How to Build Computer Vision Solutions for Industrial Applications

This blog discusses building computer vision solutions at scale for manufacturing and industrial facilities.

How to Build Computer Vision Solutions for Industrial Applications

Computer vision is the field of artificial intelligence that allows algorithms to interpret visual inputs, such as digital images and video data. With the ability to rapidly analyze and derive insights from vast troves of data with fewer errors, computer vision is already transforming manufacturing and industrial facilities. Today, computer vision is being integrated into visual inspection systems, safety and security monitoring, inventory tracking and supply chain monitoring to improve quality control, optimized processes, and enhanced operational efficiency.  

The increased adoption of computer vision is part of the digitization of Industry 4.0. To transform the factories of today into the smart manufacturing facilities of the future, manufacturers must be able to quickly process the projected 75% of enterprise data that will be created at the edge in 2025. Doing this requires building the digital fabric between their information technology (IT) and operational technology (OT) assets, which is heavily dependent on being able to analyze data at its point of collection in a timely manner. This means that manufacturers need computer vision insights at the edge.  

This blog will explore the trends driving computer vision to the edge in manufacturing. We’ll discuss the current challenges with building the connection that facilitates AI insights at the edge and examine a better way to integrate AI at scale to drive better data-driven decisions. Finally, we’ll examine some use cases and examples of how computer vision can be used to power smart manufacturing solutions for the future.

Industry 4.0 and Edge Computing: Challenges for Manufacturers

Edge computing enables data processing and storage at the point of data collection, which yields faster insights, lower latency, and better security. Edge computing can enable real-time decisions when needed because data doesn’t have to travel to/from the cloud before the next step in the process, and 5G enables better connectivity for sensor-driven analysis. Edge computing is becoming more mainstream, and McKinsey research predicts that IoT will be a $250B market by 2025 as sensor cost reduction, data storage cost reduction, device ubiquity, and better connectivity drive increased demand.

A single manufacturing facility could have hundreds or even thousands of different IoT assets generating data that must be analyzed and communicated rapidly to execute critical functions and processes. But in the US, the average age of industrial assets is close to 23 years, a figure that is only increasing according to the Bureau of Economic Analysis. Updating and adding cutting edge technology like AI to existing assets presents numerous challenges depending on the system or machine, regardless of existing compute or digital capabilities.

Integrating AI capabilities can’t take a one size fits all approach. One critical factor is the high variability in machine components, including different hardware specifications, processors, resourcing, network and bandwidth constraints, and power limitations. Another key consideration is data transfer between different devices and machines.

Manufacturers refer to projects that improve their existing machines and infrastructure as brownfield projects, and retrofitting existing factories with capabilities like computer vision presents a massive undertaking. It is within this context, coupled with an aging and shrinking workforce, that manufacturers are turning to solutions like computer vision to optimize their processes, enhance operational efficiency, and drive innovation towards Industry 4.0.

Building Computer Vision Solutions at Scale

Fortunately, going back to the concept of one size fits all, it’s possible to design an architecture that’s edge-centric and facilitates running computer vision models wherever insights are needed. This approach obfuscates the unique machine or infrastructure complexities. One of the biggest mistakes organizations make today is investing in a new IoT solution that is a viable solution to one problem – take a “smart camera” for example. At first glance, it might seem like a quick and easy way to add a new piece of tech and capability to a factory, but in reality, this represents yet another machine to add to the complicated web of systems that need to be connected in the future. There’s a better way.

What’s needed to lay the foundation that can connect computer vision models and insights to existing machines and infrastructure? First – forget about the infrastructure aspect at all. Today, requirements might dictate AI insights in the cloud and at the edge, and in the future, that could extend to hybrid cloud, on-prem, or any complex infrastructure combination. Instead, consider a solution that scales computer vision models to many locations efficiently, enabling the flexibility to build AI applications anywhere. This kind of solution should (1) include a central management hub for all computer vision models, (2) deploy to any machine or device, (3) perform low latency inferences to enable real-time insights, and (4) operate in a disconnected environment in an instance where a machine goes offline.

Prioritizing these four elements removes the dependencies and concerns mentioned in the previous section and lays a foundation that can support all computer vision solutions today, with the flexibility to support more in the future.

