The Rise of Augmented AI Component Counting and Managing
By Eddy Lin, CEO
During the Industry 4.0 revolution towards fully automatic robots, material management has remainedan enclave populated by humans, including component counting, sorting, and storing. While component counting is not new, management has been trying for years to automate the process and eliminate labor.
Current issues with component counting include:
● Difficulties locating components
● Human errors in components issuing
● Stocks expiration, salvation component
● Components exposed to humidity and air
● Missing expensive components
● Misplaced components
● No traceability when mistakes happen
These problems need to be dealt with manually and have been the root of many challenges.
To deal with these, the pioneers of the SMT industry began implementing smart counting and smart storage solutions. Factories worldwide are using machine learning storage solutions, such as smart racks, more and more because they are managed through networked computer systems.
Each production line must address three challenges:
1. Components must be recounted on their way back to the warehouse ERP or EMS systems.
2. Incoming materials must be scanned and entered into the system, preferably with a unique identification (UID).
3. Where are they stored?
Challenge 1: Counting Components with X-ray
Here, we discuss only the modern method of using X-ray machines to count components. Thistechnique is used to acquire the X-ray image of a reel, a tray, or a tube (for this article, we will use reels only). Figure 1 shows X-ray images of reels and a tray. To truly illustrate the benefits of X-ray in component counting, simple examples such as 0402 and 0201 cannot be used. Instead, the examples include a difficult connector reel and a tray.
The X-ray source is considered a spotlight source. The cone beam from the focal spot creates a paradox distortion on the outer radius of the reel. Therefore the images of the components are convoluted. One image can produce more than a dozen algorithms, but how does the computer find the correct one to use? That is where artificial intelligence (AI) comes into play. The AI deconvolutes the overlapping images, searching for a known pattern, and counting the matching algorithm. Figure 2 shows the counting results. The process usually takes 300 ms.
Figure 3 is another example of counting a complex reel.
The X-ray counter sometimes has an internal camera to read the barcode. After the camera deciphers the barcode, the computer uploads the barcode information, such as the UID, as well as the quantity to the central system and print label as instructed by the central system. Some companies no longer require new labels because the UID and real-time tracking system are sufficient.
For the camera to read the barcode, it first takes a picture of it, enhances the contrast, and sends it to the computer for processing. The computer algorithm can read 1D, 2D, data matrix, and optical character recognition (OCR). Therefore, the information, including the date code (typically printed as a number instead of barcode) can be read. The information then must be separated into the correct catalog based on the barcode rule. If the barcode rules do not exist, the computer can use a customed template to separate the barcodes based on the barcode sequence. This leads into the second challenge as the barcode reading and MES integration are the same asincoming material processing.
Challenge 2: Incoming Material Processing
When receiving the reels, trays, and other components, whether new or consigned, the factory must register them in the system. Generally, the current process involves manually scanning the label. However, this can be handled automatically with a high-resolution optical system. The barcode camera system must have a large field of view (up to 15 in radius), a high depth of field (up to 80mm), and high pixel counts. Once the camera locates the label, the software crops, straightens, and prepares it for reading.
Figure 4 is an example of a captured image.
There typically are two ways of deciphering the barcode: barcode rule or AI teaching. If barcode rules are known, the computer can categorize the reading by criteria, such as Name, Numeric Only, MinLength, MaxLength, Match Index, Match Content, Prefix String, and Contain String, among others.
However, for a new reel without an internal company label, the rules are unknown. AI prevails in performing such tasks. A trained AI finds the label, reads it, categorizes the reading results based on templates, and even reads the date code by OCR.
Once the AI software successfully reads the label, the label results need to be uploaded to the central system using either real-time mode or batch mode. The following diagram (Figure 5) shows the typical flow of most customers; however, customization is available.
Challenge 3: Where Are They?
Now the reels are counted, the labels are scanned, and data is uploaded to the central system. But where are they? This is the third challenge. Implementing a smart storage system will help keep track of every component that comes in and out of the warehouse. Thesesmart racks include sensors and color LED lights to help with this.
Figure 6 illustrates two racks, the Porter rack (left, Figure 6a) and the Station rack (right, Figure 6b).
Both of these systems have wheels and can be placed anywhere on the production floor, with connections to the central system both wired or wirelessly. The Porter rack is designed with a battery to ensure constant connection during transport. It also can be used as a stationary combination rack.
There are several advantages of implementing smart racks:
● Inform the central system where the reels are
● Inform the system of unauthorized reel removals
● Once a work order is received, light up the LED to show the needed reels
● Multicolor LED for several kitting jobs simultaneously
● Once a component is below the minimum level, the system receives a request to replenish the rack
● FIFO, or the reels with less quantity first
Conclusion
There are solutions available, either semiautomatic or automatic, forlabel scanning, X-ray counting, and smart storage to improve production. Before implementing, however, companies must internally lay out the process and consult with the supplier about it. The initial integration process can be time-consuming, but provides numerous long-term advantages.
While some code writing can be done before the order is placed, thefinal finishing always requires onsite cooperation once the machines are linked to the central system. If a customer does not have an internal software programmer to write the codes, such as matching the PO number on the label to the internal PO number, Scienscope can provide a customization system with a reasonable fee to speed up the process.
For more information about Scienscope’s advanced technologies, visit www.scienscope.com.