Something went wrong while submitting the form. Did you complete all fields?
Something went wrong while submitting the form. Did you complete all fields?
Computer vision (CV) systems are increasingly being adopted and integrated into various stages of the manufacturing process.
Essentially, computer vision is a branch of AI that specializes in analyzing images or videos to identify objects and determine their relative positions with a level of accuracy that matches or surpasses human capabilities.
By mimicking the precision of human experts – while avoiding the errors, fatigue and high costs associated with manual monitoring – computer vision systems are being used to improve quality, safety and operational efficiency in factory processes at scale.
First, image or video data is captured from a source—this is a imaging device, recording video or taking pictures, depending on the application. This data is processed at the edge e.g. filtered or enhanced before being stored alongside historical data.
Next, the image data is fed into an AI-powered computer vision system, which uses a variety of machine learning methods to identify objects and detect irregularities. Depending on the implementation, these irregularities could signify anything from minor deviations that are within acceptable tolerances, all the way to critical operational problems requiring immediate intervention.
Computer vision systems like these can be extended: for example to classify anomalies by type and severity, to monitor trends over time, or to provide predictive alerts to prevent potential problems before they arise. They can also offer real-time feedback, allowing for rapid responses when needed.
Capture high-quality image data from various contexts, using multiple imaging devices or cameras, to create a comprehensive dataset for analysis
Preprocess the collected data by filtering, normalizing, and enhancing it to ensure optimal quality and consistency for AI analysis
Apply AI models like CNNs or autoencoders, to identify known objects or irregularities that deviate from the wanted norm in real-time
Classify findings by type or severity, determining the appropriate response: alert sending, investigation, or automatic next step
Send immediate alerts and integrate automation systems to respond quickly, minimizing risks and ensuring smooth operations
CNNsCNNs excel in visual recognition tasks by learning spatial hierarchies from images, making them ideal for manufacturing. Trained on large datasets, they can distinguish between defective and non-defective products and address key questions in computer vision processes: classification (identifying the defect), positioning (locating it), and segmentation (counting defects). Their advantages include:
Supervised learning: High-accuracy classification using labeled images.
Robust to position variations: Effective detection regardless of feature position e.g. for video data.
Targeted feature learning: Recognizing complex defect patterns and normal variations, making them highly effective for specific object identification.
Auto-encodersAuto-encoders leverage their ability to learn and reconstruct data to detect irregularities by identifying and classifying unknown objects in manufacturing products. Trained on product images free of irregularities, they highlight deviations from normal patterns in new images. Their advantages include:
Unsupervised learning: Detecting anomalies without labeled datasets, valuable when labeled data is scarce.
Unseen defect detection: Recognizing deviations from learned normal patterns, flagging potential defects even with new or unexpected variations.
Robustness to noise: Ignoring random noise, focusing on underlying structures to minimize false positives.
Modern manufacturing processes rely heavily on computer-based automation at the shop floor level. The term "Industry 4.0" refers to the digital transformation of manufacturing, characterized by the integration of smart, autonomous systems powered by data and machine learning. This advancement enhances production processes, making them more efficient, productive and less wasteful.
Computer Vision is one part in this toolbox, delivering clear benefits within manufacturing facilities:
The implementation of computer vision applications in manufacturing covers a wide range of use cases from production lines and quality control to operational safety and packaging. Manufacturing giants as well as boutique manufacturers can significantly benefit. According to Deloitte labor productivity and production output can be increased by about 12% and 10%, respectively.
PRODUCTION LINES
In modern manufacturing, the integration of computer vision (CV) in defect detection has significantly advanced production lines, marking a shift from traditional methods to the sophisticated systems of Industry 4.0. A prime example is Hitachi High-Tech's launch of the DI4600 Dark Field Wafer Defect Inspection System, designed for semiconductor production lines. This system offers enhanced detection capabilities, enabling highly accurate monitoring of defects and particles on patterned wafers which are difficult to detect for the human eye. The DI4600 improves throughput by approximately 20% through faster wafer transfer and optimized inspection operations, which is crucial as semiconductor devices become increasingly complex and miniaturized.
This technology plays a critical role in maintaining high yields and reducing manufacturing costs, especially as semiconductor production volumes continue to rise. The DI4600’s ability to monitor cleanliness and detect defects at a high precision level ensures that semiconductor devices, such as DRAM, FLASH, and logic semiconductors used in AI computing and autonomous driving, meet stringent quality standards.
Similarly, Volvo Cars has implemented a CV-based Atlas quality inspection system developed by UVeye since 2020. Installed at the end of the assembly line, this system uses over 20 cameras within an aluminum tunnel to capture hundreds of images per second. The AI algorithms then assess surface quality, detecting 10% to 40% more defects, including microscopic scratches and dents, than conventional manual inspections. The system's ability to identify defects as small as 0.2 millimeters highlights the essential role of computer vision in enhancing both the quality and efficiency of modern production lines.
OPERATIONAL SAFETY
Operational safety is a critical concern in industries, particularly in high-risk occupations like transportation, material handling, and construction. In 2020 alone, 4,764 U.S. workers lost their lives on the job, with nearly half of these fatalities occurring in transportation, material moving, construction, and extraction occupations. To address these risks, computer vision (CV) technology is increasingly being employed to enhance worker safety by monitoring hazardous areas, identifying unsafe behaviors, and improving emergency responses.
