What are the types of asset inventory management systems?
RFID Knowledge

What are the types of asset inventory management systems?

PublishedShanghai Zejie Smart Technology Co., Ltd.

The writing and pen era is not far away: the manual and bar-coding trophies system

Late last year, an electronic components distributor asked us about its year-end count. The warehouse held nearly 80,000 SKUs (stock keeping units, meaning product types) and relied on Excel and manual counting. Seven people spent two full weeks on each count, and errors often exceeded 3%. A scanning-enabled system cut the count to three days and discrepancies below 0.5%. This showed us how much identification methods affect counting speed. By method, asset inventory systems fall into roughly five groups: barcodes and QR codes, RFID, AI vision, positioning, and combinations of multiple methods.

Barcode and QR code systems are the most widely used. Put a printed barcode or QR code label on each asset, then scan it with a scanner or phone camera during the count. The asset ID is recorded automatically. Hardware costs are very low: a basic scanner costs tens of yuan, each label costs a few fen, and an existing office printer will do. The drawback is that each of 10,000 assets needs its own close-range scan. The electronics component warehouse mentioned earlier cut its count from two weeks to three days by replacing paper entry with scanning, but that worked because everything was on a shelf with the label facing out and within reach. At outdoor worksites, in pipe cavities, or in high-bay storage, scanners cannot always reach and productivity drops.

Remote lot reading: RFID Radiofrequency Identification Category

RFID, or radio-frequency identification, uses a tiny chip in each tag to communicate with a reader over radio waves. It works well where there are many items or they are hard to reach. At an auto parts plant we worked with, the warehouse had four levels of high racks, with parts bins on pallets more than 2 m high. During counts, staff climbed up and down to scan tags, taking more than ten minutes per location. With RFID, staff could stand in the aisle and read dozens of bins at once within 3–5 m, cutting the count to under two minutes per location.

These systems cost considerably more in hardware than barcodes. RFID tags cost from a fraction of a yuan to several yuan each, depending on whether they are ordinary paper tags or on-metal tags. Fixed readers go from several thousand to over CNY 10,000; handhelds cost from a few hundred to CNY 2,000-3,000. Tagging 30,000 assets at a midsize manufacturer typically costs CNY 50,000-150,000 in tags alone. Run the numbers before choosing RFID. If two years of manual stocktaking labor cost more than that, it is worth adopting. If assets are few and concentrated, barcode scanning may be more economical.

RFID is also limited. Metals and liquids are very disruptive to radio frequency signals, with reading rates falling to about 70% on the metal shell, requiring specialized anti-metallic labels; between stacking labels, they're also shielded by adjusting antenna angles and power. The most common failure we have seen in the project is that customers purchase civilian labels cheaply, and then they can read only half of them in the cabinet.

Replace the scrutinizing gun with a camera: AI visual recognition system

Visual recognition takes two years to actually get into the inventory scene. It's a thought that goes without labeling, taking pictures directly with a camera, allowing the AI model to identify what is in the map, and which asset number. The more mature scenario is: the finished packaging on the shelf (AI through overpacking, printed by means of SKU), the desk-chair computer in the office area (AI through the form of an external identification device) and the tray and cargo on the warehouse ground.

A customer who was a clothing electrician tried to get a visual count in the sorting area: camera on the shelf, every few minutes, AI automatically compared the number of opposites and the number of system accounts, and the difference over the threshold was reported. Manpower fell from two shifts down to one person ' s inspection rate, dropping from 2% to less than 0.3%. But it was also clear that short panels for the visual system were: recognition accuracy and shot angle, light, cover-up directly related. The same clothes were hanging on the rack and stacked in boxes, AI recognition rates were cut off; retroactive metal pieces, comparable products with similar colours, were also easy to roll over.

The visual system is structured in two main components: camera hardware and AI models. The cameras themselves are not expensive (a couple of hundred pieces of industrial camera for 2 million pixels), but the AI model needs training, with each industry, and each warehouse layout being adjusted separately. Shanghai-based Sievet technology usually uses 2-4 weeks to fine-tune a model with field samples, with higher input than direct purchase of mature products, but with better matchability, and recognition rates above 95%.

Where's the missing thing?

The LPOS addresses not "no" but "where." There are three common techniques: GPS (for outdoors, for example, engineering vehicles, containers), UWB superb broadband positioning (accuracy within 30 cm, often used in plant and hospital equipment management) and Bluetooth Beacon (3-5 m accuracy, low cost, rough location).

On a municipal engineering project, more than 300 machines were distributed across over ten sites citywide, and simply finding them took a week per count. Once each machine had a GPS module and SIM card, the platform produced a location heat map and the count could be done online. Staff only had to check condition and appearance. The count went from a week to a little more than half a day. Communications and location tags are the main costs. For 300 units, annual communications cost about CNY 30,000–60,000, and tags cost CNY 150,000–300,000 up front.

UWB is highly accurate in positioning, but needs to deploy the base station (i.e. a signal receiver attached to walls or ceilings), a 5,000-metre workshop that would probably cover 20 to 40 base stations, with a single base station of several thousand dollars, and construction wiring, which would have significantly higher pre-commitment than GPS. Bluetooth Beacon is going to the compromise route, with a cheap label (one or two hundred), but with limited positioning precision, which is for this requirement "in terms of which floor, which area."

No silver bullet: Multi-modular combinations are becoming mainstream

In the actual project, there is a growing number of single-identification methods, most systems are combined. When we do an inventory of a large retail chain, we end up with a combination of "bar code + RFID + vision": the door store makes small-scale inventories using a scrutinized rifle, the warehouse uses RFID to make bulk count, the whole shop table behind the door shop is automatically identified by Ai, and three data are aggregated into the same backstage. The benefits of this combination are long distance - bar-code costs are used in more than SKU scenarios, and RFID uses visuals at the back of the warehouse and shelf, high or dense areas where no easy to move.

The complexity of the combination system is even more complex. To harmonize data formats in three modes of identification, to avoid double counting, to ensure that timetamps are aligned, and to be more detailed in the data processing modules on the back, this is why there have been more and more systems with the concept of "clip-top" in the last two years -- medium stations that aggregate data from different methods of identification, reconcile, clean up, and collect at the front end. When we do projects like this, we usually run through existing processes for customers, then access different identification modules in stages, instead of putting them all together.

What do you think of the type? What kind of identification do you need?

Three factors drive the choice of identification method: asset volume, storage density, and budget. Below 10,000 assets, if they are organized and individually accessible, barcode or QR scanning is generally sufficient and costs the least. For more than 10,000 assets, dense storage, or items stored high up, RFID batch reading has clear advantages. If assets are varied, look different, and are hard to tag, machine vision is worth testing. If they are spread across several sites and need continuous location tracking, a positioning system is a better fit.

There is a rough reference line on the budget: bar-coded, single-asset assets put in a few cents to a few cents; RFID, each of them, was a few cents to a few cents; visual classes were mainly spent on front-end cameras and backstage models, ranging from 50,000 to 500,000; positioning categories ranged from 100,000 to millions. How exactly does it fit? Without a standard answer, the key is to figure out where the pain points of their own inventory are going -- whether they can't be found or not, whether they can't read or not -- and then to which identification method to solve them.

If you are evaluating the inventory system, you can use a typical set of assets for a small-scale test to identify the way it is done. You can run one or two real scenes before deciding.