Warehouse automation systems have evolved with new generations of mobile robots and multi-directional shuttles that offer a high degree of operational flexibility. New technologies demand advanced application expertise. We hear industry jargon such as “Artificial Intelligence” and “Machine Learning” used frequently. Is it marketing hype? What is it and why do we need it?
Gone are the days of calculating automated systems’ performance based on fixed paths, speeds, and distances. Not long ago, automated warehousing and distribution systems often employed equipment and system designs that were constrained by fixed paths and routes for product flow. For example, stacker cranes and shuttle systems that drove in aisles with no competing traffic and conveyor systems with predictable speeds and paths. Simple calculations using speeds, accelerations, and distance could calculate performance of individual subsystems with some certainty. Note: there is a real beauty in predictability… those technologies absolutely have their place!
In many cases, system performance issues related to order sequencing were compensated for by employing accumulation and sequencing systems. These accumulation/sequencing systems had knock-on effects of increasing travel distances for product through the system, causing potential for errors, product damage, and high system costs due to space required and cost of equipment. Increasing system performance used to mean drive faster, convey faster, sort faster, release faster. Faster, faster, faster…
Enter the new age of flexibility. New equipment on the market such as mobile robots and multi-directional shuttles are autonomous, independent, and with many possible routes between origin and destination. That flexibility means there are endless combinations of ways to complete a task. Picking the shortest or closest path may not be the quickest and increasing equipment speed doesn’t necessarily result in greater output.
Performance of individual pieces of equipment or subsystems is not indicative of overall system performance. While “rules of thumb” are valid for a gut check, they aren’t adequate when it comes to justifying large investments and trusting the system will deliver when needed most.
Enter Artificial Intelligence and Machine Learning
In a nutshell, Artificial Intelligence is: “developing computer systems that are able to perform tasks normally requiring human intelligence such as: problem solving, planning, ability to manipulate and move objects”. Artificial Intelligence is best used when describing equipment level capability. But cool equipment isn’t the end game; improved system performance and operational excellence is the is the goal.
While Artificial Intelligence is important, it is a generic term when used in the context of warehouse automation equipment and software. It falls in the category of “industry jargon” such as “Industry 4.0” and “Internet of Things”. Broad terms that make good marketing slogans, but don’t really tell you much. Machine Learning is where it’s at!
Machine Learning is “learning with supervision using observation and experience” and comes in a few different flavors: supervised (known input, algorithm, predictable output), unsupervised (known input, algorithm, no predictable output), and reinforcement.
Reinforcement learning is used to design advanced automation systems. Reinforcement learning uses trial and error where many iterations of the design are tested in the model to discover the optimal design solution. Reinforcement learning requires data, algorithms/rules, and a model (simulation). Collect the data, input the algorithms/rules, prepare the model, try it out, collect feedback. Adjust, and try again.
There are several factors at play: an experienced Expert system designer, the environment (inputs, outputs, system layout, etc.), and the capability of the equipment. The Expert system designer relies on their experience to adjust the environment (design) using constraints of equipment to arrive at the best possible system outcome.
It would seem to be common sense that implementing advanced technologies requires the pre-requisite of advanced system design, but unfortunately many complex systems are still implemented based on “equipment capability” and “rules-of-thumb”. In the excitement to install the latest in robotic technology, we breeze past the hard work of ensuring the system will operate as intended and meet the business needs.
Machine Learning is the first step in designing and implementing an automated system. Analysis and simulation will determine the rules and algorithms by which each system should be controlled that results in the desired performance. Those rules and algorithms get applied to the control software systems specific to each system design.
After go-live, reinforcement learning can be used to maintain peak efficiency, even in the face of changing order profiles, SKU mixes, seasonality, or other unforeseen changes in the business. These flexible systems are more adaptable to change than fixed-path systems, and the tool to leverage that flexibility is continuous machine learning.
Are you planning to build an advanced warehouse automation system? If so, there are several steps leading to a successful outcome: data analytics, simulation model, and an experienced Expert system designer. Don’t start your project without them!
This article was originally published on www.esotericstaffing.com. Esoteric matches specialized robotics, engineering, and supply chain talent with the needs of the Fourth Industrial Revolution.
