Defect Detection Defect Detection

Defect Defection with Computer Vision Whitepaper

White Paper on Visual AI for Defect Detection: Discover the Tremendous ROI Benefits of Computer Vision for Defect Detection

Imagine a team of inspectors which can monitor defects for every manufactured item with laser-point accuracy, 24-hours a day. This team never gets tired, never gets distracted, and works for a fraction of the cost of human quality control inspectors.
Moreover, this team can instantly scale to handle inspection tasks of any size or complexity. This team doesn’t have to be imagined, as this high standard of defect detection is already available to manufacturers, and it does not require human inspectors, nor months of development.The team is not a team at all, but rather a sophisticated machine vision system for defect detection.
This computer vision technology uses a network of cameras, edge devices, with highly customizable AI models to help manufacturers eliminate their quality inspection bottlenecks, dramatically reduce costs, and produce more and safer products faster – all while meeting and exceeding their quality control standards. With new computer vision platforms, the customization of AI models is now making these deployments ever more flexible.

GET THE DEFECT DETECTION WITH COMPUTER VISION WHITEPAPER

AI for Defect Detection with Computer Vision

Financial Impacts of Manufacturing Defects

In Quantifying the Full Costs of a Product Defect, Alec Baker identifies the following key ways that product defects cause enormous costs for manufacturers and businesses:

  • Notification: The cost of tracking down and notifying retailers and customers about defective products.
  • Transportation and repacking: The extra transportation costs associated with sending and returning defective products, then the cost of sending the repaired or replaced products back to the customer.
  • Destruction and disposal: The cost of destroying and disposing of defective, unusable products, and the wasted materials and labor required to build them.
  • Product replacement and repair: The cost of materials and labor required to replace and repair defective products.
  • Lost revenue and reduced sales: Customers that purchase damaged products are less likely to buy from that manufacturer again, which reduces sales revenue. The reputational damage that comes from news spreading about defective products also hurts sales revenue.
  • Marketing and public relations: The cost of marketing and public relations efforts to rebuild customer trust and rehabilitate sales. In its 2017 Product Recall Report Allianz Global Corporate & Specialty offered additional perspectives on the costs of manufacturing defects and their associated product recalls:
  • Recall costs and reputational damage: “The biggest single cost of a product recall event is the loss of sales and business interruption, both from the recall itself and the reputational damage.”
  • Product liability and personal injury lawsuits: “Defective product incidents have caused insured losses in excess of $2 billion over the past five years, making them the largest generator of liability losses.”

Considering the tremendous financial costs of defective products, the ability to identify defects quickly and effectively before they leave the manufacturing facility can bring enormous ROI benefits. However, due to the limitations of human quality control staff, there is only so much that manufacturing facilities can do. A certain number of defects will invariably go undetected, and potentially trigger some or all the above-listed costs.

Challenges of Traditional Defect Detection Strategies

Manual visual inspections by humans have long been required to ensure that manufactured products meet the highest quality control standards. These human inspectors were traditionally required to visually examine details like color, texture, shape, quality, position, labeling, etc. in industrial and CPG manufacturing settings. However, human visual inspectors come with several downsides and disadvantages:

  • Human limitations: Human quality control inspectors are limited in their capacity to detect defects and prone to making mistakes and errors depending on energy levels and attention spans.
  • Constant training and retraining: Human quality inspectors require training and retraining to adapt to new quality control requirements for different types of products.
  • Scaling challenges: As manufacturers try to scale their operations, they need to hire new quality control inspectors – representing additional delays, costs, and training burdens. Also, as operations scale, existing inspectors may become overworked and fatigued.
  • Prone to varying levels of accuracy and human errors: Human quality control inspectors are prone to experiencing eye fatigue, repetitive motion injuries, and other physical stresses as they endure long hours and struggle to apply greater attention to detail.
  • Prone to boredom and distraction: Inspectors engaged in monotonous visual tasks are prone to boredom and distraction, which leads to declining levels of defect detection accuracy.