One key application of CV in operational safety is posture detection. For instance, TuMeke has developed a CV-powered ergonomic risk assessment platform that uses smartphone videos to analyze warehouse activities, such as lifting boxes. The platform generates a risk summary that identifies unsafe postures, helping to prevent injuries by promoting better ergonomics.
Another crucial application is fatigue detection, especially in transportation. Cipia offers CV-based driver monitoring systems like Driver Sense, which detects signs of drowsiness and distracted driving in long-haul truck drivers, significantly enhancing road safety. Additionally, their Cabin Sense solution monitors passenger safety by detecting posture, seat belt usage, and other critical factors.
PHARMACEUTICALS
Quality control is crucial in manufacturing, as it ensures products meet high standards, reducing costs and enhancing a brand's reputation. In industries like pharmaceuticals, where stopped production can cost up to $22,000 per minute, robust inspection processes are essential. Computer vision (CV) technology plays a key role in this by providing precise, automated quality control.
For example, England-based Pharma Packaging Systems has developed a CV-based solution specifically for tablet counting and quality inspection. This system uses advanced CV algorithms to analyze tablet images, checking for correct dimensions, color, and quantity. Any defective tablets are automatically rejected from the production line, ensuring that only products meeting the highest standards reach consumers. This integration of CV in pharmaceutical quality control not only improves efficiency but also maintains the stringent quality necessary in the industry which ultimately can save lives.
WAREHOUSES
Packaging plays a crucial role in product presentation, safety, and usability, often accounting for 10-40% of a product's retail price. However, human error in the packaging process can be costly. Computer vision (CV) and robotics are transforming this space by increasing accuracy, reducing costs, and optimizing resource allocation.
Amazon, a leader in warehouse automation, has significantly expanded its use of robots, with the number of robotic units in its fulfillment centers skyrocketing from 350,000 in 2021 to 750,000 by mid-2023. Amazon claims to be the world’s largest manufacturer of industrial robots, with advanced systems like Robin and Sparrow playing pivotal roles in streamlining the packaging process. Robin assists in scanning and sorting packages, while Sparrow optimizes fulfillment by handling individual products more efficiently.
Amazon’s Sequoia robotic system further enhances productivity by identifying and storing inventory 75% faster than human workers, and speeding up order processing by up to 25%. This automation not only improves lead times for shipping but also reduces the physical strain on employees, leading to a 15% reduction in incident rates and an 18% decrease in lost-time incident rates at robotic sites in 2022 compared to non-robotic plants.
Despite the rise in automation, Amazon continues to create new job opportunities, with 700 new categories of skilled roles that didn’t exist before. This integration of CV and robotics in warehouse automation exemplifies how technology can boost efficiency while supporting human workers, ultimately leading to more effective and safer packaging processes.
The use of CV systems such as Hitachi High-Tech's DI4600 and Volvo Cars' Atlas increase data storage volumes, as they require large training datasets to develop the CV system, which will generate vast amounts of high-resolution image data. The implementation of CV systems brings on the challenge of a data tsunami, which can reach petabyte-scale volumes. It is essential for firms to store these volumes efficiently as they are crucial for quality, compliance and overall operations’ progress.
UltiHash eliminates redundant data on a byte level, independently from data type and/or format, allowing companies to save significantly on their data volumes.
In CV systems whether for training or production, datasets are highly similar. Deep learning implies training on large datasets of similar images to reliably spot defects, and in production environment, the products analyzed are consistent with the training data. UltiHash leverages this similarity to offer up to 60% space savings, helping companies cut costs and reduce resource usage across on-premises and cloud infrastructures.
In pharmaceutical manufacturing, CV systems for quality control depends on fast data access. These systems generate and process large volumes of high-resolution image data, where delays in data retrieval from storage can cause significant bottlenecks, disrupting the production flow and risking costly downtime. However, when using managed services, such as cloud storage, inconsistent data access speeds can complicate this issue, potentially leading to further production delays.
UltiHash’s lightweight algorithm and tailored architecture for AI operations ensure high throughput and low latency, enabling fast and predictable data access for both read and write operations.
In industries like pharmaceuticals, where paused production is extremely costly, robust inspection processes are essential. UltiHash addresses the critical issue of data access speed by enabling fast and consistent access to high-resolution image data. This prevents production delays, whether the data is stored in the cloud or on-premises, and supports the quality control needed to carry out smooth operation such as high-quality medicine packaging.
Integrating new tools into a company’s tech stack can be challenging. In manufacturing and packaging, the complexity rises with advanced robotics like Amazon's Robin and Sparrow, combined with ingestion tools like Apache Kafka and machine learning tools like Weights & Biases. These integration issues can create inefficiencies, turning automation benefits into headaches, highlighting the need for solutions that seamlessly connect diverse technologies.
UltiHash’s S3-compatible API and Kubernetes-native design ensure seamless integration with enterprise infrastructure - cloud or on-premises.
UltiHash enables seamless integration across cloud and on-premises environments. In addition to supporting the AWS tools, it integrates seamlessly with wide range of tools like Apache Kafka, AWS SageMaker, and Prometheus, as well as open table formats like Apache Hudi, Apache Iceberg, and Delta Lake. This allows users to build a broad tech stack with all the tools they need, ensuring seamless integration and maximizing the benefits of automation.