Preface
Welcome back to AI Manufacturing. In the previous article, we discussed some of the challenges facing the manufacturing industry and outlined relevant solutions. In this article, we will delve into the practical operations in the cloud and present architecture diagrams, along with explanations and introductions to the cloud products involved. It is hoped that through this architecture showcase, readers can gain a deeper understanding of the advantages brought by the cloud and how cloud technology can inject new vitality into traditional manufacturing.

To start, here is an architecture diagram for our readers:

In the previous article, it was mentioned that in this case, cameras were installed in the grinding factory to capture real-time photos of multiple faces of the steel material. These photos were then transmitted to the cloud platform's object storage via dedicated lines or the internet.

Next, manual custom labeling was performed to define categories such as defective and non-defective for the photos. These labeled photos were then used to train machine learning (ML) models using cloud-based ML tools. Once the model training was completed, it was deployed to cloud servers. Subsequently, all photos awaiting judgment could be transmitted over the network to the cloud servers for classification, and the results were sent back to the operators' client screens.

The author conducted a pros and cons analysis of this architecture as follows:

Pros:

1. Most hardware resources are deployed in the cloud, saving deployment time and maintenance costs.
2. Machine learning can be trained using pre-built AI models and customized adjustments can be made according to specific needs.
3. The factory side only needs to introduce automated camera equipment such as cameras and connect them to transmission lines.

Cons:

1. Requires stable transmission lines to avoid service interruptions.
2. Higher latency may result in longer waiting times.

So, are there any alternative solutions? Yes, please refer to the diagram below:

As shown in the diagram above, this architecture also requires the installation of cameras, and the captured photos need to be transmitted and stored in the object storage bucket on the cloud platform. The model training process is carried out using the cloud platform's machine learning tools. Once the model training is completed, it is directly deployed on edge devices at the factory. During the decision-making process, relevant decision results can be collected and feedback can be provided to optimize the model.

The pros and cons of the above architecture are as follows:

Pros:

1. Lower network configuration requirements.
2. Improved computation and result return speed.

Cons:

1. Longer time required for updating decision-making models.
2. Additional costs for purchasing edge server hardware.
3. Costs associated with hardware maintenance personnel.

Both of the above cloud architectures use the cloud platform's tools to establish and train AI models. The difference lies in whether the decision-making model is deployed "on top of the cloud platform" or "on edge devices at the factory."

The subsequent effects will also vary. If constructed on top of the cloud platform, it will save a considerable amount of manpower and hardware costs, but there may be slight delays in network transmission, which could reduce decision-making speed. If the model is trained using cloud platform tools and then deployed to edge devices, network transmission speed will be improved, but additional hardware costs will be incurred.

There is no one-size-fits-all solution. The best solution is the one that fits the specific needs. The two solutions provided above are for reference, serving as a starting point for further exploration. We look forward to leveraging the power of the cloud in the future of AI smart manufacturing, working together to achieve the vision of the AI digitization era.