Laser Process Signal Reconstruction for Unknown Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring laser machining processes, particularly those using deep neural networks, struggle with detecting anomalies and determining quality features with limited error examples, leading to unpredictable behavior under unknown error types and complex interpretations.
Innovation Solution
The method employs an autoencoder formed by a trained deep neural network that reconstructs process signal data sets to detect anomalies based on reconstruction errors, allowing for plausibility checks of quality features and reducing the need for extensive error examples during training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for anomaly detection in laser machining processes, then the ability to detect quality features is improved, but the system requires extensive error examples for training and exhibits unpredictable behavior under unknown error types
Solution Approach 1:
Instead of training the neural network to directly detect anomalies (which requires extensive error examples), the patent inverts the approach by training the network to reconstruct normal process signals. Anomalies are then detected by measuring the reconstruction error, allowing the system to reliably identify unknown error types without requiring them in the training data.
Solution Approach 2:
The patent introduces an autoencoder as an intermediary system between the raw process signals and the anomaly detection task. The autoencoder learns to compress and reconstruct normal process signals, serving as a mediator that transforms the detection problem into a reconstruction error measurement problem, thereby improving reliability for unknown error types.
2Device complexity
If traditional anomaly detection methods with reference curves and envelopes are used, then the system is simple to implement, but it cannot handle laser processing processes under changing conditions and requires significant equipment complexity for post-process monitoring
Solution Approach 1:
The patent replaces static reference curves and envelopes with a dynamic neural network model that adapts to changing process conditions. The autoencoder learns the underlying patterns of normal process signals and can dynamically adjust to variations in processing parameters, workpiece properties, and environmental conditions while maintaining simple in-process monitoring equipment.
Solution Approach 2:
The patent substitutes complex post-process mechanical measurement systems with an in-process optical monitoring system using photodiodes or cameras combined with neural network analysis. This replaces the need for separate measuring cells and physical contact measurements with a non-contact, real-time optical system that is both simpler and more adaptable.
3Measurement precision
If post-process monitoring systems are used for quality control, then quality assessment can be performed according to standards, but significant equipment complexity and separate measuring cells are required
Solution Approach 1:
The patent merges the quality assessment function into the laser processing system itself by using the same optical path and sensors for both process monitoring and quality evaluation. This eliminates the need for separate post-process measuring cells while maintaining standard-compliant quality assessment through neural network analysis of process signals.
Solution Approach 2:
The patent creates a virtual model of the process signals through neural network reconstruction, which serves as a digital copy that can be analyzed for quality assessment without requiring physical contact or separate measurement equipment. This virtual modeling approach maintains measurement precision while eliminating complex hardware requirements.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
Method and system for monitoring a laser processing process, the method comprising the steps of: inputting at least one process signal data set (31) of the laser processing process into an autoencoder (400) formed by a deep neural network; generating a reconstructed process signal data set (51) by the autoencoder; determining a reconstruction error based on the at least one process signal data set (31) and the at least one reconstructed process signal data set (51); and detecting an anomaly of the laser processing process based on the determined reconstruction error; as well as laser processing method comprising the method; and laser processing system comprising the system.