The Potentials of AI Vision for Industrial Deployments

AI vision does not only replicate human vision but can also go beyond that in offering highly accurate accounts of environmental features that are not readily visible to the human eye. In industrial settings at the edge, Vision AI means high-end quality control of manufacturing tasks and enhanced automation capabilities.

While edge AI has been around for a while, enhancing edge capabilities with computer vision is still a novelty. Those who have ventured into improving production processes, safety, and quality with the help of AI vision, however, are already reaping the benefits.

When equipped with AI vision solutions, industrial enterprises can take full control of their assets on the edge and build a truly collaborative foundation for a multitude of use cases. This will allow them to tackle the challenges of a dynamic setting that includes many unknowns.

What AI Vision Can Do For You?

What makes AI vision so attractive for industrial deployments? It is not simply the high accuracy and adaptability of machine vision algorithms at the edge but also the sustainability of adopting computer vision applications. Below is a breakdown of what you can expect.

You only install the hardware once

Once you have installed and set up your smart camera system at the IoT edge and have connected to your platform, you are good to go. Taking it from there, you can use this solid foundation to build a variety of use cases on top of your existing solution, and add or switch between different Vision AI apps without disrupting ongoing operations.

No need to change existing assets

Using your existing legacy assets at the edge, you can use the platform to train, deploy, and improve on ML models without being dependent on specific types of hardware, IT specialists, or external providers. This means that your existing assets stay where they are – and you can use any hardware-agnostic platform to orchestrate assets within your device landscape.

No need to buy additional sensors for new use cases

Equally so, it is not necessary to buy new sensors or any new hardware if you want to build and test additional use cases. Once you have your hardware in place, you can easily adapt to changing demands by simply installing new apps (that is, machine learning models and whole artificial intelligence solutions packaged as ready-to-use lightweight IoT applications), testing, debugging, and updating from within one platform with a full overview of what happens.

You can add extra apps immediately or incrementally

Adding, removing, and improving applications is equally seamless. You can add multiple deep learning apps at once to track various performance parameters and serve a variety of use cases. And you can transition between different levels of complexity gradually, adding and testing one app at a time.

Handle complex environments

Platform-enabled Vision AI applications are at their best when they have to work in complex environments and capture a variety of issues that may evade human inspection. Using smart cameras installed at the edge, you can get Vision AI apps to inspect barcodes, check for assembly or packaging errors, detect safety issues, count people on the shop floor, and perform inventory management all at the same time, with high levels of precision.

Allows for seamless scaling

By using just one platform, you create a solid foundation for your use cases, starting with data import to ML model creation and deployment back to the AI edge. Thanks to containerized applications, testing and deployments within production environments can take place without disruption, minimizing downtime.  

The Case for AI Vision in Industrial Manufacturing

Just about every step of the manufacturing process can be enhanced and made more secure by implementing AI vision. Even more automation, real-time defect detection, and productivity tweaks are just a few of the possibilities. Below are some ideas of what you can achieve with computer vision.

Quality Assurance

Powered by AI vision, manufacturers can make sure processes are optimally orchestrated and executed. Waste reduction is at a minimum and the operating conditions on the shop floor are nearly ideal.

This makes for maximal visibility and less dependence on human expertise. Quality control is taken over by smart algorithms specifically trained to detect outliers, perform analyses, and send out alerts.

Root Cause Analysis

Computer vision at the IoT edge can help detect patterns that have eluded human observers. And it can implement complex action plans to tackle the ramifications of these discoveries. In unearthing the root cause for a specific phenomenon or behavior, manufacturers learn from errors quickly and avoid future mistakes.

When multiple errors occur, Vision AI applications installed at the industrial edge can trace the chain of events and provide important insights to help remedy the situation. Inspection via computer vision is especially valuable when looking for certain manufacturing defects across the entire value chain. Here, as well, AI vision proves to be more accurate than the human eye.

