YOKOGAWA's Reinforcement Learning AI
The industrial AI platform and YOKOGAWA's reinforcement learning AI are now available as a package. YOKOGAWA's Reinforcement Learning AI is characterized by being able to learn in a small number of trials and being resistant to disturbances. It frees you from adjustments that rely on intuition and experience, and contributes to process automation.
Overview
YOKOGAWA's reinforcement learning AI, which succeeded in autonomous control of plants using AI for the first time in the world*, can now be implemented with the e-RT3 Plus.
The industrial AI platform and YOKOGAWA's reinforcement learning AI are now available as a package. We provide an environment where customers can work independently on their own AI control. The AI control learning service is provided in a cloud environment, so you can always use the latest AI control learning service. It is possible to reduce the need for new infrastructure development and system construction.
* Based on Yokogawa Electric survey conducted in February 2022 regarding AI that directly changes the manipulative variable in the chemical plant.
Features of YOKOGAWA's reinforcement learning AI
With PID control which widely used in control, high-precision control can be achieved when conditions are straightforward, but becomes difficult when there is a large disturbance in the control state or when there is a long dead time. In many cases, adjustments must be made based on intuition and experience. AI easily utilizes empirical knowledge through learning, and can frequently derive answers even if a theory has not been established.
YOKOGAWA's reinforcement learning AI “Factorial Kernel Dynamic Policy Programming” (FKDPP)* is very practical because it requires fewer trials than general reinforcement learning.
* This is a reinforcement learning AI algorithm jointly developed by Yokogawa Electric and Nara Institute of Science and Technology (NAIST), and was the first in the world to be recognized by IEEE International Conference on Reinforcement Learning.
- Optimal control according to customer control goals is possible
- Learns after a small number of trials
- Applicable to general control
Expected benefits of AI control
- Securing the Future Workforce and Shifting to Automation
By automating process*1, which required human coordination, with AI, we will solve the labor shortage at the site.
*1: A process that relies on the intuition and experience of skilled engineers / A process that requires human adjustment every time a disturbance occurs - Contribute to improving control
It is possible to control with suppression of overshoot. It is possible to reduce the load on the equipment by improving control. - Both productivity and energy saving
By control that shorten the settling time, it is possible to secure productivity (quality, yield, etc.) while reducing wasted energy.
详细介绍
Product Specifications
| Category | Name | Model |
|---|---|---|
| CPU module | F3RP70-2L | OS-Free CPU module |
| License | SFRL18-MPC | AI control learning service for F3RP70 |
| SFRL19-MPC | AI control license for F3RP70 | |
| Software Package | SFRM19-MDW | AI control software package for F3RP70 |
* Purchase this product if you plan to use your existing OS-Free CPU module.
For new purchases, we offer the F3RP70-2L/L09 OS-Free CPU module with an AI control license.
 |
*1:AI control learning does not guarantee the generation of the best control model.
The performance of a control model varies depending on the characteristics of the equipment and the quality of the data provided.
To use the AI control learning service, please apply*2 at our dedicated website. After applying, a user account will be created in the cloud environment and you can begin using the service. The service is an annual subscription. To create an AI control model, you need to prepare either an actual machine or a simulator. If employing a simulator, you can create one using the simulator*3 function included with the AI control learning service.
*2: The product serial number shipped from the factory is required to apply. Please contact us for the sales area, as it is necessary to confirm the use of the cloud service.
*3: The simulator corresponds to single input and single output per the system identification method. If the controlled equipment has multiple inputs and outputs, it is necessary to prepare a separate simulator.
 |
 |
|
| Simulator
Conditions can be automatically set to easily generate control models |
AI control learning Generated control models are displayed graphically for review |
- AI control learning service and AI control software package sites will be released in conjunction with product first delivery scheduled for May.
- About AI related products : Please go to AI Product Solutions page.
Application
In combination with proven industrial controllers, our AI control can be used in a wide range of applications, including plant and machine control.
- Supports control periods from 0.01 seconds, and equipment control that requires high speed
- Multiple control loops can be processed in a single controller
- Hybrid control that takes advantage of the features of P ID and other conventional control methods
Main markets
- Resources and energy (petroleum, chemical, natural gas, electric power, renewable energy, etc.)
- Food and agriculture
- Materials (textiles, pulp and paper, paints, etc.)
- Pharmaceuticals
- Electronic equipment (semiconductor manufacturing equipment, etc.)
- Water and sewage systems
 |
Application examples
Temperature, pressure, water level/flow control etc.
[Demonstration] Furnace heating control
Overshoot reduced by a fifth and settling time shortened by a third
 |
Application examples
 |
Point!
- While ensuring quality, overshoot (which is difficult to adjust) is suppressed while reducing dead time
- Contributes to energy saving by improving efficiency
- Improvement of transient characteristics prevents unnecessary load on the heater and contributes to extending the life of the equipment
新闻
-
新闻 2023年2月27日 横河电机推出用于边缘控制器的自主控制AI服务
- 优化控制，提高生产率，节约能源 -
想要了解更多的信息，技术&解决方案？
联系我们