Of all the manual tasks that are performed in a warehouse, the flagging and management of damaged boxes must be among the most tedious. And when the company in question is Intel, the huge volume of boxes to be individually inspected presents an especially daunting challenge. 

Intel’s warehouse receives and stores a steady stream of boxes containing raw materials, components and finished goods. Each must be manually inspected for damage by the facility’s operations team. When a damaged box is detected, it gets held up for another level of validation by the inventory quality control team. And when IQC confirms the damage, the errant container moves on for further validation by a factory materials engineer, who, depending on availability, might take up to two months before assessing its condition and deciding on the proper action to be taken. Options include sending it for retesting, scrapping the contents or repackaging in a new box. If both box and contents are damaged, Intel will then file a claim. 

That lengthy and arduous process was resulting in significant “inconsistency and uncertainty,” Intel said in its submission to the 2023 Supply Chain Innovation Award competition. Human inspectors were tending to reject or place holds on an excessive number of boxes that were showing some degree of external physical damage, even when the contents might be perfectly intact. The result was unacceptably high costs for testing, materials and shipping. 

The numbers add up fast. In 2022, Intel’s warehouse in Malaysia received more than 30,000 inbound boxes of raw materials, and filed for more than $5 million in damage claims during that period. What’s more, Intel said, the actual cost of manual inbound box inspection at the facility was probably even higher than estimated. 

In 2022, a core team launched a pilot to test computer vision as an alternative to human inspection of damaged boxes. It included members of Intel’s Customer Fulfillment, Planning and Logistics Group, as well as the company’s Supply Chain Innovation Lab. The hope was that deploying computer vision in conjunction with artificial intelligence and machine learning would provide Intel’s Global Logistics Organization with a more efficient and accurate means of inbound inspection. There would continue, of course, to be a number of boxes held or rejected for damage, but in the end, fewer good units would be inspected, held and returned. 

Intel was no stranger to computer vision, which it had been using for years in microprocessor manufacturing. The trick lay in melding the technology with an AI and ML control system. 

The team decided at the outset to create a solution internally, rather than buy a system off the shelf. Early design was divided into two parts: cameras and other vision devices to capture multiple images of each box, and AI and ML models to compare them with a repository of “good” and “bad” box images, to make a recommendation on whether to hold up the box in question. 

Computer vision eliminates the need for all those multiple levels of manual validation, Intel said. If the inner box or its contents are determined to be damaged, the process of filing a claim can be fast-tracked from months to a matter of minutes. 

From the start, the pilot team established multiple objectives: 

• Eliminate double buying of components, based on the assumption that a certain percentage would be rejected; 

• Quickly assess outer-box damage, within a second or less per inspection; 

• Reduce costs associated with inspection, materials and shipping; 

• Increase flexibility, possibly by extending computer vision to other items or external inspections, and 

• Ensure a low-cost setup in the pilot phase. 

The system they came up included a pilot workstation equipped with four Logitech cameras, capturing all sides of the box, and connected to a computer. The raw images are stored and used to produce “inferred” results, which are displayed in the user interface. 

According to Intel, the initiative met its objectives, keeping setup costs low while yielding cost savings of $4 million in the first year. Inspection and disposition of boxes was a matter of milliseconds. 

Intel plans further enhancements to boost system accuracy and apply it to additional commodities across warehouses, cross-docking stations and manufacturing sites in multiple countries. There’s also the possibility of using computer vision to read labels and inspect other types of containers. 

“The computer vision pilot has shown tremendous improvement over the previous manual inspection process,” Intel said, “and the team continues to examine ways to further improve system performance and reliability.” 

Resource Link:

Intel, https://www.intel.com/content/www/us/en/homepage.html