Monitoring Device Training for Learned Control System Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In monitored control systems with learned components, it is challenging to define and control the range of correct operation, especially when the value of adjustable control parameters is unknown a priori, making it difficult for monitoring devices to accurately evaluate the operation of action devices with multiple actuator modules and complex operations.
Innovation Solution
A monitoring device that communicates with the action device to receive command signals and sensor data, computes an operation indicator using learned values from iterative learning procedures, and employs a classifier to label and train on trial datasets to determine the correctness of the action device's operation, enabling accurate monitoring of action device performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning is used to learn control parameters iteratively, then the action device can perform complex tasks with optimized parameters, but it becomes difficult to define and control the range of correct operation
Solution Approach 1:
The monitoring device is trained in advance using trial datasets collected during the iterative learning process. This preliminary training enables the monitoring device to recognize correct operation patterns before actual monitoring begins, resolving the contradiction by preparing the monitoring system ahead of time rather than attempting to define correct ranges during operation.
Solution Approach 2:
The system collects trial datasets during the learning process and uses them to train the monitoring device. This feedback loop allows the monitoring device to learn from the action device's learning process, enabling accurate monitoring of complex tasks while maintaining reliability through data-driven training.
2Productivity
If the action device performs multiple task trials with different control parameter values, then the learned parameters improve task performance, but it becomes difficult to evaluate whether the operation is correct or abnormal
Solution Approach 1:
Trial datasets are collected during the learning process and used to pre-train the monitoring device. This preliminary action enables the monitoring device to establish baseline knowledge of correct operations before actual monitoring begins, making it possible to evaluate complex tasks with multiple parameter variations.
Solution Approach 2:
The monitoring device creates a model of correct operation by copying patterns from trial datasets. This model serves as a reference for evaluating whether current operations are correct or abnormal, enabling easy evaluation of complex tasks without requiring deep understanding of each parameter variation.
3Ease of operation
If the monitoring device uses traditional monitoring methods with predefined ranges, then it can easily determine correct operation, but it cannot accurately monitor action devices with learned parameters that are unknown a priori
Solution Approach 1:
The monitoring device trains itself using trial datasets collected during the action device's learning process. This self-service approach eliminates the need for manual definition of correct parameter ranges while achieving high monitoring accuracy, resolving the contradiction between ease of operation and measurement precision.
Solution Approach 2:
The system transitions from fixed predefined monitoring ranges to dynamic learned ranges derived from trial datasets. This parameter change allows the monitoring device to adapt to learned parameters that are unknown a priori while maintaining monitoring simplicity through automated training.
Data Source
AI summary
Monitored control system includes an action device and a monitoring device for monitoring the operation of the action device. The action device includes actuator modules, a sensor module and a control module to command the actuator modules to perform a predefined task using the sensor data. The monitoring device receive monitored data from the action device and compute an action device operation indicator indicative of a correct or abnormal operation of the action device on the basis of the monitored data. The action device operation indicator is computed on the basis of a learned value of an adjustable monitoring parameter. The learned value is determined from a set of labelled data generated from trial datasets recorded by the action device during an iterative learning procedure.