Building AI without Programming for Improving Productivity at Factory

See How to use the “Learning Center” from examples

In recent years, more and more companies started to implement digital tools to improve productivity. One of the key features of the “Learning Center” provided by AI inside is that people in charge of the site can create AI themselves with their best knowledge and understanding. With this, companies can implement AI best matching their business issues.

However, is it possible for someone without expert knowledge or experience in programming to create AI for resolving business issues? How can it be included in the operation? In this article, we use the development example of AI by Dai Nippon Printing Co. (DNP), our collaborative partner of the “Learning Center.”

“Learning Center” and “Design Thinking”

The “Learning Center” is a service in which anyone can create highly accurate AI models with no programming. It’s possible to develop AI at a low cost and in a short period without requesting AI vendors or developers. This allows users to create the AI based on their issues, and it can be used regardless of industry or sector.

Steps for creating AI on the “Learning Center”

Still, many find it challenging to have a concrete idea of what AI can do. This is where we use “Design Thinking,” to come up with ideas on what can be automated and made efficient with AI, as well as how to implement them. In our previous article, we shared details of our workshop where we used design thinking in defining requirements for creating AI, so please take a look.

Using AI to Understand Retention in Factories and Lead to Improvements

After this design-thinking-based workshop, DNP decided to create an “AI for Congestion analysis in the factory” to understand the retention points and the factory’s situation. The goal is to use AI to visualize what bottlenecks are causing extra time to wait for something in the factory and optimize the flow of people and things.

For the case of DNP, the time waiting for the elevator within their factory was an issue. So we started developing an AI to understand the details of the issue such as what is taking time and how much waiting time occurs at what time.

By just labeling with intuitive control, complete the preparation for training

Once it’s decided on what kind of AI to develop, the next step is to collect data (images) to create the data for training, which is used to train the AI. After collection, the images are uploaded for training to the “Learning Center,” then execute “annotation”. “Annotation” is the part where we label the images with the people and things we want the AI to recognize. The “Learning Center” has an instinctive user interface that allows users to add labels with simple input and clicks.

This time, we extracted images from videos taken by cameras installed in front of the elevators. We labeled each image as “Worker,” “Handcart,” “Palette,” and other people and things that pass in front of the elevator.

Image provided by: Dai Nippon Printing Co.

Click the “Start Training” button, and the AI will automatically train

Once the annotation work is done, by simply clicking the “Start Training” button, the AI will automatically start training through our deep learning network. Users can check the progress of the training through the “Learning Center.”

Annotation screen on the “Learning Center”

The accuracy of the AI model done with training can also be analyzed in the “Learning Center.” To improve the accuracy to the level required for business,

Users can add images and continue training, to improve the accuracy to the level required for business. One of the appealing features of the “Learning Center” is that users can easily update the AI themselves.

The “AI for congestion analysis in the factory” created by DNP was annotated on 500 images. The person in charge of the site took half a day to process this. After taking over one day to train the AI, we were able to achieve over 90% accuracy.

Integrate AI in worksite

Afterward, to start using the AI after additional training and improvement of the accuracy, an API key will be provided on the “Learning Center.” By using this API key, it’s possible to embed the AI into on-site systems. In the future, we aim to link other existing systems aside from the API key with just a few clicks.

By linking the “AI for congestion analysis in the factory” to cameras in front of the elevators in the DNP factory, we can now understand more details of the waiting time in data such as when, what, and how much time. With this data, DNP will work on improvements within the factory.

For the implementation of AI, up until now, it was required to request specialists such as AI vendors or system development departments, which often resulted in large projects costing tens of millions of yen and taking years to complete.

The “Learning Center” can be used free of charge from issuing accounts until the creation of training data. After designing the AI model, a fixed fee is charged for training (create) and deployment (use) for each month’s use. Training can be used for 100,000 yen per month, and deployment for 30,000 yen per month, leading to significant cost reduction from conventional AI development. In addition, if the AI is simple, it can be created and implemented within one to two weeks, making it easier than ever to use AI to improve business.

“Learning Center” can be used in a variety of situations

In this article, we shared the example of how the “Learning Center” was used to improve production at DNP’s factory. The current “Learning Center” is capable of creating AI for object detection using image recognition technology, and excels at operations such as “checking with the visual confirmation.” For example,

  • AI that determines the damaged areas from images and videos of buildings and vehicles.
  • AI that identifies the shape and number of items in a shipment and automatically inspects them
  • AI that supports diagnosis based on the type of condition of diseases

and other varieties of AI.

Furthermore, aside from object detection, we are planning to incorporate AI for natural language, voice recognition, and other applications, which will continue to expand the range of use. AI inside also plans to build a marketplace to share AI created in the “Learning Center.”

In the future, when improving business, how about considering AI for business efficiency, just like people consider using Excel? That’s how much AI has become a familiar presence to us. As the first step to familiarize AI more and for the “Learning Center” to be used by more people, we will continue to share application examples. Please look forward to them!

Contact for “Learning Center”:
AI inside Inc. (https://inside.ai/en/contact/contact-form/)

This is the official account of AI inside Inc.