Two of the biggest needs of the electronics industry are to automate production processes and improve quality. However, some applications are too complex, time-consuming, and costly to program into a rule-based algorithm. Using testers to handle different types of parts and make judgment-based decisions can also lead to errors and slow down production.
Deep Learning - first deep learning technology designed to solve difficult electronic applications, providing human-like testing with the speed and robustness of computer systems, in a single Field maintenance facility makes it easy to train new parts and fault types in the field.
During final assembly verification, it is extremely important to inspect the PCB (printed circuit board) to determine the presence and correct placement of LEDs, microprocessors, and other components. Deficiencies or errors in this position can seriously affect the performance and life of the PCB. It is essential to detect these errors before the PCB is assembled into the device or sent to the customer.
However, automated testing can be difficult when there are slight changes in the appearance of the PCB. Factors such as subtle light contrasts, changing viewing angles, and glare on metal surfaces can make it difficult for machine vision systems to distinguish components. In particular, parts that are close to each other are as difficult to distinguish as independent parts. Programming these checks into a rule-based algorithm requires a lot of time and effort, is error prone, and is challenging for the field engineer to maintain the system.
Deep Learning provides an automated solution for field PCB assembly checks, rivaling manual human inspection. The tool uses machine learning to identify components from marked "good" images, building a reference model of their normal shape. From there, the tool generalizes the distinguishing characteristics of components based on size, shape and surface characteristics, helping to locate and count components on the PCB quickly and accurately. . Thanks to this technology, it becomes more efficient and reliable to verify the correct assembly of the board, which optimizes the manufacturing process and ensures product quality.
Assembling components such as mouse diodes onto the board without affecting their electrical connections requires perfectly gluing solder resistors onto the board. Even minor defects in solder paste can lead to wire breaks, short circuits and other electrical problems. These flaws can vary in size, shape and appearance, which are often caused by glare. It is very difficult to program the automation to withstand significant part changes under these conditions.
Deep Learning quickly identifies solder resistance on diodes and checks for defects using a defect detector in monitor mode.
Aesthetic inspection can be complicated as defects occur at different levels, from composition to packaging and casing. Scratches, dents, and other defects may not affect functionality but can significantly affect quality and consumer perception. Among them, obvious defects can lead to rejection, while smaller defects are acceptable.
Engineers can use Deep Learning training and monitoring defect detection to automatically look for specific defects such as scratches, while removing unimportant anomalies and variations. This tool is optimized to work effectively on images with low contrast or poor image quality.
Pre-assembly insert inspection is an important step to detect defects in the internals and battery before mounting the case. Locating and inspecting the battery is challenging during assembly due to its instability. However, deep learning vision software simplifies the automatic detection and identification of defects in the metal surface of a battery.
Deep Learning technology helps manufacturers check battery integrity and distinguish between aesthetic and functional defects before a phone is assembled. Engineers can train software on "good" and "bad" images of labeled defects. The tool will learn the normal form of the battery and the acceptable natural variations. The parameter can be dynamically adjusted during training so that the model correctly detects and classifies functionally problematic images. Once deployed, the fault detection tool will automatically identify and remove the faulty battery.
Machine vision is used in semiconductor manufacturing to monitor quality and detect defects. Scratched, twisted, bent or missing pins should be noticed and removed by the manufacturer. With many types of potential defects, it is not efficient to program rule-based tests. Instead, deep learning vision software helps to reduce semiconductor defects and improve productivity without using extensive error libraries.
Deep Learning simplifies the identification of anomalous features without training on "bad" images. The engineer only needs to train the software with a sample of "good" images in unsupervised mode. This technology automatically learns the normal shape and position of the chip's wires and pins, thereby detecting and characterizing all misaligned features as faulty. This saves time and reduces the complexity of finding defects.
Classification of electronic components is often difficult when they are divided into many classes with many different visual variations. This includes parameters such as capacitor type (ceramic and electrical), size and color which vary by manufacturer and specifications. Cylindrical shapes and lighting can also be complex. To solve this problem, VisionPro ViDi provides a deep learning-based solution that allows to automate the classification of multiple components in a single image.
Using the fault detection tool, a software-trained engineer monitors the annotated images, classifying the capacitors and gold capacitors as "good" parts. At runtime, the tool will segment and classify all capacitors and gold capacitors as one category. Deep Learning can distinguish different capacitors based on their color and markings, even if they look the same on single images.
