Variable Damper Suspension Control With ML Instruction Limiting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional suspension control methods, such as skyhook control law or BLQ, are based on linear or bilinear systems and may not be optimal. Additionally, machine learning-based control methods lack clarity in the control instruction derivation process, making it difficult to verify the correctness of the learning results, which can lead to vehicle state instability due to learning errors or random failures.
Innovation Solution
A suspension control apparatus that includes a first instruction calculation portion using machine learning to output a first target amount based on learning results, and a control instruction output portion that limits and outputs the control instruction if the first target amount leads to a greater vehicle state amount than a predetermined amount.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning-based control is used to improve control accuracy, then control accuracy is improved, but reliability deteriorates due to unclear verification process and potential learning errors
Solution Approach 1:
The control instruction output portion monitors the vehicle state amount after applying the control instruction derived from machine learning. When the vehicle state amount exceeds a predetermined threshold, the system feeds back a corrected control instruction to prevent instability, creating a closed-loop verification mechanism that maintains reliability while using ML for accuracy
Solution Approach 2:
The control instruction output portion acts as an intermediary between the first instruction calculation portion (ML-based) and the force generation mechanism. It verifies the control instructions by checking whether they lead to excessive vehicle state amounts, and corrects them if necessary, thus mediating between ML accuracy and system reliability
2Productivity
If machine learning control instruction is directly output without verification, then productivity is improved by simplifying the control process, but reliability deteriorates due to inability to verify learning results
Solution Approach 1:
Instead of implementing complete verification of the machine learning process, the system applies a simplified verification by checking only whether the control instruction leads to excessive vehicle state amounts. This partial verification approach maintains high productivity while providing sufficient reliability assurance for critical safety parameters
Data Source
AI summary
A controller controls a suspension apparatus including a variable damper that adjusts a force between a vehicle body and a wheel of a vehicle. The controller includes a first instruction calculation portion and a control instruction output portion. The first instruction calculation portion outputs a first instruction value of a damping force, which corresponds to a first target amount, using a learning result acquired from machine learning in advance by inputting a plurality of different pieces of information. The control instruction output portion limits the first instruction value and outputs it as a control instruction in a case where the first instruction value works in a direction leading to a greater vehicle state amount than a predetermined amount due to control of the variable damper based on the first instruction value.