Full automation for complex visual tasks

Because manufacturing processes are inherently complex and dynamic, achieving even modest levels of automation has been a challenge throughout. With the backing of a comprehensive IoT & AI platform, Vision AI applications at the edge can respond to these dynamics. They continuously adapt to new variables.

New incoming IoT data can constantly deliver new insights and each AI model at the edge can be quickly adapted by rolling out instant OTA updates. This way, the automation cycle is continuously renewed without disruption.

What Are the Typical AI Vision Use Cases?

The capacity of industrial Vision AI technology to enhance quality assurance, root cause analysis, and industrial automation can go a long way. Specifically, computer vision at the edge, embedded into a complete infrastructure for the creation, deployment, and oversight of Vision AI applications, can take over a wealth of classic tasks in manufacturing, across the entire value chain.

Most AI vision applications use the visual data to train a machine learning model to do well-established things such as facial recognition, object detection, image classification on the basis of visual input from multiple sources, image recognition, visual inspection, image analysis, or even more sophisticated video analytics. Taking it from there, various computer vision solutions can be built. On the basis of the pre-trained model, companies can expand to creating custom models specifically tailored to their unique need.

Typical use cases include:

• Component assembly. These tasks can be now fully automated and managed without or with minimal human intervention.
• Packaging. Vision AI applications can oversee the entire packaging process and inspect the final packaged material for defects.
• Barcode verification. This task can be solved with a simple AI vision app installed on smart cameras. And it saves thousands of hours of manual inspection.
• Defect detection. Image processing capabilities – with IoT cameras installed at the edge – make it possible to achieve higher accuracy in defect analysis and prevention.
• Predictive maintenance. Instead of costly visual checks by human employees, equipment monitoring can be performed by computer vision systems. These will track specific metrics and send out alerts when outliers are detected.
• Safety protocols. Computer vision can take over the oversight of employee safety on the manufacturing venue, effectively reporting on changing environmental conditions and other health hazards, adherence to safety compliance protocols, and any untypical movement of people or machinery.

To All Developers: Creating Vision AI Solutions

You’re a developer and may wonder what is our take on Vision AI and how we integrate Vision AI capabilities into our IoT platform? In what follows, we sketch out the basics.

Solving the problem with containers

One way to get on top of the complexity of IoT development initiatives is to use containerized applications. This means that you will run your apps in lightweight units called containers. Containers allow you to run your applications in an isolated environment while having everything you need for these apps to run–including libraries, code, tools, and runtime.

This way, you are empowered to deploy and scale your applications in any type of environment.

As you separate your applications from your other infrastructure, you can also run multiple containers at the same time. Also, you can share containers.

At Record Evolution, we use Docker. Each app must have a Dockerfile to create a custom Linux configuration starting from a standard base image like ubuntu or Alpine Linux. Alternatively, you can add Dockerfiles for specific target system architectures “Dockerfile_armv7” or “Dockerfile_arm64” if you want to make your app run on armv7 (Raspberry Pi) or arm64 (Jetson) systems.

Using editable code

In the IoT development environment, you can edit the source code of your app using the web code editor. The code editor is your personal cloud IDE based on Microsoft’s VS Code. Code editing is based on the git workflow–the industry standard for creating maintainable and auditable code for collaborative development.

This means that every change you make to the code is only present in your personal environment until you commit and push your code to the master branch of your application.

The highlights:

• You build and run your code on a remote device during development
• You log into a running app container directly on a device to debug

Integrating with GitHub, GitLab, and Bitbucket

By default, each app on the platform has a central git repository. That repo is used as the common code point for all developers of the app. This central git repository is managed within the platform.

When creating an app, however, you can select another service. Currently, we support GitHub, GitLab, and Bitbucket.

You can directly build and push your GitHub app code into the IoT App Store and deploy on devices.

Also, you can move an internal repository to any of these external services at any time. This way, you can use your local editor setup and push to your GitHub repo, while remaining able to run and debug code on actual devices using the platform.

We look forward to joining forces with Vision AI developers.

‍