Manufacturing technology advances have extended the realm of automation beyond plant floor sensors, controllers, vision systems, and robotics into closely connected data collection and analysis using artificial intelligence. These technologies enable a “smart factory” to self-optimize and adapt to conditions in real-time. Despite these advances, warehouse and related material moving operations tend not to be nearly as modernized as plant floor operations.
“If you look inside the most modern environment, warehouse or factory, material handling, broadly speaking, is mostly analog,” says Matthew Rendall, CEO of Otto Motors, a maker of autonomous mobile robots (AMRs). “Any place where a forklift truck is driving something around, it is highly analog. That means the amount of data you have at your fingertips to analyze is limited. For decades, operators have been grasping at low accuracy, low frequency, and expensive-to-capture data trying to figure out how to run a continuous improvement program.”
For example, you can go into a factory or a warehouse today and still see industrial engineers sitting in lawn chairs at an intersection in the plant with a clip board, pencil, and stopwatch to monitor material flows, says Rendall. “It is an expensive thing to request of a highly trained industrial engineer, so it doesn’t get done frequently,” he says.
Which means the process, by its nature, is not as exact as it should be; plus, it’s rarely updated.
But this antiquated, analog surveying method is shifting in response to the decreasing cost of computer storage, increasing compute power, and new tools that target warehouse and distribution operations.
The AMRs from Otto Motors are designed for material handling in the manufacturing and warehouse operations in a free-movement manner. In an automotive warehouse, for example, there could be 30,000 parts moving across the facility. “It’s a symphony of motion,” Rendall says. “[With data from the AMRs] we are able to harness insights about how those 30,000 parts are moving in order to create a smarter traffic network.”
The foundation of Otto Motors’ AMR is a proprietary version of the simultaneous localization and mapping (SLAM) algorithm, which relies upon cameras and laser scanners to develop a photorealistic floor plan.
“Think of an AMR as having a photographic memory. As it’s driving around it is constructing a picture-perfect representation of that environment in its memory,” Rendall says. Using that underlying localization and mapping, you can plot out where a vehicle has traveled. “As a result, the side benefit [of an AMR] is that we have one of the most sophisticated data collection machines a factory or warehouse has ever seen, patrolling the floor 24/7,” Rendall says.
The AMRs operate with fleet management software which includes a feature called “factory replay,” providing a time-lapse recap of the entire day’s production. The software can take a 16-hour operation and collapse it down to five minutes, which provides an industrial engineer with a birds-eye view of the floor and the ability to rewind, fast-forward, and zoom in to a certain time during an incident—as well as before and after—to extract insights about ultimate root causes of problems on the floor.
Another company, 634AI, offers an AI-driven technology called Maestro, which is designed to safely orchestrate the movement of AMRs and coordinate situational reactions to create a more efficient and safer environment. Using off -the- shelf, Power-over-Ethernet ceiling-mounted cameras, and proprietary computer vision technology, Maestro pulls video streams from different cameras to create a grid on the floor.
Those videos are stitched together to create a real-time map of the facility. Then, deep learning AI coupled with computer vision algorithms draw semantic analytics from data on the factory floor. “This data can be productivity-related information providing real-time safety alerts, near-miss analyses, task allocation, and the ability to instigate and navigate a robot’s autonomous capabilities live on the factory floor,” says Shlomi Hatan, 634AI’s vice president of business development and operations.
Every factor in the process is identified, classified, tracked, and managed by Maestro, including raw materials, mobile robots, forklifts, boxes, and even people. “Designed to be hardware-agnostic, Maestro is interoperable with other systems to enhance company workflows on a universal scale,” Hatan explains. “The general rule is, if Maestro sees it, it can be tracked and controlled in real time.”
Hatan adds that Maestro’s control capabilities can be extended to deliver custom, AI-generated productivity propositions. For example, Maestro can display forklift slow zones to locate bottlenecks and interferences for vehicles and workers, informing them of obstacles to help shorten travel times. It can also track the traveled distance and driving hours of forklifts and robots to identify underutilized resources or assure timely maintenance of equipment. It can even alert operators of incorrectly positioned materials and pallets.