The boring and monotonous nature of visual inspection tasks is particularly concerning. The longer human workers try to focus on mundane visual tasks – such as looking for misplaced bottle caps, monitoring live video feeds, checking for painting errors, or watching for broken parts – the more bored, fatigued, and error-prone employees become. Research shows that work-related boredom brings a host of negative consequences for both employees and employers, such as:

  • Job dissatisfaction
  • Absenteeism
  • Employee turnover
  • Counterproductive work behavior
  • Work-related injuries
  • Less effort
  • Poor performance
  • Errors and mistakes

Despite the above negative outcomes for employees and employers, the use of human quality control inspectors is a vital part of ensuring quality in manufacturing operations, and especially important in precision manufacturing. Nevertheless, the error-prone nature of human quality control inspectors – and the rising cost of hiring and training new inspectors – creates a serious bottleneck when it comes to achieving fewer defects, better automation, and greater scalability of manufacturing processes.

Get More Computer Vision Resources

For data sheets, videos, and the full suite of resources from Chooch AI, please visit the Computer Vision Resources page.

ROI Benefits of Computer Vision for Defect Detection

Advanced visual AI technology can eliminate the errors, inefficiencies, and costly consequences associated with employing human workers to complete visual inspection tasks. By overcoming the human limitations associated with completing these tasks, computer vision for defect detection achieves the following ROI benefits for manufacturers:

  • Streamlines workflows: By speeding up and improving the visual inspection process, visual AI empowers manufacturers to speed up their entire manufacturing process, reduce inspection-related slowdowns, increase yield, and improve overall product quality.
  • Identifies defects faster: Computer vision solutions find defects faster than their human counterparts, allowing manufacturers to locate and resolve defects before they leave the factory and become more expensive problems.
  • Boosts inspection accuracy: Compared to human inspectors, computer vision technology consistently achieves more accurate inspections with finer-grained detail and fewer inspection errors.
  • Reduces defect-related costs: Better defect detection with computer vision allows manufacturers to dramatically reduce defect-related costs caused by reputation damage, reduced sales figures, product liability litigation, insurance fees, replacement costs, transportation costs, repackaging, and marketing/PR activities.
  • Reduces downtime: Fewer delays in production and reduced downtime related to fixing recalled defective products.
  • Reduces scrap: Reduces scrap resulting from defects being found too late in the production process.
  • Reduces labor and training-related costs: Completes the same inspection tasks for a fraction of the cost of human labor. Eliminates the need to hire and train new employees. Frees up existing employees to focus on more important tasks.
  • Automates monotonous visual tasks: Frees employees from needing to complete monotonous visual tasks and avoids the negative consequences of work-induced boredom.
  • Facilitates immediate scaling: Scales instantly to satisfy the needs of a growing manufacturing business without the bottleneck of hiring/training new quality control inspectors.
  • Augments human capacity: Improves the capacity of human quality inspectors to perform tasks more successfully, efficiently, and productively.

How Computer Vision for Defect Detection Works

Computer vision systems for defect detection use strategically placed high-resolution cameras to capture video or image data of manufactured items. They then use sophisticated AI models to evaluate the visual data and deliver instant insights, metrics, and defect alerts to key employees and decision makers. This allows the end-user to identify and replace defective products more efficiently and successfully, thereby avoiding the costs and workflow inefficiencies that manufacturing defects cause.

Every visual AI system requires a “base training” – a process that involves capturing hundreds of thousands of annotated images. By using advanced tools to automate most of the image/video annotation process, end users can quickly provide the visual data that the AI models need to distinguish between a “defective” and “non-defective” product while identifying the various types, classifications, and severities of defects. Later, end users can rapidly finetune the training of existing visual AI models with additional data for new and different use cases.

When using the Chooch Enterprise Platform for defect detection, end users can generate a custom, lightweight visual AI model and deploy it to the cloud or to local “edge devices” at the manufacturing facility. End users can manage their deployments on existing high definition cameras through a cloud-based dashboard. Scaling these systems to other facilities and updating the models to apply to slightly different use-cases is a fast and straightforward process.

As a final note, deploying visual AI defect detection systems to a local edge device offers several key advantages. First, it allows all data and processing to stay onsite without touching the internet, so there are fewer risks of data breaches and greater security. Second, for mission-critical, 24/7 applications, edge devices offer greater resilience, better bandwidth, faster processing, and a reduced risk of downtime. On the contrary, there are many situations that do not require a high degree of data security, bandwidth, or speed. In these cases, cloud-based deployment may be sufficient to meet the manufacturer’s defect detection needs.

Common Use Cases for Defect Detection

The flexibility and scalability of the Chooch Enterprise Platform means that clients can deploy and redeploy visual AI defect detection strategies in a wide variety of industries – by making quick adjustments to pre-existing models. The following use case examples illustrate the versatility of Chooch’s computer vision solutions for defect detection in manufacturing.

