AI-powered machine vision for manufacturing and industrial environment monitoring applications

In traditional industrial and manufacturing environments, monitoring worker safety, improving operator efficiency, and improving quality inspections are all manual labor. Today, AI-based machine vision technology has replaced many inefficient, labor-intensive operations and improved reliability, safety, and efficiency. This article will explore how to further improve performance by deploying AI cameras, because the data used to enable AI machine vision comes from the camera itself.

AI-enabled machine vision

According to IoT Analytics, the AI ​​machine vision market in manufacturing and industry was valued at approximately $4.1 billion in 2020 and is expected to grow to $15.2 billion by 2025, a compound annual growth rate (GAGR) of 30%, compared to a 6.5% CAGR for traditional machine vision deployments. This high CAGR is due to the fact that the applications of next-generation real-time edge AI machine vision are not limited to quality assurance and product inspection applications.

Worker safety is a top priority in manufacturing and industry, and AI-enabled smart cameras help automate monitoring and inspection in these environments. The safety of employees, contractors, and other third-party operators working in potentially unsafe environments, such as dangerous machinery and hazardous materials, must be ensured. Behavior and position (POSE) detection generates information that can indicate whether machine operators are in danger, following standard operating procedures (SOPs), or working in a way that provides productivity and efficiency. Finally, automated optical inspection (AOI) can improve the speed and accuracy of quality control, even for hard-to-see products such as contact lenses.

AI helps smart worker safety

Fatalities caused by industrial settings are not unheard of around the world. Facilities must also consider non-fatal workplace injuries when evaluating worker safety. In addition to emotionally traumatic accidents, there are often factors such as economics that need to be considered.

Industrial and manufacturing industries often use human supervision and light curtains to ensure worker safety. However, humans cannot be everywhere and do everything, so there is a risk of error, and safety light curtains have their own limitations.

Electronic Fence

In modern smart factories, people often work around potentially dangerous equipment, such as robotic arms. Safety light curtains protect personnel from harm by creating a sensing screen at and around machine access points. However, they take up a lot of floor space and are difficult to deploy and lack flexibility. In some cases, safety light curtains have limited response times, which brings other problems.

Traditional machine vision solutions use flexible and easy-to-deploy IP cameras and artificial intelligence modules, but the latency is still relatively large, making them unsuitable for application scenarios that require immediate response.

Figure 1: Safety light curtains take up floor space, are difficult to deploy, lack flexibility, and are sometimes limited in their responsiveness. AI cameras minimize latency, reduce deployment space and bandwidth requirements, and are easy to deploy and maintain.

ADLINK's NEON-2000 series of all-in-one AI cameras can solve the problem of latency. It captures images and performs all AI-related operations before sending results and instructions to related equipment (such as a robotic arm) (see Figure 1). Compared with light curtains and traditional machine vision facilities, the use of all-in-one smart cameras can minimize latency, reduce deployment space and bandwidth requirements, and are easy to install and maintain.

Real-time machine vision AI provides benefits for enhancing worker safety by alerting workers to enter unsafe areas and recording that information for retraining workers. Data recorded at past times may also be helpful in the future. For example, if a worker approaches a hazardous area, the robot arm does not need to shut down completely, but enters a functionally safe process loop. Routines such as this can not only improve worker safety, but also improve factory operational efficiency.

Smart refueling

When a refueling truck arrives at a manufacturing plant, it may pose many safety hazards, which can be easily solved with intelligent AI vision. First, if the brakes are not applied correctly or fail, it may cause the vehicle to roll over. The AI ​​machine vision system is trained to monitor the movement of the vehicle and can immediately issue an alarm when its state changes.

Facilities must also consider the location of operators during the refueling process, as there are different types of zoning violations. It becomes critical to ensure that all on-site workers understand the safety risks that exist. For example, it is necessary to place traffic cones at the four corners of the vehicle and ensure that operators refueling the vehicle are wearing appropriate personal protective equipment - AI vision can perform all safety checks to confirm that all processes are correct. (See Figure 2)

Figure 2: While it’s not always possible for a supervisor to be on-site to enforce safety procedures, AI machine vision can immediately sound an alarm if someone intrudes into a hazardous area.

