Spindle Collision Detection via Torque-Acceleration Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting spindle collisions in machine tools, such as abnormal load detecting functions and acceleration sensors, face challenges in setting reference values, leading to either false alarms or missed detections, especially in distinguishing between heavy cutting and rapid traverses.
Innovation Solution
A control device combines abnormal load detection and acceleration data, utilizing a machine learning device to learn the relationship between estimated load torque values and spindle acceleration values during normal operation, allowing for accurate spindle collision detection without requiring trial-and-error setting of reference values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a reference value for abnormal load detection is set low, then spindle collision detection sensitivity is improved, but false alarms increase during normal heavy cutting
Solution Approach 1:
The patent combines abnormal load detection with acceleration sensor detection to form a composite detection system. The control device integrates both detection methods and performs comprehensive judgment, allowing the system to distinguish between heavy cutting (high load, low acceleration) and actual spindle collision (high load, high acceleration), thereby reducing false alarms while maintaining detection sensitivity.
Solution Approach 2:
The patent changes the detection parameters by introducing acceleration values as an additional dimension for judgment. Instead of relying solely on load current thresholds, the system now considers both load current and acceleration sensor outputs, enabling dynamic adjustment of detection criteria based on the combined state of both parameters.
2Reliability
If a reference value for abnormal load detection is set high, then false alarms are reduced, but spindle collision detection accuracy deteriorates
Solution Approach 1:
The control device merges abnormal load detection with acceleration sensor detection to create a more accurate detection system. By combining both detection methods, the system can set higher load thresholds without missing collisions, because acceleration data provides additional confirmation of actual collision events.
Solution Approach 2:
The acceleration sensor acts as an intermediary that mediates between the load detection system and the final collision determination. It provides additional information that helps resolve uncertainty when load values are borderline, enabling more confident judgment at higher thresholds.
3Speed
If an acceleration sensor is used for spindle collision detection, then detection speed is improved, but distinction between rapid traverse and actual collision becomes difficult
Solution Approach 1:
The patent merges acceleration sensor detection with abnormal load detection to resolve the ambiguity between rapid traverse and actual collision. The control device combines both acceleration values and load current values for comprehensive judgment, allowing it to distinguish rapid traverse (high acceleration, low load) from actual collision (high acceleration, high load).
4Measurement precision
If machine learning is introduced to learn the relationship between load torque and acceleration, then automatic detection accuracy is improved, but device complexity increases
Solution Approach 1:
The machine learning device enables the control system to automatically learn and adapt to the specific characteristics of each spindle through self-service. The system collects operational data during normal operation, learns the relationship between load torque and acceleration for that specific spindle, and uses this learned knowledge for automatic collision detection, eliminating the need for manual parameter setting and reducing operator burden.
Data Source
AI summary
A control device includes a machine learning device that learns a state of a spindle during normal machining without a collision of the spindle, and the machine learning device includes a state observation unit that observes spindle estimated load torque data indicating an estimated load torque value for the spindle and spindle acceleration data indicating an acceleration value of the spindle as state variables representing a current state of an environment and a learning unit that learns a correlation between the estimated load torque values for the spindle and the acceleration values of the spindle during the normal machining with use of the state variables.