Machine Tool Thermal Displacement Estimation With Reliability Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning methods, despite providing high estimation accuracy under normal conditions, fail to deliver reliable results when faced with unlearned input data, leading to inaccurate thermal displacement estimates in machine tools, which in turn result in inappropriate correction amounts and reduced machining accuracy.
Innovation Solution
A fluctuation amount estimation device that employs a neural network with a parameter storage system for machine learning, where the estimation unit repeatedly estimates fluctuation amounts with omitted parameters, and a reliability evaluation unit assesses the reliability of these estimates based on variation, allowing for objective judgment and appropriate usage of estimated values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning is used to estimate thermal displacement amount, then estimation accuracy is improved under normal conditions, but reliability deteriorates when input data is unlearned or abnormal
Solution Approach 1:
The system performs repeated estimation multiple times and uses the variation in results as feedback to evaluate reliability. When the variation exceeds a threshold, the system identifies the input as unlearned or abnormal data, preventing inaccurate correction. This feedback mechanism allows the system to maintain high accuracy for normal data while reliably detecting when the model should not be trusted.
Solution Approach 2:
Instead of performing a single estimation, the system performs excessive action by repeating the estimation multiple times with the same input. This redundancy allows for reliability evaluation through variation analysis, enabling the system to distinguish between high-reliability and low-reliability estimates without requiring additional sensors or hardware.
2Reliability
If repeated estimation with omitted parameters is performed, then reliability evaluation capability is improved, but device complexity increases
Solution Approach 1:
The system uses its own internal variation as the evaluation criterion rather than requiring external validation mechanisms. By comparing the variation of repeated estimation results against a threshold, the system self-evaluates reliability without needing additional sensors, reference systems, or complex external verification infrastructure.
3Productivity
If correction is applied based on unreliable estimated values, then productivity is maintained through continuous operation, but manufacturing precision deteriorates due to inaccurate correction amounts
Solution Approach 1:
The reliability evaluation acts as a feedback gate that controls when correction should be applied. When reliability is high, correction is applied to maintain precision. When reliability is low (indicating unlearned or abnormal input), the system withholds correction, preventing degradation of machining accuracy while allowing continuous operation to proceed.
Data Source
AI summary
Provided are a fluctuation amount estimation device (9) capable of evaluating reliability of an estimated value and a correction amount calculation device (1) including the fluctuation amount estimation device (9). The correction amount calculation device (1) includes the fluctuation amount calculation device (9), a correction amount calculation unit (5), and a correction amount output unit (7). The fluctuation amount estimation device (9) includes a parameter storage (3) storing parameters as constituent elements of a neural network obtained by machine learning, an estimation unit (2) estimating a fluctuation amount relevant to a position of an element arranged in a machine tool (11) or a fluctuation amount of a distance between elements arranged in the machine tool (11) for each physical condition information of the machine tool (11) by means of the neural network with a parameter freely selected from the parameters being omitted, and a reliability evaluation unit (4) evaluating reliability of estimated multiple fluctuation amounts based on the estimated fluctuation amounts. The correction amount calculation unit (5) calculates a correction amount for the estimated fluctuation amounts based on the fluctuation amounts, and the correction amount output unit (7) outputs the calculated correction amount to outside.