Getting products to customers at accelerated rates is one key to Amazon’s booming success. But the company found that relying on a mix of system integrators to deploy cell/area zone equipment in its growing array of fulfillment centers caused a range of integration issues with its enterprise systems. This delayed commissioning and, in turn, order fulfillment.
Ever dissatisfied with delay, Amazon worked with its suppliers to develop a modular, standardized fulfillment center design that could get these systems up and running more quickly. It turned to Rockwell Automation to provide a scalable, cut-and-paste, information-enabled automation architecture that would also streamline integration with Amazon’s own cloud services.
The cooperation has since created standardized control templates that have reduced commissioning time by up to 95%, and helped speed project delivery. The new solution is being rolled out in new Amazon facilities in North America and Europe.
Bart Schouwenaars-Harms of Amazon Warehouse Services played a key role in launching the modular designs and information-enabled control system. He provided an overview of how the company leverages Rockwell Automation solutions at this week's TechED event in San Diego.
"We needed to come up with an architecture that works for us and requires little effort when launching a new fulfillment center," said Schouwenaars-Harms. "That's where Rockwell Automation's equipment comes in for us. The focus was on the physical side—the hardware. How do we use the hardware, and how do we interface with our higher level software?"
The historical designs of a fulfillment center, or FC for short, included a top-level warehouse management system (WMS) running on rack-mounted server hardware at a regional data center. At the local level, warehouse control services (WCS) ran as a software layer controlling the equipment. At the bottom were the machines, sensors and actuators.
"The facility and machines were a rigid, monolithic design,” explained Schouwenaars-Harms. “The FC would typically have tens of PLCs, each one controlling a large area, which gave us a single point of failure. If one PLC went down, we lost operation of a large area of the FC."
“We moved to an information-enabled design to get more information out of the machines,” continued Schouwenaars-Harms. “We started at the bottom with a modular machine design, enabling copy-and-paste actions," he said. "We went to our suppliers and said we want you to build one of these machines very well and then copy-and-paste it. We didn't want every project to be custom or a new design, we wanted something that worked on a worldwide level for a single machine."
Amazon looked at using more cloud-based services, but its WCS remains on site due to latency issues. "We chose to use local servers and Greengrass," said Schouwenaars-Harms. Greengrass is the software layer that allows one to deploy cloud technology at the edge. The former WMS migrated to the Amazon Web Services (AWS) cloud.
Modular machine design
"From a system integrator’s point of view, we stopped being a user, and in 2016, we became a development partner," said Schouwenaars-Harms. "We worked with the suppliers of the custom, monolithic designs and provided input in how the new modular-equipment facilities were to be designed."
One of the requirements was for its integrators to follow international standards. "We use ANSI ISA-S88 batch management," said Schouwenaars-Harms. "We use batch control even though we are not really a batch company. We consider each box, every shipment as a batch with a size of one."
Depending on the facility, there are varying numbers of packaging lines, but the overall design remains the same, only varying in scale. Amazon resists custom machine designs and instead fits what it can into the site or area while still adhering to common designs. There are common, cut-and-paste designs for different types of process cells such as a pack labeler, and units such as glue systems or equipment modules such as a conveyor diverter.
Amazon also identified the critical paths. "If we have seven of something, we work with the integrator to get one operational," said Schouwenaars-Harms. "Once the first one is tested and accepted, training can start, duplicates can be made and the equipment can start being used. Our go-live point has moved from a single big facility-level event to individual legs or bits of a line.”
Industry 4.0 at work
"Industry 4.0 is a big play for everybody, and it means different things to different companies," said Schouwenaars-Harms. "At Amazon, we wanted to get more information and wanted our WMS running in the cloud and to have more control over our machines. That doesn't mean we want to fully control the machine from the cloud. But we did want to mimic all the pushbuttons, knobs and adjustments used by our associates and control it through higher-level software systems."
The move to a modular, information-driven architecture had only a small cost impact. The design moved from tens of PLCs to hundreds. However, the cost impact was only about a 1% increase, while the modular approach resulted in quicker commissioning.
Energy usage, overall equipment effectiveness (OEE), and alarming data are important inputs for Amazon as well. At first, the inclusion of power monitoring on all of the centers panels was difficult to justify. "But there are definitely benefits,” said Schouwenaars-Harms. “If we can see the power consumption on a machine, we can relate that to the number of parcels on a machine or their weight. A spike or drop in power consumption is an indicator that something may be wrong, and the machine needs to stop for maintenance."
“We use to have grease monkeys who carefully maintained the equipment, but not a lot of manufacturing facilities have those guys anymore," he said. "We need to cover that function with intelligent sensors—vibration is the big one—but motor current, speed and energy consumption are becoming more important. With less people in each facility, the machine needs to tell you it needs maintenance."
Amazon sees a future with Rockwell Automation, and that future is local compute and intelligence at the edge. "Intelligence is moving from servers to the machines themselves. For us, the compute modules that can be installed in a ControlLogix or CompactLogix slot are a great product. If you move your machine, the intelligence goes with it."