General Use Cases

The table below shows additional use cases for visual AI defect detection in the manufacturing industry:

Visual AI Use case for Defect Detection

Products Potential Defects
Nonferrous Metals Wires, cables, aluminum, stainless steel Scratches, cracks, dirt, dents
Building Materials Wood boards, sashes, metal fittings, tiles, other materials Scratches, cracks, surface defects, dents
Electronic Parts Electronic materials, electronic components, circuit boards, electrical panels, other items Scratches, chips, cracks
Auto Parts Material parts, resin parts, fabrics, other materials Scratches, dents, dirt, cracks
Raw Materials Chemical fibers, rubber, glass, paper, pulp products Scratches, cracks, dirt, dents
Food Processed foods, beverages, food packaging, bottling Foreign objects, labeling errors, leaks, packaging damage, missing bottlecaps
Medical Pharmaceutical medicines, medical devices, surgical equipment, wound dressings, syringes, other items Foreign objects, labeling errors, cracks, defects, dirt, impurities, sanitary issues

Specific Use Cases

  • Industrial manufacturing: Industrial manufacturing facilities must adhere to industry standards and government regulations while providing defect-free products to their customers. Visual AI solutions can detect the most important industrial manufacturing defects such as cracks in casings, industrial welding errors, missing parts, broken products, scratches, structural integrity problems, and painting errors.
  • CPG manufacturing: Consumer-packaged goods (CPG) manufacturers also need to adhere to industry standards and government safety regulations to provide quality products to their customers and prevent product related injuries and lawsuits. Visual AI systems can assist CPG manufacturers to identify defects like spoiled or unsightly food products, packaging defects, missteps in the packaging process, labeling inaccuracies, misaligned bottle caps, misshapen products, leaking products, and broken glass.
  • Pharmaceutical and medical device industry: Pharmaceutical and medical device manufacturers adhere to the highest safety and cleanliness standards that exist because the failure to detect defects and errors can lead to disastrous consequences. Pharmaceutical manufacturers can use computer vision to monitor every step of the manufacturing process to ensure the use of correct compounds and ingredients, and to ensure that all products are appropriately labeled and safe for human consumption. Medical device manufacturers can also use computer vision to catch defects like misshapen products, cracked or broken products, or unsterile products.
  • Airline industry safety: According to a report from Boeing, the airline industry spends about $40 billion annually on inspections and maintenance of jets, engines, and related equipment. These expenses relate to “the costs of the labor and materials required to perform servicing, repair, modification, restoration, inspection, test, and troubleshooting tasks during on-airplane and shop maintenance activities.” Visual AI defect detection systems dramatically increase the accuracy and efficiency of airline industry inspections – offering a better way to detect problems before they become dangerous or expensive.

In-Depth Use Case Example #1: Misplaced Bottle Caps

Beverage factories need to monitor and inspect their production lines for misplaced bottle caps. Due to the human limitations of quality control inspectors, they need to have several tiers of inspection controls beyond real-time inspections as bottles pass through the production process. These controls typically involve additional human inspectors reviewing video footage of bottles to identify cap placement errors.

These secondary quality control measures are far from perfect. Often, by the time they find a misplaced bottle cap, the defective bottle has already left the factory. The tremendous efforts required to fix the problem can result in the destruction of 1,000 more bottles of product, additional transportation costs to return and replace the bottles, and other logistical costs.

Why does a single bottle cap defect result in so many costs? For most beverage manufacturers, it is less expensive to dispose of an entire lot of product than to try to find the individual box – among countless boxes – that contains a single defective bottle. Product manufacturers may also need to recall delivery trucks, ask retailers to hunt for defective products and incur other costs related to tracking down the misplaced or missing bottle cap.

The problem gets worse if the secondary controls fail to identify the defect before the bottle makes its way to the customer. One customer photo of a moldy, capless bottle can go viral on social media and result in lasting reputational damage, reduced sales, and increased marketing/PR expenses related to rebuilding consumer trust in the brand.

With the Chooch Enterprise Platform for defect detection, CPG and industrial manufacturing businesses now have access to advanced defect detection technologies that far exceed human ability – and they can develop and deploy them in a matter of weeks. Chooch visual AI models (running on locally installed edge devices) capture and review live video from cameras installed along production lines. Then, the AI models instantly interpret the visual data, sending their conclusions, reports, and defect alerts to staff who can pull defective products from the factory line before they are packaged into boxes.