The smartphone uses the OEM USB connector for charging and data transfer. This connector should be thoroughly inspected prior to installation, including power and data contacts, shields, and mount connectors. Common errors on the USB connector include fire, short shot, dust, scratches, and displacement. Even minor errors on the connector can have a big impact on the connection performance, stability and longevity of your phone.
The Deep Learning defect detection engine is trained on diverse image data, including both functionally important appearance variations and those that do not affect function. During training, cosmetically acceptable defects are ignored to focus on defects that may affect function, making the process simpler and faster. When deployed on a line, lighting and part placement are important to capture as many defects as possible.
Decoding character strings on chips assembled on PCBs is a challenge for machine vision systems, especially when faced with glare and low-contrast images. Our Optical Character Recognition (OCR) has been optimized to handle reflective surfaces and recognize characters regardless of distortion, misalignment, and poor image quality.
With Deep Learning, reading distorted characters is no longer a problem. A deep learning-based OCR approach saves time and reduces over-labelling. The software only requires the engineer to set the region of interest and the character size. The tool's pre-trained font library automatically decodes characters and reads strings without training. For difficult character reading situations, the software can be directly retrained using variant characters.
Once fully assembled, before packaging, the smartphone undergoes a quality check to detect any defects such as scratches, cracks, or discolouration on the case and glass. Although not related to function, these problems can adversely affect the appearance of the product.
Deep Learning teaches from images of unacceptable disability as well as acceptable aesthetic variation. The tool learns to categorize them and can adjust the parameter to meet market requirements. Combined with exceptional lighting and proper presentation, a Deep Learning-trained tool checks every aspect of a smartphone, from the display to the strap and back, for dents, scratches, and dents. , and change the color. This thorough inspection ensures that only aesthetically perfect products are packaged.
The speaker grille is a perforated piece of metal that protects the mobile phone's sound module from dust and damage while minimizing the impact on the sound. The mesh should be visually inspected for cracks, scratches, holes, deformation, broken edges, foreign matter such as dirt or hair, and damage caused by the punching process. This mesh is visible to the end user, and therefore aesthetic damage must also be detected.
Deep Learning's defect detection engine is trained with a variety of speaker grilles to ensure the range of variation of normal parts. It analyzes and marks points that fall outside the acceptable range when scanning through meshes. This tool has the ability to combine with traditional machine vision, especially suitable for precise alignment and measurement of speaker grilles. The combination of deep learning and traditional vision tools surpasses manual testing, dramatically improving speed and maintaining consistency in the production process.
Before the camera module is installed on a mobile device, the surface of the module must be checked to ensure that there are no foreign objects, scratches, stains or dust on the lens.
Deep Learning is trained on a variety of defect-free lenses to understand variation of normal parts. In Unattended mode, it scans through the lens string and indicates those that are out of acceptable range, while minimizing false positives.
Lens defects often stem from specific causes such as contamination by dust, smudges by oil or fingerprints, and misalignment within the lens. User can define defect type or size accurately by monitoring mode. In this mode, the user trains the system with examples of good and bad parts, to clearly recognize defect areas and label them as scratches, stains, contamination, and otherwise.
This information is applicable to reverse manufacturing process control. For example, if there is a permanent scratch, it may be caused by a machine deviation or production speed deviation. By identifying the root cause, manufacturers can quickly respond and minimize the generation of bad parts.
When the battery cell is sealed, liquid electrolyte is poured in through the small hole in the cap. The hole after being filled will be closed with a sealing pin or a lid, then welded tightly to fix it.
The exact position of the sealing pin was determined by measuring the height of the pin using a laser or a 3D displacement sensor. However, detecting defects during soldering, such as dents, burns, pinholes and breaks, is not easy through traditional machine vision due to various possible defects and different reflections of the battery cover.
After the weld is complete, Deep Learning detects many potential defects on the sealing pin weld. Apply trained deep learning on a variety of sealing pins to understand normal part variability, including acceptable levels of cosmetic defects, highlights, and shadows.
Cylindrical and button batteries are sealed by welding the cap to the housing containing the electrode and electrolyte. Low heat methods such as laser or resistance welding are often preferred to protect the battery internals and maintain sealing accuracy. Defects in this process can lead to electrolyte leakage and reduced performance.
Welding problems can appear anywhere around the perimeter and can have a variety of shapes. Dimples, scratches and other defects can appear anywhere on the surface. Both types of defects are difficult to distinguish on the reflective surface of the cell. All of these problems make it difficult to use traditional machine vision to accurately detect defects.
