In the United States alone, container volume in ports is soaring, and even in a backlog. Container imports to the port of Savannah, for example, increased 32% from 2016 through 2020, according to the Georgia Ports Authority. Warehouses contending with the back-log of containers within the global supply chain, as well as labor shortages and growing e-commerce demand are buying and leasing more real estate than ever, and investing in automation technology and AI systems to keep up.

As more and more warehouses turn to ADLINK for guidance in developing and deploying these industrial AI systems, we wanted to share where we’re seeing warehouses gain the most value and quickest returns on investment when it comes to automated technologies for the warehouse.

Automation in the Warehouse

We’ve been talking about Industry 4.0 and automation within manufacturing since about 2011, and within the materials handling space since before 2016. So, how are we doing? We are further along in manufacturing than in the warehouse. In a 2019 Modern Materials Handling survey, over 50% of warehouse distribution center managers said they still use mostly manual processes.

Over the last few years my team and I have visited 100’s of warehouses and not one warehouse is fully automated. Often we see a seemingly simple, yet very common challenge when trying to adopt an emerging technology like AI: where is AI going to yield the biggest return? With any AI project, it’s important to focus on a business outcome and use case. While this seems obvious, we find that often this initial step can be challenging for our customers to figure out. This step is the most critical, otherwise, the project is likely to fail.

We’re finding that machine vision is one of the easier technologies to implement and leverage AI, as well as one to see the quickest returns within warehouse fulfillment because of the flexibility. AI machine vision will meet you wherever you are in the automation journey (see figure 1 below), building upon infrastructure and processes (manual, automated and autonomous) you already have.

How Does AI Machine Vision Work?

Machine vision is image-based analytics for warehouse applications like inspection, process control, safety and robot guidance; giving human vision to machines such as conveyors, robots or a computer just logging what it sees.

There are two different levels of machine vision, rules-based and AI-based.

Rules based machine vision – standard machine vision has the ability to detect when something is wrong but cannot tell us what exactly is wrong (classification), nor direct a system to take an action once information is received. Examples of rules-based machine vision include reading text, a barcode, identifying fill levels, or reading gauges.

AI based machine vision – AI machine vision identifies things seen before without a unique identifier. AI machine vision classifies what it sees, becomes smarter over time, and can also create automation workflows to take action.

So how does the AI machine vision system work?

1. Gather Images

If you took 200 pictures of your face and gave them to me I could build a facial recognition model, and it would be pretty accurate. But if the camera used is pointed at a certain angle, capturing images at night, with grainy image resolution, like an ATM camera for example, the system is going to struggle. So gathering images actually has to be the first part of the process.

2. Mark Objects

This is actually a manual process. In order to train an AI system, you have to help it. You have to tell the system where you to start looking, and what to look for.

3. Training

Training is an automated process, it runs in a neural net. The system takes the images that are marked and it distorts them. It stretches them, it flips them, it changes the colors, it blurs them, and it turns the one image captured into 10,000 images. And the more images it consumes, the smarter the system gets.

4. Test

And then the system creates a model. The model is essentially a set of algorithms that look at new images coming in, that then communicate what is sees providing a confidence level.

5. Act

After the model is created, we must do something with it. So what, if the machine sees someone standing in an unsafe location? Action must occur in an automated way. Let’s take a look at some examples.

AI Machine Vision for Warehouse Fulfillment

The more efficient a warehouse is among the primary areas of fulfillment – receiving, storing, picking, packing and shipping – the higher the quality assurance, on-time deliveries and happy customers, along with lessened pressure, decreased costs and employee burnout.

These examples use both, rules-based and AI-based machine vision.

Receiving and Storing – Machine vision with AI automates and improves the traceability of all assets throughout their lifecycle. Machine vision and beaconing equipped forklifts, for example, remove the human error during the receiving and storing process. Automating the scanning and data entry required for inventory management allows forklift drivers and receiving staff to make speed and safety their first priority.

Picking – Whether this is used for assembly or order fulfillment, robots, equipped with machine vision and AI accurately pick parts and place them in totes or boxes. AI allows users to take this a step further, by identifying the best “pick point” and placement orientation speeding up the packaging process and maximizing package volume.

Packing – Machine vision with AI audits each order for correctness. After the order is picked, these solutions automatically validate the order against the contents. This can support and/or replace a quality assurance technician allowing them to be reallocated to other tasks throughout the warehouse. Orders marked as incomplete/incorrect can be marked for re-packing prior to shipping.

Shipping – Machine vision and AI continue to audit each order for correctness during palletization. All packages placed on a pallet are marked based on unique identifiers, shape and size. Validating pallets for order correctness and fulfillment traceability can net massive gains in profitability. The industry is plagued with a human error during the palletization process, and high-mix orders cannot currently benefit from the cost/capabilities of robotics. ADLINK’s Smart Pallet solution, for instance, is a turnkey solution for semi-automated palletization leveraging AI machine vision.

Safety and Security – Machine vision and AI are used to determine if and when employees should be allowed entry, have the proper personal protective equipment (PPE) or in unauthorized areas. This same technology can be used to identify slick spots, trip spots, and/or pinch points.

Is AI Machine Vision the Solution for Me?

An easy question to ask yourself to determine if machine vision will work for a certain process is: if I film this process, am I able to see, deduce, infer, read whatever I want a machine to see, deduce, infer and read? If the answer is yes, then AI machine vision can be done. If you can see it with your own two eyes, then we can train a system to do it.

We can help you with a workshop and a roadmap to help you build a digital experiment pilot in as little as 2 weeks. Not every use case requires AI based machine vision and we can help you determine what process would benefit from AI machine vision the most and where it may be best to start. When implementing AI machine vision technology, we recommend the following:

Make lots of small bets, not one big experiment

For automation investments to go smoothly and quickly (not lasting years), look at tackling processes first, in an incremental way. For example, can you imagine how costly, lengthy and painful managing a “traceability study for an entire product lifecycle from receiving to shipping” would be as a first AI project?

Instead, make several small bets. Start by automating things that make human tasks faster, easier, more efficient, such as palletization, packing, or fork lifts for inventory accuracy. And note those technologies can work with each other as you scale automation technology horizontally across your business.

Move intelligence into or near things

Whether embedded, distributed or intelligent computing, edge computing enables AI to generate, consume, compute and distribute data in real-time to make decisions and take action when it matters most.

For instance, a machine vision AI system designed for robotic quality inspection may need to inspect a product for defects in less than .05 a second before the conveyor moves another product under its lens. The margin for error here becomes quite slim, requiring the movement of data to happen instantly. Edge computing is where data meets people, places and machines so this type of instant, real-time action occurs. 

This is a team sport, engage your ecosystem and partners

We don’t know the answers to everything and no vendor should ever tell you they do. And if you’re working with anyone who refuses to work with another vendor, another OEM, another partner you already have because they believe it’s competing technology, you may want to look elsewhere.

Learn more about ADLINK’s AI machine vision edge hardware here.

Learn more about our ADLINK AI machine vision edge software here.