Machine-Learned Process Control for Fast Response and Low Overshoot

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing control systems, such as those using PID control, face challenges in achieving high-speed control with minimal overshoot, especially in complex systems like industrial plants and process control.

Innovation Solution

A learning processing apparatus and method that utilize machine learning to generate a control model, which outputs a manipulated variable corresponding to an indicated variable and a process variable. This control model is used to create controlling data that maps combinations of indicated and process variables to manipulated variables, enabling efficient control.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If PID control is used for temperature control, then the control system is simple and reliable, but the response speed is slow and overshoot cannot be minimized

Engineering Contradiction:

Improveresponse speedVSAvoidcontrol system complexity

Core Design Contradiction:

SpeedVSDevice complexity

Solution Approach 1:

The patent replaces traditional PID control algorithms with machine learning-based control models that have been trained to optimize response speed and minimize overshoot. The control model learns optimal control strategies from historical data and applies them in real-time, achieving faster response without proportional-integral-derivative calculations.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms control parameters from fixed PID values to dynamic values generated by the control model based on learned patterns. The control model outputs manipulated variables that adapt to changing system conditions, enabling optimized performance across different operating ranges rather than relying on static parameter tuning.

Inventive Principle:

Principle #35Parameter changes

2Loss of time

If traditional control methods are used, then the control logic is straightforward, but overshoot is significant and regulation time is long

Engineering Contradiction:

Improveregulation timeVSAvoidcontrol precision

Core Design Contradiction:

Loss of timeVSManufacturing precision

Solution Approach 1:

The control model is trained in advance on historical operation data to learn optimal control strategies before actual control is needed. This preliminary learning phase enables the model to predict and prevent overshoot conditions, reducing regulation time and improving precision during actual control operations without requiring complex real-time calculations.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The control model incorporates feedback mechanisms where the actual process variable is continuously compared with the target value, and the model adjusts manipulated variables based on learned relationships. This feedback loop, combined with the model's learned understanding of system dynamics, enables faster convergence to target values with minimal overshoot.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12253835B2Learning processing apparatus, control apparatus, learning processing method, control method, non-transitory computer-readable medium having learning program recorded thereon, and non-transitory computer-readable medium having control program recorded thereon

Publication Date: 2025.03.18 YOKOGAWA ELECTRIC CORP

AI summary

There is provided a learning processing apparatus comprising: a learning processing unit configured to generate a control model that outputs a manipulated variable corresponding to an indicated variable and a process variable of a predetermined system by means of machine learning; a generation unit configured to generate controlling data that indicates a correspondence relation of a combination of the indicated variable and the process variable to the manipulated variable corresponding to the combination by using the control model; and a supply unit configured to supply the controlling data to a predetermined control apparatus.