Given the large scale of production and the sophisticated nature of these defects, manual inspection faces the same challenge.
Deep Learning has been trained with a variety of sealing pins to detect weld defects. When scanning through the battery covers, it analyzes and marks which welds exceed the acceptable range, while minimizing positive results due to cosmetic defects.
The battery electrodes connect to the external circuit via metal tabs. These metal lumps, thin and soft, are made of a variety of metals and break easily. Scratches, bumps, holes and dust are common faults. Damaged tabs can reduce or prevent connection, which is why they need to be defect-free before soldering to the electrode.
In addition, improper soldering also leads to problems such as burns, bumps, lack of balls. Poor solder interferes with or degrades connection and performance.
Deep Learning has been trained with a wide variety of unbroken tabs and precision welded tabs to detect solder defects. Apply trained deep learning on a wide selection of valid tabs to understand the variability of normal parts, including acceptable levels of cosmetic defects.
To ensure a solid electrical connection, the conductor needs to be soldered in place or to the terminal. The welding process combines two dissimilar metals by heat to create a solid connection. It is important that the solder is fused to the right level, not too much, not enough, in good shape and in the correct position. For manufacturers, managing multiple lines for multiple electronic components requires all connections to be secure.
Given the high variability and heterogeneity of weld connections, including shape, location, colour, reflectivity, structure and surface markings, testing can result in high false positive rates. , aka overkill. This leads to the removal of fine parts. Mismarking good welds is an error that requires manual inspection, delays line speed, and often causes errors in weld defect identification.
The wide variety of sizes, colors and other characteristics between parts from batch to batch causes a wide range of variations and makes it difficult to distinguish between good and bad connections, making the method traditional rule-based testing becomes ineffective
Users can train Deep Learning's fault detection engine on a variety of spot solder connections to understand the full variation of normal parts. The tool scans through the weld and analyzes it, and eliminates false positives. The Deep Learning classifier is then trained with labeled weld defects, to classify specific problems such as irregular shapes, blowholes, cracks, burns, and surface contamination. face. Classified defect classes can be applied to control upstream processes and reduce defects over time.
Multilayer Ceramic Capacitors (MLCC) contain multiple stacked capacitors with metal terminals that connect to an integrated circuit board. MLCCs are subject to many manufacturing defects such as cracks, blisters, chips, contamination and coating gaps. These failures affect not only the faulty MLCC but also the surrounding components and the integrated circuit board.
MLCCs are small, numerous, and have many small defects that vary in shape and location. Shiny surfaces interfere with traditional machine vision.
Manual inspection and automatic optical inspection (AOI) machines are used. The AOI examines all six aspects of the MLCCs, then humans test the statistical sample. However, both of these methods have a high rate of overkill or are time consuming. The overall process is expensive, slow, and lacks data useful for process improvement.
Cognex has developed a special aesthetic optical inspection (COI) machine for MLCC inspection using both custom lighting and deep learning vision technology. Custom lighting modules are applied for MLCC inspection, eliminating extraneous surface variations and focusing on defect detection on capacitor bodies and connectors.
After the MLCCs have undergone AOI testing, the COI machine continues to test to reduce false positives and remove unnecessary good components from the manufacturing process. This combination delivers speed, accuracy, and provides process-optimized data compared to manual testing.
The deep learning classifier is trained on images of a variety of defect-free and defective MLCCs. This tool learns to classify the many types of defects that may appear, while understanding the full variation of normal parts. After training, it is capable of inspecting the entire MLCC and immediately marking any defects outside the acceptable range or identifying good parts that were previously marked as defects.
Classified defects can also be applied to control the upstream process, helping to reduce defects in the part over time.
Deep Learning is a vision solution based on unique Deep Learning technology, specially designed for automation in factory environment. Built on state-of-the-art machine learning algorithms, this technology has been tested and optimized, and has proven effective in real projects.
Different from the rule-based approach of traditional machine vision, Deep Learning not only solves the testing challenge in a similar way, but also learns to recognize patterns and anomalies through reference images. . Deep learning's automation and scalability enables complex testing applications, including defect detection and final assembly verification, that traditionally relied on manual work. of human.
>>> Xem thêm: EDGE LEARNING VÀ DEEP LEARNING
----------------------------------------------------------
CONTACT INFORMATION
New Ocean Automation System Company Limited
Website: New Ocean Automation System
Hotline: 1900 0224
