast area of computer-science with many applications, but it’s already all around you. Virtual personal assistants (e.g. Siri, Alexa, etc.), facial recognition and self-driving technology are just a few high­profile examples of machine learning that is becoming embedded in our day-to­day lives.

It is, however, essential to note that Machine Learning is an umbrella term for a variety of sub-technologies that vary considerably in complexity, sophistication and accessibility. Of course, at one end of the spectrum, technology giants such as Tesla and Google are spending billions on building supercomputer powered, deep neural networks to create next-generation technologies. But at the other end, small businesses are using simple, DIY-models to improve operational decisions and efficiency. Machine Learning, like any form of Artificial Intelligence, does not have to come with a significant price-tag and can be used to combat a myriad of challenges.

Although currently under-utilised in the warehouse, machine learning’s potential impact in distribution centre design and management is virtually limitless. The language of logistics is data, and consequently, there are big-datasets available in most warehouse operations. The promise of machine learning is that it can take this information and not only make connections virtually impossible for humans to identify but actually to get better at doing so as time goes by and more data becomes available. These insights can then be fed back to either an IT-system or management to help improve operational performance.

Taking SEC Storage as an example, we have developed a machine learning platform that allows us to analyse our clients’ picking data and throughput information to optimise the selection of pick-faces for new and existing SKUs. Not reliant on fixed logic-based algorithms like traditional warehouse management systems, our program instead leverages historical and current data to assess new SKUs and make accurate predictions about the most efficient type and location of pick-face to utilise. In a recent application for a major high street retailer, this system helped increase pick-efficiency by 34% by ensuring that SKUs were located in the most optimally positioned, and sized pick-faces. Doing so resulted in both reduced travel distances and replenishment frequency, and better still, the system is still learning and improving as more and more data becomes available to it.

In reality, given the right data, machine learning is sufficiently flexible that it can tackle most warehouse management problems in virtually any warehouse operation. Other examples of how we are or will use machine learning in warehouses over the coming years are as follows:

Improving the quality of forecasting for planning and predictive purposes.

Identifying SKUs that should be clustered together due to previously unforeseen connections in buying habits.

More intelligent methods of classifying products, a particular example would be improving ABC banding analysis for fast, medium and slow¬moving products.
Training self-driving vehicles to safely navigate warehouse environments and work alongside human operatives.

Image recognition software trained to recognise and identify products based upon their physical characteristics. As examples, this could help ensure order-pick accuracy or allow a robot to determine the correct item on a shelf to pick.

Supporting augmented reality applications that provide layers of additional data and information to an operative’s standard view using special wearable eyewear.
Embedding speech recognition into processes to speed up and enhance information interchange and allow actions to be performed ‘hands-free’. Virtual personal assistants and Voice-WMS applications are common examples.

Machine learning is undoubtedly a sophisticated, cutting edge tool, and indeed, the underlying mathematics and programming can be challenging to engage with for the layperson. However, the common misconception that this technology will only provide a return on investment for large companies with deep pockets is something we should lay to rest. In truth, machine learning applications can often be developed relatively affordably since much of the underlying code required is now freely available to developers and can be highly efficient to produce. In many cases, having sufficient data is actually more problematic than the development of the code itself.

This widespread accessibility and the vast array of potential applications has led Forbes to predict that the machine learning industry will grow 1200% in seven years, as it becomes increasingly adopted by companies from all sectors, worldwide. And, in their logistics trend radar report, DHL predicts that machine learning will be the highest-impact technological trend over the next ten years, ranking it above robotics, self-driving vehicles, and cloud logistics just to name a few.