In-Depth Case Example #2: Aircraft Parts Manufacturing and Maintenance

A large-scale aircraft parts manufacturing and/or aircraft repair operation needs to employ expert mechanics to visually review several points of inspection for specific aircraft parts to ensure the proper functioning of equipment and identify potentially dangerous defects. It is not unusual for the visual inspection of a single aircraft part to require the following:

  • 4 expert employees who separately review the same visuals (or points of inspection)
  • $100 an hour in wages per inspector 113 visuals (or points of inspection) to review per aircraft part
  • 80 seconds to inspect each visual (or point of inspection) to determine if the part meets inspection standards
  • 10 person-hours (36,160 seconds) to complete the inspection of one part
  • 2.5 hours total to complete the job with 4 inspectors working concurrently
  • $1,000 total inspection labor cost to visually inspect a single aircraft part

When computer vision technology performs aircraft part inspection tasks, no human labor hours are required. In contrast to the above cost and efficiency figures for employing a team of human inspectors, here is what a visual AI inspection system can achieve:

  • Zero expert human employees required to carry out visual inspection tasks
  • $0.001 cost per inspected image*
  • 0.2 seconds of processing time for each visual (or point of inspection) compared to 80 seconds per visual for human inspectors
  • Concurrent scaling of visual AI inspection processes so that a visual AI system can process all 113 points of visual inspection at the same time
  • 0.2 seconds to complete the entire part inspection process
  • $0.11 total processing cost* to complete the full inspection (note that additional training, equipment, and set up costs apply to developing a computer vision strategy of this type).

*Based on theoretical projections.

Comparing the cost of using visual AI to the cost and efficiency of human inspectors, the ROI benefits are clear. A single part inspection by human mechanics requires four skilled employees working for 2.5 hours each at a price of $100 each per hour. This comes to 10 human hours and a total operational cost of $1,000 per inspected part. Meanwhile, a computer vision model for defect detection can complete the same part inspection task in 0.2 seconds without any human labor.

Chooch Defect Detection and Pre-Trained Computer Vision Models

Chooch’s computer vision platform allows end-users to manage pre-trained visual AI models for defect detection – and the use-cases are practically endless. These pre-trained models are available now for immediate integration into manufacturing workflows. They can provide real-time defect detection insights, reports, and alerts for the following use-cases:

  • Misplaced or misaligned bottle caps
  • Industrial weld errors
  • Improper packaging
  • Labeling errors
  • Paint issues
  • Broken parts
  • Food quality issues
  • Misshapen products
  • Custom use-cases as required

As for compatibility features, the Chooch Enterprise Platform generates inferences that consist of a .json and .jpeg file. These inferences are compatible with virtually any business analytics platform, on-premises, or cloud based system through API or MQTT broker. The Computer Vision platform can also notify key personnel and decision makers through text message and email alerts as soon as a defect is found.

Finally, the Chooch Enterprise Platform includes an easy-to-use dashboard with the following features:

  • Visual, point-and-click controls that allow non-technical users to manage and customize computer vision deployments
  • Easy management, downloading, and viewing of defect detection models based on camera, edge device, or location
  • Instant deployment of lightweight containerized computer vision models to edge devices at different locations
  • Easy controls for managing video networks spanning multiple locations – like manufacturing floors, logistics centers, warehouses, mining operations, and construction sites.
  • Temporal, or time based, rules engine to create business impactful alerts and minimize alarm fatigue.

About Chooch

Founded in 2015, Chooch is a Silicon Valley-based computer vision company that offers a ready-now visual AI platform. Chooch solutions accurately detect and evaluate objects and actions in images and video for a wide variety of use cases – including defect detection, diagnostic medical analysis, PPE safety monitoring, employee/visitor fall detection, fire/smoke/heat detection, maintenance/repair inspections, and many other scenarios.

Deployable in the cloud or on the edge, the Chooch Enterprise Platform generates data sets, rapidly trains new models, and provides end-users with instant and accurate inferences. Chooch solutions have been successfully deployed in the widest range of verticals, including the U.S. government, aviation, media, geospatial, healthcare, and security/IoT. Chooch enterprise partners include NVIDIA, Deloitte, AWS, HPE, DELL, Lenovo, and ADLINK.

If you would like to learn more about Chooch and its range of Visual AI solutions, please visit www.chooch.com for more information.