Instant alerts from AI-powered machine vision systems can alert operators to safety breaches and prevent injuries. It also creates accountability; if someone enters an unsafe area without wearing personal protective equipment, the recorded image can flag the error and educate the employee to prevent future mistakes.

Behavior and location detection

For the manufacturing industry, “cycle time” is a key performance indicator for production efficiency. It represents the time a team spends on a production project until the product is ready to ship or before. Using AI camera technology to monitor employee behavior and location helps enforce standardized procedures (SOPs) and improve employee efficiency, shortening cycle time.

Figure 3: Behavior and position detection on electronics manufacturing lines helps increase productivity and improve the balance between orders, quantities, and lines.

Behavior and position detection from live video plays a vital role in overlaying digital content and information on the analog world. Behavior and position use a set of skeletal landmarks (such as a hand, elbow, or shoulder) to describe the position and movement of the body.

AI machine vision allows factory operators and workers to focus on how body position affects their work. Behavioral and positional data is a great training tool for operators on how to position their arms and hands to be more ergonomic and efficient; it can also improve people’s posture, another significant benefit. (See Figure 3)

Tracking whether operators are at their workstations on the production line can also be automated and schedules verified. Monitoring their active adherence to standard processes ensures quality management and line balance.

AI Smart AOI Smart AOI based on artificial intelligence

Manual inspection of product quality can be time-consuming and eventually lead to bottlenecks in the production line. Traditional AOI (automatic optical inspection) machine vision, with its superior accuracy and high efficiency, can detect easy-to-find product defects faster than professional quality control personnel. However, when defects are difficult to detect, such as those on contact lenses, these machine vision systems have difficulty meeting actual needs in terms of accuracy and consistency.

Although most manufacturers use random sampling to test whether products have defects, this method is not applicable on the contact lens production line because every lens needs to be inspected. Quality control personnel can only inspect a maximum of 4,000 lenses per shift, thus creating a production bottleneck. In addition, false detection and missed detection are inevitable.

Since contact lenses are transparent, the use of machine data for inspection has always been a major challenge for the industry. Traditional AOI relies on fixed geometric algorithms to find defects, but it is difficult to obtain high-quality images from transparent objects, resulting in inspection performance that is unacceptable to customers.

AI-based smart cameras collect data to train AI algorithms and continuously iterate on the performance of inspections to provide better solutions. AI-based smart systems can identify common defects such as burrs, bubbles, rough edges, particles, scratches, etc. (see Figure 4), and keep inspection logs for customer reference.

Figure 4: AI-based smart AOI can detect even tiny defects in transparent contact lenses, significantly improving inspection efficiency compared to the previous manual quality control process

Compared with manual visual inspection, each AI-based smart camera can inspect more than 50 times the number of contact lenses, and the inspection accuracy is improved from 30% to 95%.

in conclusion

By leveraging the powerful, real-time data generated by AI-enabled machine vision technology, manufacturers can gain more uptime, gain the ability to perform preventive maintenance, improve productivity, and ensure worker safety, among other benefits.

The AI ​​machine vision applications highlighted in this article require AI algorithms for deep learning. Software experts who develop AI algorithms need an intelligent and reliable platform to perform AI model reasoning. AI cameras pre-installed with EVA (Edge Vision Analytics) software solve many common problems with traditional AI vision systems, improve compatibility, speed up installation, and minimize maintenance.

To successfully deploy an AI vision project, engineers may need up to 12 weeks to conduct a proof of concept (PoC). Selecting optimized cameras and AI inference engines, retraining AI models, optimizing video streams, etc. all require a long learning period. However, EVA software, with its pipeline structure advantages, simplifies all steps and shortens the PoC time to less than 2 weeks, making it an ideal starting point for launching an AI vision project.