Running Computer Visions Models in Production Applications At-Scale

Diving a level deeper, the next step is understanding what is needed to run computer vision models, or any kind of machine learning model, in production, at-scale. Again, adopting an edge-centric approach can help tame the chaos and allow for accelerated solution development in a repeatable way.

Components of a scalable edge-centric AI system.
  • First, containerizing models provides the flexibility and portability to run on many kinds of machines and infrastructure. Containers store libraries, scripts, and data files in an immutable format; a container runtime like Docker is also needed to take advantage of model containers.
  • Next, these model containers should be uploaded to a model store, which makes it much easier to move models to different locations, maintain model versions, and disconnect from machines as needed. In the smart camera example, this would be impossible to do, because the computer vision algorithm is embedded in the camera itself as part of a closed, proprietary system.
  • Using a model management hub allows for greater flexibility, sharing, and reuse of models as needed.
  • Finally, a data transfer protocol like MQTT, REST, or gRPC can be used to enable high speed, low latency, and secure inference for the system. Protocols such as MQTT can serve as the messaging mechanism between different machines, allowing communication and the ability to distribute telemetry information with low overhead and bandwidth consumption, flexibility, scalability, and easy implementation.

Use Cases and Examples for Computer Vision in Manufacturing

The main value in adopting this architecture for AI systems is that it can be used to support the development of any computer vision solution for any use case, all managed from a central location. Computer vision based visual inspection systems are more accurate and reliable than human inspection systems, increase speed and efficiency for defect detection, reduce costs and waste, and can proactively mitigate maintenance procedures to reduce asset downtime and improve operational efficiency and output. All these solutions can be used to improve product quality, improve operations, and yield better outputs.

In addition to these top-level benefits, centralizing how computer vision-enabled applications are built can enable faster development of solutions, cloud cost savings, and more. By using a solution like Modzy to build intelligent applications at scale, it’s possible to build, update, and maintain any kind of computer vision solution:  

  • Image Recognition and classification: Broadly speaking, image recognition and classification is used in applications such as automated surveillance, autonomous vehicles, robotics, and industrial quality control. It can be used to perform tasks like quality control, object tracking, robotic guidance, and inventory management for industrial automation, and improves efficiency, accuracy, and productivity in manufacturing and logistics operations.
  • Visual inspection and defect detection: Computer vision can be used to automate visual inspections on production lines by analyzing images or videos in real-time to identify defects, measure dimensions, or detect assembly errors. Automated visual inspection and defect detection allows manufacturers to maintain high-quality standards and prevent faulty products from reaching consumers.
  • Predictive maintenance: Computer vision can be used for predictive maintenance to monitor and analyze equipment and machinery conditions, identify potential faults or failures, and enable proactive maintenance actions. By proactively monitoring equipment conditions, it can be used to reduce equipment downtime.
  • Worker safety monitoring: Computer vision can be used to detect safety hazards and unsafe workplace behavior, like detecting the absence of PPE in image or video feeds. Faces can even be blurred to anonymize data and protect worker privacy.
  • Inventory management and monitoring: Computer vision can be used to perform optical character recognition (OCR) analysis and scan barcodes or QR codes for automated inventory management. It can also be used to continuously monitor visual data to keep track of stock levels in real-time, while also identifying visual defects or other anomalies to ensure only products that meet standards are counted as part of inventory levels.
  • Supply chain monitoring: computer vision can be used to analyze data from cameras and sensors to track shipments as they move through the supply chain, monitoring things like packages, containers, and vehicles. It can also be used to monitor facility operations and monitor video feeds as goods move through a facility.
  • Video surveillance: Computer vision can be used to quickly analyze video feeds and detect anomalous activity, unauthorized access, improper handling, and more, allowing for faster alerting and response when needed.

Build Your Computer Vision Solutions with Modzy

Computer vision powers the intelligent edge for manufacturing, improving quality control, optimizing processes, and enhancing operational efficiency. Architecting a flexible, scalable system that supports the development of endless computer vision-enabled solutions can help ensure efficiencies in how computer vision solutions are built and maintained over time. Modzy is the platform to deploy, connect, and run AI models anywhere, supporting the development of intelligent applications for enterprise and the industrial edge. With 15X faster model deployment, APIs for easy integration, and automated governance and monitoring, teams count on Modzy to build AI solutions for endless use cases.

Want to learn how Modzy can accelerate the development of your AI solutions? Book a demo today.