Collecting all this information internally and creating a map of material movement can also help companies with supply chain struggles. “There’s an interesting relationship between transparency and automation,” Rendall says. “We can’t influence what time the parts arrive on the loading dock. But if we are responsible for all the material handling that happens once the materials hit the receiving dock, we can use a QR code or RFID tag to scan the inventory. That inventory is only touched by a machine, and when that’s the case, you should be able to see within inch- and second-level accuracy where every nut, bolt, and screw is inside the operation. So you have a much better handle of what inventory you are working with inside your operation to more intelligently use the resources available to you.”
Beckhoff Automation, for example, offers a shuttle control system for AS/RS applications in compact form factors. This system includes TwinCAT Machine Learning software to reduce energy consumption while optimizing acceleration and deceleration of the shuttles. “The machine learning (ML) functionality automates this so there is no human intervention required to achieve the process improvements,” says Doug Schuchart, Beckhoff’s global material handling and intralogistics manager.
In addition, the EtherCAT communication protocol used in this system allows information to be collected in large quantities and stored in a database on a Beckhoff controller. That data can then be transmitted to the cloud.
“Once at the enterprise level or the cloud, data science software can be used to develop ML inferences for equipment optimizations and predictive maintenance applications. These ML inferences can then be deployed using TwinCAT Machine Learning to be executed in the PLC in real time,” Schuchart explains.
Dealing with dock delays
AMRs coupled with AI can increase productivity
and safety inside operations. And machine
learning coupled with PLCs can improve order
fulfillment and throughput. But what happens
when the truck doesn’t show up at the receiving
dock? Traditional warehouse management systems
(WMS) don’t have the ability to recalculate
everything when there are shipping delays, or if
there are too many shipments and not enough
docks, or if inventory is in the wrong building.
To address such issues, AutoScheduler.AI has developed a cloud-based intelligent warehouse orchestration platform that integrates with existing WMS, ERP, and even yard management systems to provide dynamic dock scheduling, proactive cross-docking, and prescriptive analytics that balance inventory flow and drive labor efficiencies. “Our goal is to be the brain of the warehouse,” says AutoScheduler.AI CEO and co-founder Keith Moore.
The work started as a project with Procter &Gamble (P&G). P&G operates one main plant in Ohio that uses seven nearby satellite warehouses for storage. Across this campus, the project noted more than 250 outbound full-vehicle shipments per day and, of those, 85% were drop and hook, and 15% were live load. These operations require manual efforts that are completed based on who is scheduled and the need for the day. But volatility in the production schedule and volume made it difficult to plan, resulting in an inefficient operation.
Using the AutoScheduler.AI technology, P&G doubled shipments from the plant directly to customers without increasing inventory, reduced shuttle moves involving outside warehouses by nearly 50%, reduced workforce planning from eight hours to 20 minutes per day.
That experience prompted Moore to turn this into a scalable system for any warehouse or distribution center operation to provide analytics that describe what is happening, predict what will happen, and prescribe an optimal plan based on that information.
Every few minutes, AutoScheduler looks at the current situation and then runs “what-if” scenarios based on inventory and constraints to maximize flow through the building. “We pull information from the WMS and other systems, run the optimization, and push it back into the execution system. So when someone scans a pallet…our plan flows straight through to the floor, and nobody even knows we exist,” says Moore.
The AutoScheduler technology is exactly the kind of cognitive toolsets that Stephen Laaper, a principal and manufacturing strategy leader at Deloitte Consulting, says is coming to day-to-day warehouse operations.
“Supply chains will continue to be pressed and stretched in various ways for the foreseeable future,” Laaper observes, noting that more actionable information is imperative. “Because of that, the nature of these solutions are becoming increasingly important.”
